{"id":"e0fbc116-42bc-4707-8ecc-5e2f4773a92a","entityType":"agent","slug":"clawhub-franklinxkk-ai-delivery-spec","name":"AI Delivery Spec｜需求判断与交付","canonicalUrl":"https://www.xpersona.co/agent/clawhub-franklinxkk-ai-delivery-spec","canonicalPath":"/agent/clawhub-franklinxkk-ai-delivery-spec","generatedAt":"2026-10-10T02:10:06.419Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-09T10:56:40.890Z","emptyReason":null},"description":"Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变 Skill: AI Delivery Spec｜需求判断与交付 Owner: franklinxkk Summary: Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变 Tags: latest:5.5.2 Version history: v5.5.2 | 2026-09-29T23:","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 2.9K downloads reported by the source. Last updated 10/9/2026.","installCommand":"clawhub skill install s1788610zxyr699qednfs41d7n88xqj6:ai-delivery-spec","sourceUrl":"https://clawhub.ai/franklinxkk/ai-delivery-spec","homepage":"https://clawhub.ai/franklinxkk/skills/ai-delivery-spec","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/franklinxkk/ai-delivery-spec","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/franklinxkk/skills/ai-delivery-spec","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":69,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements,"},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-09T10:56:40.890Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T10:56:40.890Z","emptyReason":null},"stars":null,"forks":null,"downloads":2887,"packageName":null,"latestVersion":"5.5.2","tractionLabel":"2.9K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-09T10:56:40.866Z","emptyReason":null},"lastUpdatedAt":"2026-10-09T10:56:40.890Z","lastCrawledAt":"2026-10-09T10:56:40.866Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-10T10:56:40.866Z","lastVerifiedAt":null,"highlights":[{"version":"5.5.2","createdAt":"2026-09-29T23:57:25.062Z","changelog":"- Major cleanup: reduced file count significantly by removing legacy, maintainer, and workflow files. - Added runtime-manifest.json and build_review_workspace.py script to support review workspace operations. - Updated and streamlined example requirements documentation for medium-review-handoff. - Improved documentation focusing on clarity, minimal duplication, and directly actionable requirements for all roles. - Specification emphasizes direct business intent, reduces procedural overhead, and clarifies boundaries for requirement management.","fileCount":121,"zipByteSize":580647},{"version":"5.5.1","createdAt":"2026-09-22T14:09:25.622Z","changelog":"- Added maintainer validation and test files to improve version 5.5.1 evidence workflow. - Introduced .gitattributes and .gitignore for repository management. - Added an explicit LICENSE file. - Removed outdated skill-card.md. - No functional or rules changes to the main skill content.","fileCount":180,"zipByteSize":697324},{"version":"5.5.0","createdAt":"2026-09-08T13:28:29.726Z","changelog":"AI Delivery Spec 5.5.0 introduces a new governance and lifecycle structure for requirements management with major repository and documentation changes. - Adopted the Apache-2.0 license and updated SKILL.md for clearer boundaries, scope, and process guidance. - Added extensive contributor and maintainer guides, templates, and automation under the `.github/` and `maintainer/` directories. - Introduced implementation notes, evaluation schemas, industry evidence, example projects, and runtime/configuration files for governance, review, and assurance. - Removed outdated or duplicate files: LICENSE, runtime-manifest.json, and skill-card.md. - Streamlined documentation and reference structure to support requirements handling, change management, and product lifecycle in a concise and practical workflow.","fileCount":175,"zipByteSize":678977},{"version":"5.4.9","createdAt":"2026-09-02T11:43:02.011Z","changelog":"ai-delivery-spec 5.4.9 - Added references/domains/domain-media-knowledge.md as a new domain knowledge reference file. - Removed obsolete skill-card.md file. - Updated the main documentation with streamlined review/PRD/acceptance stage instructions and clarified product/technical tracing details. - Clarified that milestone gating must use explicit commands, with clear status states and validation requirements. - No changes to intent shortcuts, skill description, or entrance logic.","fileCount":118,"zipByteSize":602218},{"version":"5.4.8","createdAt":"2026-08-28T13:52:53.812Z","changelog":"ai-delivery-spec 5.4.8 — Adds lightweight intent shortcuts and refines core process - Introduced concise, multi-intent description with support for /ads, /dig, /prd, /proto shortcuts in SKILL.md. - Added documentation and logic for four shortcut entrypoints, clarifying their routing and default behaviors. - Updated SKILL.md for greater focus on usability, language locking, artifact criteria, and scenario coverage. - Removed outdated ./github/assets/lifecycle-bridge.svg and skill-card.md files. - Added new review-prototype HTML example for handoff scenarios.","fileCount":117,"zipByteSize":562990},{"version":"5.4.7","createdAt":"2026-08-26T00:14:54.026Z","changelog":"**Added support for structured review workspace documentation and schema.** - Introduced `references/review-workspace.md` and `schemas/review-workspace.schema.json` for organizing review workspace references and their schema validation. - Improved review-mode documentation and explicit review context handling, with new details for visual/product separation in prototype reviews. - Refined explanation of delivery depth (`direct`, `standard`, `governed`), risk facets, and evidence levels; deprecated old numeric artifact levels. - Clarified rules around stage entry/exit, artifact persistence, and cross-role handoff. - Removed obsolete `skill-card.md` file.","fileCount":117,"zipByteSize":544031},{"version":"5.4.6","createdAt":"2026-08-12T07:21:21.635Z","changelog":"ai-delivery-spec 5.4.6 - Removed the redundant sample file: `skill-card.md`. - Documentation clarifies: - Default language handling, YAML/JSON summaries, and machine value display updated for brevity. - Fine-tuned rules on context loading, pre-check (surface scan), and handling of edge/negative cases. - More precise instructions on file and stage handling; clarifies not to inject full templates/examples/maintainer content. - No changes to core logic, features, or lifecycle behavior—revision focuses on documentation tightening and maintenance cleanup.","fileCount":115,"zipByteSize":468104},{"version":"5.4.5","createdAt":"2026-08-10T23:39:04.932Z","changelog":"**Summary:** This release introduces a \"quick modify\" minimal path for small, low-risk requirement changes and adds supporting assets and documentation. - Added “快速小改 (quick modify)” path to process simple, local, and reversible requirement changes using minimal workflow. - Updated documentation to clarify default use of the lightest feasible delivery process for non-disruptive changes. - Added: README, CHANGELOG, runtime manifest, lifecycle SVG, and a new change contract script. - Removed: skill-card.md. - No impact on core workflow for standard or complex requirement deliveries.","fileCount":115,"zipByteSize":456925}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s1788610zxyr699qednfs41d7n88xqj6:ai-delivery-spec","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-10T02:10:06.416Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-franklinxkk-ai-delivery-spec/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-09T10:56:40.890Z","emptyReason":null},"readme":"Skill: AI Delivery Spec｜需求判断与交付\n\nOwner: franklinxkk\n\nSummary: Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变\n\nTags: latest:5.5.2\n\nVersion history:\n\nv5.5.2 | 2026-09-29T23:57:25.062Z | user\n\n- Major cleanup: reduced file count significantly by removing legacy, maintainer, and workflow files.\n- Added runtime-manifest.json and build_review_workspace.py script to support review workspace operations.\n- Updated and streamlined example requirements documentation for medium-review-handoff.\n- Improved documentation focusing on clarity, minimal duplication, and directly actionable requirements for all roles.\n- Specification emphasizes direct business intent, reduces procedural overhead, and clarifies boundaries for requirement management.\n\nv5.5.1 | 2026-09-22T14:09:25.622Z | user\n\n- Added maintainer validation and test files to improve version 5.5.1 evidence workflow.\n- Introduced .gitattributes and .gitignore for repository management.\n- Added an explicit LICENSE file.\n- Removed outdated skill-card.md.\n- No functional or rules changes to the main skill content.\n\nv5.5.0 | 2026-09-08T13:28:29.726Z | user\n\nAI Delivery Spec 5.5.0 introduces a new governance and lifecycle structure for requirements management with major repository and documentation changes.\n\n- Adopted the Apache-2.0 license and updated SKILL.md for clearer boundaries, scope, and process guidance.\n- Added extensive contributor and maintainer guides, templates, and automation under the `.github/` and `maintainer/` directories.\n- Introduced implementation notes, evaluation schemas, industry evidence, example projects, and runtime/configuration files for governance, review, and assurance.\n- Removed outdated or duplicate files: LICENSE, runtime-manifest.json, and skill-card.md.\n- Streamlined documentation and reference structure to support requirements handling, change management, and product lifecycle in a concise and practical workflow.\n\nv5.4.9 | 2026-09-02T11:43:02.011Z | user\n\nai-delivery-spec 5.4.9\n\n- Added references/domains/domain-media-knowledge.md as a new domain knowledge reference file.\n- Removed obsolete skill-card.md file.\n- Updated the main documentation with streamlined review/PRD/acceptance stage instructions and clarified product/technical tracing details.\n- Clarified that milestone gating must use explicit commands, with clear status states and validation requirements.\n- No changes to intent shortcuts, skill description, or entrance logic.\n\nv5.4.8 | 2026-08-28T13:52:53.812Z | user\n\nai-delivery-spec 5.4.8 — Adds lightweight intent shortcuts and refines core process\n\n- Introduced concise, multi-intent description with support for /ads, /dig, /prd, /proto shortcuts in SKILL.md.\n- Added documentation and logic for four shortcut entrypoints, clarifying their routing and default behaviors.\n- Updated SKILL.md for greater focus on usability, language locking, artifact criteria, and scenario coverage.\n- Removed outdated ./github/assets/lifecycle-bridge.svg and skill-card.md files.\n- Added new review-prototype HTML example for handoff scenarios.\n\nv5.4.7 | 2026-08-26T00:14:54.026Z | user\n\n**Added support for structured review workspace documentation and schema.**\n\n- Introduced `references/review-workspace.md` and `schemas/review-workspace.schema.json` for organizing review workspace references and their schema validation.\n- Improved review-mode documentation and explicit review context handling, with new details for visual/product separation in prototype reviews.\n- Refined explanation of delivery depth (`direct`, `standard`, `governed`), risk facets, and evidence levels; deprecated old numeric artifact levels.\n- Clarified rules around stage entry/exit, artifact persistence, and cross-role handoff.\n- Removed obsolete `skill-card.md` file.\n\nv5.4.6 | 2026-08-12T07:21:21.635Z | user\n\nai-delivery-spec 5.4.6\n\n- Removed the redundant sample file: `skill-card.md`.\n- Documentation clarifies: \n - Default language handling, YAML/JSON summaries, and machine value display updated for brevity.\n - Fine-tuned rules on context loading, pre-check (surface scan), and handling of edge/negative cases.\n - More precise instructions on file and stage handling; clarifies not to inject full templates/examples/maintainer content.\n- No changes to core logic, features, or lifecycle behavior—revision focuses on documentation tightening and maintenance cleanup.\n\nv5.4.5 | 2026-08-10T23:39:04.932Z | user\n\n**Summary:** \nThis release introduces a \"quick modify\" minimal path for small, low-risk requirement changes and adds supporting assets and documentation.\n\n- Added “快速小改 (quick modify)” path to process simple, local, and reversible requirement changes using minimal workflow.\n- Updated documentation to clarify default use of the lightest feasible delivery process for non-disruptive changes.\n- Added: README, CHANGELOG, runtime manifest, lifecycle SVG, and a new change contract script.\n- Removed: skill-card.md.\n- No impact on core workflow for standard or complex requirement deliveries.\n\nv5.4.4 | 2026-08-08T15:46:31.393Z | user\n\n- Major refactor: removed 96 files and added 4 core files.\n- Reduced codebase size by removing documentation, maintainer tools, examples, templates, workflows, and CI/config files.\n- Introduced new core scripts: three validator scripts and a scaffolding terms reference.\n- Updated and significantly condensed documentation to a bilingual, concise form.\n- Focused project scope on essential requirement management kernel functionality, removing ancillary and maintenance materials.\n\nv0.1.26 | 2026-07-30T14:52:58.299Z | auto\n\nai-delivery-spec 0.1.26\n\n- 升级主版本为 AI Delivery Spec 5.4.3，文档对应更新\n- schemas/spec-config.schema.json 增补/微调配置项，增强配置校验\n- CLI/scripts 增强对路径漂移和 resume_context 校验\n- 修复和完善 test_product_experience.py、test_v540_readme_commands.py 测试场景\n- 评审与门禁 validator 策略细化，提高静态门禁准确性\n- 文档/示例同步完善（README.md、examples/spec.config.example.yaml）\n\nv0.1.25 | 2026-07-29T23:00:15.589Z | auto\n\n- Major update: Improved focus on fast convergence, practical usage for all requirement changes, and minimal files.\n- Enhanced description and usage scope, requiring all requirement/PRD/prototype work to invoke this skill regardless of clarity/scale.\n- Clarified: Stages are routing maps, not execution checklists. Now always enter the user's goal stage directly if clear, without redundant prior files.\n- In chat, outputs are concise and practical—internal YAML and IDs shown only for handover, audit, or validation.\n- Added section: Small iterations should minimally modify first, with real diffs and contracts only as necessary.\n- Static gating (quality checks) now runs only at explicit milestones, not every stage, and groups diagnostics by root cause.\n\nv0.1.24 | 2026-07-28T00:00:15.862Z | auto\n\nai-delivery-spec 0.1.24\n\n- Introduced new handoff/static gate validators for PRD, prototype, and handoff phases.\n- Added support for Stage 0 inventory with schema and template examples.\n- Updated main documentation to clarify end-to-end deliverables, strengthen gate requirements, and expand baseline/handoff usage guidance.\n- Improved test coverage for gate and human-first handoff workflows.\n- Enhanced templates and schemas for better requirement traceability and structural checks.\n\nv0.1.23 | 2026-07-26T07:36:57.997Z | auto\n\nAI Delivery Spec 5.4.0 is a major release with requirement lifecycle \"workstations\" and stage-based outputs.\n\n- Added stage-based lifecycle: users can now specify entry/target stages (frame, explore, intake, clarify, specify, review, baseline, change, acceptance) for all requirement work.\n- Introduced new minimal, stage-specific artifacts (e.g., problem-brief, solution-sketch, requirement-brief) and templates with language-agnostic anchors, supporting non-linear and partial workflows.\n- Lifts strict ordering: users can enter or stop at any stage with valid evidence; only one minimal output per stage is generated by default.\n- Comprehensive overhaul of SKILL.md, with new guidance for artifact scope, traceability, change management, review, and gatekeeping.\n- Expanded support for assumption registers, decision records, and check/acceptance states; updated schemas, templates, and validation tools.\n- Major internal file reorganization and test coverage for new 5.4 workflow and gates.\n\nv0.1.22 | 2026-07-22T15:52:48.120Z | auto\n\nAI Delivery Spec 迎来大幅升级，正式支持中英双语规范输出与领域最佳实践，契合 ToC/ToB/ToG 等场景。\n\n- SKILL.md 全面改写为中英文对照、支持领域和角色习惯，内容结构更清晰。\n- 明确规范卡、统一 PRD、治理真相等交付形态及适用场景，具体业务分级与升级规则更细致。\n- 双语输出改为按用户请求，所有表述、本体、表格与推理均自动适应用户语言。\n- 丰富流程解释与例子，全新覆盖指标、线索、状态、投影、门禁、知识回流等模块。\n- 增加大规模项目输入分轮、端到端纵切检查点、工程/AI索引等最佳实践说明。\n- 新增大量维护、结构、模板和 issue/workflow 支持文件。\n\nv0.1.21 | 2026-07-19T09:43:09.227Z | auto\n\n- Updated SKILL.md for clarity, more detailed requirement handling, and streamlined instructions.\n- Improved intake and clarification process; now emphasizes confirming P0/P1 facts and recording assumptions with reversal paths.\n- Enhanced contract for traceability, state handling, metrics, acceptance, and review steps.\n- Gate instructions refined: auto-level PRD/prototype/hand-off reading and explicit browser/run evidence for higher levels.\n- Removed skill-card.md; documentation is now consolidated in SKILL.md.\n\nv0.1.20 | 2026-07-18T16:43:51.785Z | auto\n\n## ai-delivery-spec 0.1.20\n\n- No code or documentation changes detected in this release.\n- Version bump only; functionality and behavior remain unchanged.\n\nv0.1.19 | 2026-07-18T16:28:56.707Z | auto\n\nVersion 0.1.19 of ai-delivery-spec\n\n- No file changes detected in this release.\n- All requirements management, processes, and documentation remain unchanged.\n- No updates to user-facing features, references, or tool behavior.\n\nv0.1.18 | 2026-07-18T16:17:37.645Z | auto\n\n- Major update: streamlined requirement management language, clarified workflow, and improved gate and assurance model.\n- Overhauled SKILL.md for clarity: easier intake, explicit roles/authority, and better traceability from source to acceptance.\n- Simplified stages and file loading; load one stage/domain at a time, and clarified what should not be loaded.\n- Enhanced guidance for large/complex work, learning/candidate contributions, and handoff protocols.\n- Gate now returns clearly defined states; PASS, REVIEW_COMPLETE_WITH_GAPS, BLOCKED_BY_P0_UNKNOWN, or BLOCKED.\n- Removed obsolete \"skill-card.md\".\n\nv0.1.17 | 2026-07-16T15:14:31.390Z | auto\n\nai-delivery-spec 0.1.17\n\n- Updated requirement management kernel to version 5.2.0 with improved instructions and references.\n- Expanded guidance on loading only the active slice, now including context planning and troubleshooting sections.\n- Enhanced domain loading and evidence rules; stricter section-based loading for domain packs.\n- Added new CLI usage examples for findings explanation and resume commands.\n- Removed legacy skill-card.md file.\n\nv0.1.16 | 2026-07-15T16:00:10.920Z | auto\n\nai-delivery-spec 0.1.16 Changelog\n\n- Updated SKILL.md to version 5.1.7 with improved stage names and clearer lifecycle steps.\n- Clarified scope: skill now manages requirements only; planning, code, CI/CD, deployment, and operations are out-of-scope.\n- Revised loading guidance: lists explicit references for each need, adds Coding Agent/tool and multiple domain support, removes now-unnecessary references.\n- Expanded description and coverage of implementation contract IDs and clarified requirements for prototypes and metrics.\n- Refined outcome states returned by the light gate and updated sample usage for various profiles and levels.\n- Removed skill-card.md to reduce duplication and streamline documentation.\n\nv0.1.15 | 2026-07-15T03:42:46.273Z | auto\n\n- Refined scope and description to clarify focus on core requirement management only; removed references to unrelated tasks.\n- Added new coverage for page delivery contract and four-lens sign-off in stage references.\n- Tightened intake and clarification criteria—Ultra-Light and L2+ distinctions more explicit, simplified original instructions.\n- Expanded implementation contract section: added requirements for metrics definition, list/form controls, surface declarations, explicit API/data-flow mapping, and four-lens sign-off for L3/L4 views.\n- Removed `skill-card.md` file.\n\nv0.1.14 | 2026-07-14T02:50:52.756Z | auto\n\nai-delivery-spec 0.1.14\n\n- Streamlined SKILL.md: clearer requirement management focus, less product delivery kernel/process detail.\n- Updated description—focuses on requirement intake, clarification, specification, traceability, acceptance; omits unrelated ops, task, or code debugging.\n- New process sequence: intake → clarify → specify → review → baseline → change → acceptance → closed.\n- Emphasized minimal loading: only active references/slices, not whole repo or all examples.\n- Removed skill-card.md.\n\nv0.1.13 | 2026-07-12T15:20:42.397Z | auto\n\n**ai-delivery-spec 0.1.13** \nMajor update with streamlined delivery process and documentation.\n\n- SKILL.md significantly revised: improved guidance for product triage, artifact selection, and requirements gathering.\n- Updated terminology and rigor: introduces \"Ultra-Light\" mode, new tier definitions, and detailed consumer types.\n- Reference and runtime architecture overhauled for focused loading, domain selection, and more precise contract slicing.\n- Enhanced support for AI coding PRDs, progressive product truth management, and large-project workflows.\n- Removed outdated skill-card.md file.\n\nv0.1.12 | 2026-07-04T14:44:59.551Z | auto\n\nai-delivery-spec v0.1.12\n\n- Updated to Production Elastic Delivery Standard v4.9.14.\n- Clarified domain module loading: load only matched domains and use Domain Composition Map for multiple domains, avoiding unnecessary \"safety\" loads.\n- Improved runtime file architecture section for greater specificity.\n- No changes to core functionality; all changes affect documentation and operational clarity only.\n\nv0.1.11 | 2026-07-02T13:05:36.063Z | auto\n\nai-delivery-spec v0.1.11\n\n- Updated the production elastic delivery standard from v4.9.10 to v4.9.11 in SKILL.md.\n- No changes to logic, rules, workflow, or feature behavior; documentation update only.\n\nv0.1.10 | 2026-06-30T16:06:33.287Z | auto\n\nv0.1.10\n\n- Updated the production standard version reference from v4.9.9 to v4.9.10 in SKILL.md.\n- No other substantive changes were made; documentation content remains the same.\n\nv0.1.9 | 2026-06-29T23:34:52.992Z | auto\n\nai-delivery-spec 0.1.9\n\n- Documentation updated: SKILL.md revised to v4.9.9, reflecting minor content and version updates.\n- Outdated skill-card.md file removed for a cleaner project structure.\n\nv0.1.8 | 2026-06-28T13:54:36.377Z | auto\n\n- Added a new INFO field to triage classification for completeness tracking.\n- Clarification protocol introduced for cases where INFO is marked missing.\n- Guidance added: if key info is missing, clarify before generating; proceed with assumptions if directed, marking as REVIEW_COMPLETE_WITH_GAPS.\n- Updated version reference to v4.9.8.\n- Removed file: skill-card.md.\n\nv0.1.7 | 2026-06-27T14:06:15.714Z | auto\n\nai-delivery-spec 0.1.7\n\n- Updated SKILL.md description for clarity and conciseness.\n- Changed version reference from v4.9.4 to v4.9.7 in the delivery standard.\n- Added a clearer, more concise skill summary at the top.\n- Minor language and format edits to remove redundancy and improve readability.\n- No changes to file structure or primary workflow logic.\n\nv0.1.6 | 2026-06-27T11:57:27.063Z | auto\n\nNo file changes detected in this version.\n\n- No updates or modifications were made to the codebase.\n- The skill version remains identical to the previous release.\n\nv0.1.5 | 2026-06-27T11:56:47.894Z | auto\n\n# ai-delivery-spec v0.1.5 Changelog\n\n- Updated to Production Elastic Delivery Standard v4.9.4 (from v4.9.2)\n- Minor text, version, and heading adjustments in SKILL.md to align with new standard version\n- No changes to skill logic, runtime rules, or usage guidance\n\nv0.1.4 | 2026-06-27T07:48:50.301Z | auto\n\n- Updated standard version from v4.7.1 to v4.9.2 with expanded rules and process detail.\n- Added \"PRD Profile Selector\" and \"Product Work Path Selector\" sections for clearer output type and process path selection.\n- Adjusted runtime file architecture language: templates and domain modules are now explicitly load-on-demand, not loaded by default.\n- skill-card.md file removed.\n- Expanded and clarified coverage for different PRD profiles (Contract Summary, Human-First Full PRD, AI-Coding Full PRD) and work paths.\n\nv0.1.3 | 2026-06-25T16:05:45.568Z | auto\n\n- Version bump to v4.7.1 with expanded delivery standard.\n- Added `references/realtime-contract.md` entrypoint for products including real-time features (SSE/WebSocket, timers, alerts, polling).\n- Updated runtime architecture table for new real-time contract add-on.\n- Removed legacy file `skill-card.md`.\n- No changes made to core triage, scope, mode, or gates except for real-time artifact extension.\n\nv0.1.2 | 2026-06-22T07:46:26.335Z | auto\n\nai-delivery-spec v0.1.2\n\n- Updated SKILL.md to v4.6.3, expanding runtime file architecture and adding details on triggered add-ons (e.g., coding agent compatibility).\n- Introduced a dedicated section and routing logic for coding-agent-specific contracts and machine-readable handoff (AC-YAML, AGENTS.md).\n- Clarified the handling of add-on references and stricter loading rules for default and triggered entrypoints.\n- Removed obsolete skill-card.md file.\n\nv0.1.1 | 2026-06-19T02:13:57.657Z | auto\n\n- Removed legacy skill-card.md file; all documentation now consolidated in SKILL.md.\n- Expanded \"Learn/Retire\" coverage: now explicitly clarifies minimalism—capture post-launch metric review and sunset evidence without requiring full experimental frameworks.\n- Updated Routing, Tiering, and Lifecycle guidance for greater clarity and directness.\n- No interface or runtime changes; file loading and core delivery gates remain unchanged.\n\nv0.1.0 | 2026-06-18T15:50:20.442Z | auto\n\nInitial release of ai-delivery-spec skill.\n\n- Introduces triage process classifying requests by tier, AI, and workflow scope.\n- Defines strict entrypoints for modular loading: core delivery, prototype/testability, and advanced extensions.\n- Details gating criteria for user story mapping, demo prototype validation, product specification, and acceptance package assembly.\n- Provides specific routing rules for different artifact and module types.\n- Emphasizes testability, lifecycle tracking, and artifact handoff accountability throughout the delivery process.\n\nArchive index:\n\nArchive v5.5.2: 121 files, 580647 bytes\n\nFiles: agents/openai.yaml (394b), CHANGELOG.md (82685b), examples/medium-review-handoff/requirement.md (8079b), examples/medium-review-handoff/review-prototype.html (45052b), examples/minimal-v5/intake.yaml (1046b), examples/minimal-v5/README.md (835b), examples/minimal-v5/requirement-card.md (3974b), examples/spec.config.example.yaml (1318b), LICENSE (10701b), README.md (33348b), references/change-acceptance.md (3404b), references/context.md (2524b), references/discover.md (3187b), references/domain-coverage.yaml (18520b), references/domains/domain-ai-native.md (20725b), references/domains/domain-crm.md (18143b), references/domains/domain-data-mart.md (46038b), references/domains/domain-education-it.md (42073b), references/domains/domain-media-knowledge.md (30097b), references/domains/domain-medical-hospital-it.md (27321b), references/domains/domain-oa.md (29678b), references/domains/domain-sources.yaml (71937b), references/domains/domain-traffic.md (50394b), references/lifecycle.md (2500b), references/patterns/common-requirement-patterns.yaml (15281b), references/patterns/realtime-contract.md (13616b), references/prototype.md (4772b), references/review-workspace.md (33448b), references/scaffolding-terms.yaml (740b), references/specify.md (7434b), references/stages.md (4791b), references/templates/acceptance-run-template.yaml (1479b), references/templates/agent-handoff-manifest-template.yaml (600b), references/templates/assumption-register-template.yaml (815b), references/templates/change-request-template.yaml (1414b), references/templates/decision-record-template.md (1194b), references/templates/discovery-contract-template.yaml (1166b), references/templates/prd-light-template.md (1242b), references/templates/problem-brief-template.md (1638b), references/templates/product-truth-core-fragment-template.yaml (454b), references/templates/product-truth-index-template.yaml (469b), references/templates/product-truth-module-fragment-template.yaml (246b), references/templates/product-truth-template.yaml (5805b), references/templates/requirement-brief-template.md (2877b), references/templates/requirement-intake-template.yaml (530b), references/templates/requirement-register-template.yaml (1256b), references/templates/review-record-template.yaml (1411b), references/templates/solution-sketch-template.md (1730b), references/templates/stage0-inventory-template.yaml (6653b), references/templates/unified-requirement-prd-template.md (6658b), references/tool-adapters.md (2443b), references/troubleshooting.md (4119b), runtime-manifest.json (21087b), schemas/acceptance-run.schema.json (3831b), schemas/agent-handoff.schema.json (5025b), schemas/assumption-register.schema.json (3240b), schemas/change-package.schema.json (6896b), schemas/clarification-transcript.schema.json (2130b), schemas/context-plan.schema.json (3712b), schemas/discovery-contract.schema.json (4613b), schemas/domain-candidate.schema.json (1806b), schemas/domain-usage-log.schema.json (1029b), schemas/execution-state.schema.json (5649b), schemas/gate-result.schema.json (4062b), schemas/product-truth-fragment.schema.json (2194b), schemas/product-truth-index.schema.json (1349b), schemas/product-truth.schema.json (21351b), schemas/project-domain-capsule.schema.json (6465b), schemas/requirement-intake.schema.json (5612b), schemas/requirement-pattern-library.schema.json (1516b), schemas/requirement-register.schema.json (5753b), schemas/review-record.schema.json (4331b), schemas/review-workspace.schema.json (37457b), schemas/spec-config.schema.json (5190b), schemas/stage0-inventory.schema.json (12331b), schemas/traceability-ledger.schema.json (2073b), scripts/ai_delivery_spec_cli.py (66842b), scripts/analyze_change_impact.py (8006b), scripts/build_review_workspace.py (7787b), scripts/build_traceability_ledger.py (4304b)\n\nFile v5.5.2:SKILL.md\n\n---\nname: ai-delivery-spec\nlicense: Apache-2.0\ndescription: Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变更与验收；一句话想法、表单/表格规则和微小改动也适用。明确的纯翻译、排版、抄录或既定步骤执行由对应工具直接处理。\n---\n\n# AI Delivery Spec 5.5.2 — 需求判断与交付\n\n帮助用户管清需求，让产研和 AI 少猜、少漏、少返工。跟随用户语言；代码、字段和已有稳定 ID 保留原名。\n\n面向业务/产品、设计、前后端、QA 与 Coding Agent；从想法、存量材料或变更进入，各角色共用业务约定，不重复建文档。\n\n## 从当前目标进入\n\n识别用户要解决问题、比较方案、明确规则、获得原型，还是审查已有改变。已有决定直接继承；不因模板要求重复批准。`/ads`、`/dig`、`/prd`、`/proto` 只表达意图，宿主是否支持裸命令由实际能力决定。\n\n明确局部小改：读取相关基线，完成差异、继承边界和正反验收；没有关键未知就不提问、不建生命周期文件。不默认走全流程。\n\n按依赖推进：澄清当前决定 → 写清业务切片或走通原型 → 按需补追溯 → 验证。已有答案直接进入所需阶段。每阶段先解决业务分歧；编号、模板和门禁不能挤占业务内容。标题用业务名称，ID 留在引用位置。\n\n办公任务的目标取舍、口径、权限或流程改变同样适用；澄清后交对应工具执行。纯格式整理或既定步骤执行无需制造需求问题。\n\n| 当前需要 | 按需读取 |\n|---|---|\n| 判断问题、比较方案、澄清 | [discover.md](references/discover.md) |\n| 轻量规格、正式 PRD 与模块交接 | [specify.md](references/specify.md) |\n| 存量盘点、生成或修改可操作原型 | [prototype.md](references/prototype.md) |\n| 用户需要双态评审 | [review-workspace.md](references/review-workspace.md) |\n| 准入、处置、责任与基线 | [lifecycle.md](references/lifecycle.md) |\n| 变更、交接反馈或验收 | [change-acceptance.md](references/change-acceptance.md) |\n| 多文件、大上下文或跨会话 | [context.md](references/context.md) |\n| 机器路由、模板或检查命令 | [stages.md](references/stages.md) |\n\n不预加载全部模板或领域包。涉及行业规则，先按 [领域检索指引](references/stages.md#领域检索) 取当前问题切片和来源基线，再核实有权原文及版本；中英文同样执行。语言不决定法域，来源须匹配辖区与适用对象，不能照抄成项目真相。\n\n## 保持业务含义与决定权\n\n- 分清已核实事实、授权决定、观察、建议和未知。来源按主题、版本和授权范围判断；原型行为不自动成为产品规则。\n- 最新有效决定覆盖同主题旧内容。文档待同步、旧评审待复验不使该决定重新变成待批准。真正超出授权或存在冲突时只处理相应范围。\n- 建议暂缓、不做、缩范围写回现有产物；未获处置权不得改变需求状态。记录理由、依据和复议条件；已获授权不重复询问。\n- 未定规则保持未定。退路可以限制执行，不能借“保守默认”选定补考、口径、权限或晚到数据政策。未知只阻断依赖它的交付。\n\n## 最小充分规格\n\n实施者仍可能作出互不兼容的关键业务选择时，补足该处语义；技术实现保留合理空间。说明行为前提、允许者、业务结果、失败恢复及可判验收。按实际风险补状态、权限、指标、外部数据或历史对象约定，不按角色数或旧等级加长文档。\n\n事实只在一处人工定义，模块内就近引用或展开，使接收者连续读懂任务。业务审批和需求评审是不同对象；审核通过仅为发布前提时，不擅自合并成自动发布。\n\n关键规则用能区分错误实现的反例验证；标签缺失不能掩盖风险，模型一致不能代替来源。小改不附加全角色报告。\n\n正文 GAP 是待核实分歧，未命中也不证明通过。回读规则与来源，检查条件、反向路径和下游；不能为消除 GAP 编造政策、来源或 pass。\n\n## 变更与完成\n\n存量原型先盘点受影响页面、角色、入口、动作/处理器、状态、实体、数据源和 Mock 边界；保护未经取消的基线功能与视觉约定。产品态默认可操作；双态评审按用户需要启用。\n\n关键变更找到写入者、读取者、指标、入口、旧对象和受影响证据。候选依赖与核实结果分开，核实依赖不等于批准修改。多方修改前核对当前基线，不能静默覆盖漂移。\n\n达到用户目标就停止。检查只在需要的里程碑执行，不默认 full/handoff。静态、语义评阅、浏览器、真实系统与业务签署分别说明范围、版本及结果；没运行写未运行。小范围通过不代表全项目完成，建议被采纳不代表实现已验收。\n\n本 Skill 管需求及其产物；工程方案、排期、编码、部署和运营由相应工作流负责，只接收必要反馈与证据。私人材料与凭据不进入公共示例；外部写入遵守用户实际授权。\n\nFile v5.5.2:examples/minimal-v5/README.md\n\n# 最小需求示例\n\n这个虚构示例说明如何为制度列表增加“仅看当前有效”筛选。现有权限、时间边界、正常/空/失败结果与验收仍需说清，但不需要完整生命周期。\n\n```powershell\npython scripts/ai_delivery_spec_cli.py triage --input examples/minimal-v5/intake.yaml --format json\npython scripts/ai_delivery_spec_cli.py gate --profile prd --prd examples/minimal-v5/requirement-card.md --stage specify\n```\n\n预期为 card 路由建议及静态 PASS。建议不改变需求状态；PASS 不能证明自然语言风险发现完整、浏览器交互、真实系统或客户验收。\n\n示例保留一份可读取的旧式卡片，演示 5.4.x 内容兼容。新任务可用更短的模板或直接答复；状态、集成、指标等复杂点只补相关语义，不自动升级长 PRD。\n\nFile v5.5.2:README.md\n\n# AI Delivery Spec 5.5.2\n\n**帮你管清需求，让产研和 AI 少猜、少漏、少返工。**
\n**Manage requirements with less guesswork, fewer omissions and less rework—for your team and AI.**\n\n面向**产研团队与 AI Agent**的需求管理内核，以 Skill 形式使用。从一句话、现有材料或变更进入，帮你定清该做什么、交代清楚业务规则、找出改动影响。按当前任务生成或维护需求卡、PRD、可操作原型与验收条件，让接手者知道依据什么做、哪些还没定。\n\nA requirements management core for **product teams and AI agents**, delivered as a skill. Start with an idea, existing material or a change. Decide what needs doing, make business rules clear and identify change impacts. Create or update only the requirement cards, PRDs, interactive prototypes and acceptance criteria the task needs, so whoever takes over knows what to work from and what remains undecided.\n\n[![ClawHub downloads: 2.6k](https://img.shields.io/badge/ClawHub-2.6k_downloads-2563eb)](https://clawhub.ai/franklinxkk/skills/ai-delivery-spec)\n[![SkillHub AI score: 4.7/5](https://img.shields.io/badge/SkillHub_AI-4.7%2F5-f59e0b)](https://skillhub.cn/skills/user_12c92261/ai-delivery-spec)\n[![License: Apache 2.0](https://img.shields.io/badge/License-Apache_2.0-64748b)](LICENSE)\n\n2026-09-13 社区快照：ClawHub 约 2.6k 次下载；SkillHub 4.7/5 为 v5.4.8 历史 AI 评分，非当前版本新评测。 / Community snapshot: approximately 2.6k ClawHub downloads; SkillHub's 4.7/5 is a historical AI rating of v5.4.8, not a new evaluation of this release.\n\n**[角色价值 / Role value](#roles) · [中文上手](#zh) · [English guide](#en) · [四个快捷入口 / Shortcuts](#shortcuts) · [安装 / Install](#install) · [示例 / Examples](#examples) · [中英社区 / Community](#community)**\n\n\n\n## 各产研角色能得到什么\n\n| 核心用户 | 经常遇到的问题 | 这次能拿走什么 |\n|---|---|---|\n| **初级产品经理** | 收到一句需求，不知道该问什么、写到多细 | 关键问题、范围与边界、能开始评审的需求卡或 PRD |\n| **中高级产品 / 产品负责人** | 需求都合理，但优先做什么、跨模块如何一致还没定 | 问题证据、方案取舍、最小验证、当前决定与变更影响 |\n| **业务 / 售前 / 实施 / 设计** | 客户说法、业务规则和页面体验之间有断层 | 可确认的业务行为、可操作的产品原型、待决定事项 |\n| **前端研发** | 页面有了，入口、状态、权限和失败反馈仍不明确 | 与规格一致的交互路径、状态结果与验收条件 |\n| **后端 / 架构** | 同一句话会推导出不同口径、状态或写入方式 | 数据权威、允许的状态变化、副作用、恢复与集成边界 |\n| **QA / 验收方** | “显示正确”无法变成可重复的验收 | 正反例、权限与边界场景、变更回归范围和证据缺口 |\n| **Coding Agent** | 换个会话就丢背景，或自行补出业务政策 | 当前有效规则、来源与稳定引用、未知及可接续的任务范围 |\n\n适用于 ToC 产品、ToB/ToG 业务系统及 AI Native 场景。这些角色共用同一份业务约定，各自按需要读取；已有 PRD、需求系统和批准基线可以继续作为权威位置。\n\n## 从决定到交付，重点做好三件事\n\n**1. 更快找到当前要决定什么。** 从目标、受影响的人和事实出发，比较方案与最小验证。有依据的“先验证、暂缓、缩范围或不做”也可以完成分析；改变需求状态仍取决于实际授权。\n\n**2. 让规格可以体验，让评审有具体落点。** PRD 说明业务规则，原型呈现操作与结果，验收条件判断是否符合约定。“审批通过后可发布”要在规则、按钮行为与验收中区分发布资格和发布动作。需要双态评审时，可以边操作产品，边在当前页面旁查看规则、边界与验收依据。只补影响关键业务选择的内容，保留合理的工程实现空间。\n\n**3. 变更之后，相关产物仍然说同一件事。** 沿写入者、读取者、入口、指标和旧对象找具体依赖；区分候选影响与已核实影响，让 PRD、原型和交接引用同一规则。\n\n清晰小改直接完成；复杂需求按问题深入。工作量跟随当前目标，不要求先选 L0–L4、跑完整生命周期或填完全部模板。产品态原型默认可操作；需要面向产研的双态评审时，再开启评审标记与工作区。\n\n**先把业务讲清、操作走通，再补必要的追溯。** PRD 标题用业务名称，编号放在引用位置；快速原型不先搭全套评审设施。已定规则、待决影响和研发可自行选择的实现分别说清，避免把“有待决记录”误当成“已经可以开发”。
\n**Explain the business and make the task work before adding traceability.** Use business titles, keep IDs in references, and distinguish settled behavior, unresolved dependencies and engineering choices. A recorded question is not implementation readiness.\n\n\n\n## 安装到你的 Agent｜Install in your agent\n\n适用于能够加载 Agent Skills / `SKILL.md` 的**开发与办公 Agent**。安装、隐式调用、文件访问和浏览器能力由宿主提供，具体支持范围见[宿主适配](references/tool-adapters.md)。
\nUse it with **development and office agents** that load Agent Skills / `SKILL.md`. Installation, implicit invocation, file access and browser capabilities depend on the host; see [host adapters](references/tool-adapters.md).\n\n[Skills CLI](https://github.com/vercel-labs/skills) 支持的宿主可使用以下命令，按提示选择你的 Agent：
\nFor hosts supported by the Skills CLI, run this command and select your agent:\n\n```bash\nnpx skills add franklinxkk/ai-delivery-spec\n```\n\n安装后直接到 [中文上手](#zh) 或 [English guide](#en) 复制你的第一条任务。**日常使用不需要 Python。**
\nThen copy your first task from the Chinese or English guide. **Python is optional.**\n\n公开发布包与校验值见 [Releases](https://github.com/franklinxkk/ai-delivery-spec/releases)。仓库候选与社区渠道可能处于不同版本，请核对实际安装版本。 / Find published packages and checksums in [Releases](https://github.com/franklinxkk/ai-delivery-spec/releases). Repository candidates and community channels may differ; check the installed version.\n\n
\nZIP 与宿主导入｜ZIP and host-native import\n\n已有 ZIP 安装包？解压到宿主识别的 `ai-delivery-spec` 技能目录，让 `SKILL.md` 位于目录根部，再按宿主要求重新加载技能。
\nHave a ZIP package? Extract it into your host's `ai-delivery-spec` skill directory, with `SKILL.md` at its root, then reload skills as required by the host.\n\n宿主提供技能导入界面时，也可按其说明导入目录或 ZIP。仓库与社区各自更新版本，安装后请核对包内版本。
\nIf your host provides a skill-import interface, follow its instructions to import the directory or ZIP. Repository and community channels update separately; check the installed package's version.\n\n
\n\n\n\n## 第一次用？复制一句话就能开始\n\n安装后，在 Agent 对话中输入这句话；也可以直接换成你的真实需求。\n\n```text\n使用 ai-delivery-spec：给现有列表增加“仅看已启用”筛选，\n沿用系统已有的启用状态，默认显示全部，保留现有权限。\n把这次改动的规则和验收说明白。\n```\n\n你会得到这次修改的范围、筛选含义、正常与异常结果，以及可判断对错的验收条件。有已确认资料时直接沿用；存在关键未知时先指出需要谁决定。小改可以用一张需求卡或简短差异说明完成。\n\n**已有材料就一起给它。** PRD、截图、HTML、客户反馈或变更说明都可以作为起点；不同材料的事实与权威需要核实。资料读取、原型生成和验证能力取决于宿主实际提供的工具。\n\n### 选一句最像你现在的任务\n\n| 现在要做什么 | 可以直接这样说 |\n|---|---|\n| **想清楚值不值得做** | “使用 ai-delivery-spec：用户说流程太慢。先判断可能卡在哪里，给我能改变选择的最小验证。” |\n| **写清楚需求 / PRD** | “使用 ai-delivery-spec：基于这些已确认材料写 PRD，把角色、规则、异常和验收说明白。” |\n| **做可操作原型** | “使用 ai-delivery-spec：基于这份 PRD 做可操作的产品原型，覆盖主路径和关键失败结果。” |\n| **改现有需求或系统** | “使用 ai-delivery-spec：审核和发布要拆开，梳理旧对象、相关页面、权限和验收受到的影响。” |\n| **评审与交接** | “使用 ai-delivery-spec：站在研发和测试接收者角度审查这份规格，找出仍要靠猜的关键业务选择。” |\n| **办公流程与表格规则** | “使用 ai-delivery-spec：报销登记表要自动标出超期项，先帮我明确起算日、例外和责任人，再交给表格工具实现。” |\n\n\n\n### 四个快捷入口怎么用｜Four intent shortcuts\n\n**它们是对话中的意图简写。** 将下面任一句发给已加载技能的 Agent，附上相关材料或路径。安装技能是否同时注册裸斜杠命令取决于宿主；如果 `/dig` 等被宿主拦截，直接使用表中的带技能名写法。它们不是终端命令，也不对应四套独立流程。
\n**These are conversation shortcuts.** Send a prompt below to an agent with the skill loaded, together with relevant materials or paths. Native slash registration depends on the host; if a bare shortcut is intercepted, use the explicit skill-name prompt below. These are not shell commands or four separate workflows.\n\n| 入口 / Intent | 复制使用 / Copy and adapt | 当前结果 / Expected result |\n|---|---|---|\n| **`/ads` · 通用 / General** | `使用 ai-delivery-spec，以 /ads 处理：给现有列表增加按创建日期排序，保留权限。`
`Use ai-delivery-spec with /ads: add sorting by creation date to the existing list, preserving permissions.` | 从当前任务进入，完成必要差异与验收。 / Start at the relevant stage; deliver the change and acceptance needed. |\n| **`/dig` · 深挖 / Discover** | `使用 ai-delivery-spec，以 /dig 处理：客户说合同审批太慢，帮我找到真正的问题和最小验证。`
`Use ai-delivery-spec with /dig: customers say contract approval is slow. Help identify the problem and the smallest useful validation.` | 找关键决定，澄清会改变选择的问题；足够后回到用户目标。 / Clarify decisions that change the choice, then continue toward the requested outcome. |\n| **`/prd` · 规格 / Specify** | `使用 ai-delivery-spec，以 /prd 处理：基于附件中的已确认规则编写 PRD，按业务模块写清正常、拒绝与恢复。`
`Use ai-delivery-spec with /prd: write a PRD from the attached confirmed rules, organizing success, rejection and recovery by business module.` | 可读业务约定、待决影响及验收；小改可用需求卡。 / Readable business rules, unresolved impacts and acceptance; small changes can use a card. |\n| **`/proto` · 原型 / Prototype** | `使用 ai-delivery-spec，以 /proto 处理：基于这份 PRD 做可操作原型，并为研发测试提供当前页面的评审说明。`
`Use ai-delivery-spec with /proto: build an interactive prototype from this PRD, with contextual review notes for engineering and QA.` | 可操作产品态；明确需要时提供就近评审说明。 / A working product view, with contextual review notes when requested. |\n\n不知道选哪个就用 `/ads` 或直接说任务。单独写 `/prd`、`/proto` 不会替你确定业务规则；已有决定直接继承，只澄清真正影响交付的缺口。
\nIf unsure, use `/ads` or describe the task. A shortcut does not decide business policy for you; existing decisions carry forward, and only material gaps need clarification.\n\n不必说出“需求”才使用它。办公中的目标、规则、权限或流程改变也可以进入；明确的翻译、排版、抄录等任务由对应工具直接完成。隐式命中取决于宿主与模型，需要稳定调用时显式写出技能名。\n\n\n\n## 先看实际产物｜See the outputs\n\n| 示例 / Example | 看什么 / What to look for |\n|---|---|\n| **[最小需求卡 / Minimal requirement card](examples/minimal-v5/requirement-card.md)** · [运行说明 / Run it](examples/minimal-v5/README.md) | 给列表增加筛选：范围、权限、时间含义、异常和验收如何写在一起。 / A list filter with scope, permissions, time semantics, failure behavior and acceptance. |\n| **[模块化 PRD / Modular PRD](examples/medium-review-handoff/requirement.md)** | 两个业务模块各自写清入口、权限、状态、失败恢复和正反用例，附产品、前端、后端、测试的阅读入口。 / Two complete business slices with role entry points, guards, recovery and executable acceptance examples. |\n| **[交互评审原型 / Interactive review prototype](examples/medium-review-handoff/review-prototype.html)** | 本地打开即可操作工单指派与完成，查看就近实现要点、评审定位和独立记录；绑定上述 PRD。虚构教学数据，仅模拟业务服务。 / Open locally to assign and complete work orders, inspect implementation notes and record reviews separately. Bound to the PRD above; fictional data and a local service simulator. |\n\n需要从可运行起点改写评审原型，可用[示例构建器](references/review-workspace.md)拆分编辑后合成单 HTML，并自动绑定 PRD 摘要。它提供工作样例，不会自动生成你的业务或证明需求已验收。
\nThe [example builder](references/review-workspace.md) splits the working review example into editable files and rebuilds one HTML bound to its PRD. It is a starting point, not automatic business generation or acceptance evidence.\n\n
\n常见问题：小改、存量材料、原型与验收｜Quick FAQ\n\n- **只有一句话，或只改一个字段，也能用吗？** 能。明确小改直接交付差异与验收；模糊想法先找关键决定，不要求全套 PRD、等级或流程。 / **Can I start with an idea or one field?** Yes. Clear edits need a bounded change and acceptance; vague ideas need the relevant decision first.\n- **已有 PRD 或 HTML，要重做吗？** 读取相关基线，从当前阶段继续；继承有效决定并保护未取消的功能。 / **Must I rewrite existing work?** No. Continue from the relevant baseline, preserving valid decisions and existing scope.\n- **原型和评审态是否必交？** 按目标提供；需要原型时默认可操作产品态，双态评审按需启用。 / **Are prototypes and review mode mandatory?** They follow the task. Requested prototypes are interactive; dual-mode review is optional.\n- **检查 PASS 就能上线，领域资料能直接当政策吗？** 都不能。PASS 只覆盖已运行的检查；规则适用性、真实系统和验收各需证据。编码与上线交给相应工作流。 / **Does PASS authorize launch or adopting a domain policy?** No. Applicability, implementation and acceptance need their own evidence; coding and deployment use their respective workflows.\n\n
\n\n\n\n## English guide\n\n### Value for your role\n\n| Role | What you can take into the next conversation |\n|---|---|\n| **Junior PM** | The questions that matter, bounded scope and a reviewable card or PRD. |\n| **Senior PM / product lead** | Problem evidence, options, a minimal validation, current decisions and change impact. |\n| **Business / presales / delivery / design** | Business behavior to confirm, an interactive product prototype and explicit open decisions. |\n| **Frontend engineer** | Interaction paths, permissions, visible states and success/failure outcomes. |\n| **Backend engineer / architect** | Data authority, allowed transitions, side effects, recovery and integration boundaries. |\n| **QA / acceptance reviewer** | Positive and negative cases, boundary scenarios, regression scope and missing evidence. |\n| **Coding agent** | Current rules, source references, unresolved decisions and a scope it can resume. |\n\nFor consumer products, business and government systems, and AI-native workflows. These roles share one business agreement. Existing approved PRDs or requirement systems can remain the authoritative location.\n\n### What 5.5 focuses on\n\n1. **Find the decision that matters now.** Start with the outcome, affected people and facts. Compare options and the smallest useful validation. A supported recommendation to investigate, defer, reduce scope or decline can complete the analysis; changing requirement status still requires the relevant authority.\n2. **Make the specification tangible and the review concrete.** The PRD explains business rules, the prototype demonstrates interactions and outcomes, and acceptance criteria define how to judge them. “May publish after approval” must distinguish permission from publication in the rule, button behavior and acceptance case. Optional dual-mode review places rules, boundaries and acceptance beside the current product context while you operate it. Resolve critical business choices while leaving legitimate engineering choices open.\n3. **Keep meaning consistent through change.** Follow concrete dependencies across writers, readers, entry points, metrics and existing records. Separate candidate impact from verified impact, and keep the PRD, prototype and handoff tied to the same rule.\n\nClear local edits can be completed directly. Complex work loads only the relevant guidance. You do not need to select a delivery tier or fill every template. Existing materials let you enter at the current stage. Product prototypes are interactive by default; review markers and a dual-mode workspace are optional. [Explore the examples ↑](#examples)\n\n### Quick start\n\n**Start with the work you have.** After [installing the skill](#install), paste this into your agent or replace it with your own task:\n\n```text\nUse ai-delivery-spec: add an \"Enabled only\" filter to the existing list.\nReuse the existing enabled status, show all records by default, and preserve permissions.\nSpecify the rules and acceptance criteria for this change.\n```\n\nExpect the change scope, filter meaning, success and failure behavior, and testable acceptance criteria. Confirmed material carries forward. Missing business decisions stay explicit. A small change may need only a requirement card or a short change note.\n\nAttach an existing PRD, screenshot, HTML prototype, customer feedback or change request if you have one. The skill checks how each source relates to the decision. File access, prototype creation and validation depend on your host's available tools.\n\n### Choose your starting point\n\n| Your task | A prompt to copy |\n|---|---|\n| **Decide whether to build** | “Use ai-delivery-spec: users say this workflow is slow. Identify plausible causes and the smallest validation that would change our choice.” |\n| **Write requirements / a PRD** | “Use ai-delivery-spec: turn these confirmed materials into a PRD with roles, rules, exceptions and acceptance criteria.” |\n| **Create an interactive prototype** | “Use ai-delivery-spec: build a working product prototype from this PRD, including the main path and key failure outcomes.” |\n| **Change an existing system** | “Use ai-delivery-spec: separate approval from publishing. Identify affected existing records, screens, permissions and acceptance criteria.” |\n| **Review or hand off** | “Use ai-delivery-spec: review this specification as an engineering and QA receiver. Find critical business choices that still require guessing.” |\n| **Office workflows and spreadsheet rules** | “Use ai-delivery-spec: flag overdue reimbursements in this tracker. Clarify the start date, exceptions and responsible person before the spreadsheet tool implements it.” |\n\nUse the [four bilingual shortcut prompts](#shortcuts) for `/ads` (general), `/dig` (discovery), `/prd` (specification) and `/proto` (prototyping). Native slash-command support depends on the host; the explicit skill-name prompts work as conversation requests without requiring separate shortcut registration.\n\nYou do not need to say “requirement.” Changes to office goals, rules, permissions or workflows also apply. Straightforward translation, formatting and transcription can go directly to their tools. Implicit selection depends on the host and model; name the skill explicitly when you need a reliable invocation.\n\n\n\n## 按需深入｜Go deeper when needed\n\n| 当前需要 / Need | 入口 / Guide |\n|---|---|\n| 判断问题、澄清、比较方案 / Problem framing and options | [澄清与探索 / Discovery](references/discover.md) |\n| 写规则、做需求处置、维护基线 / Specification and decisions | [可实施规格 / Specification](references/specify.md) · [需求处置 / Lifecycle](references/lifecycle.md) |\n| 存量盘点、交互原型、双态评审 / Prototypes and reviews | [原型 / Prototyping](references/prototype.md) · [评审工作区 / Review workspace](references/review-workspace.md) |\n| 变更、验收、跨会话接续 / Changes, acceptance and continuity | [变更与验收 / Change and acceptance](references/change-acceptance.md) · [上下文 / Context](references/context.md) |\n| 交通、CRM、OA、数仓、教育、医疗、媒资知识、AI Native / Domain knowledge | [领域覆盖与证据 / Domain coverage](references/domain-coverage.yaml) |\n| 命令、宿主适配与排错 / Tools and troubleshooting | [阶段与工具 / Stages](references/stages.md) · [宿主适配 / Adapters](references/tool-adapters.md) · [排错 / Troubleshooting](references/troubleshooting.md) |\n\n领域资料按需读取，其经验与成熟度见覆盖表；项目规则仍须核实来源和适用性。
\nDomain references load on demand. The coverage file records their evidence and maturity; project rules still need applicable, authoritative sources.\n\n中英文关键词通过小型术语表检索同一份领域原文。回答跟随用户语言，法规名称与来源保留原文；中国法规、其他法域和国际标准分别核实，不能按提问语言选择适用法律。
\nChinese and English terms search the same source text through a curated glossary. Responses follow your language while preserving original source titles. Verify Chinese law, other jurisdictions and international standards separately; language does not select the applicable law.\n\n
\n可选检查工具与命令｜Optional checks and commands\n\n在技能目录内运行，使用 Python 3.10+。这些是交付检查工具；日常澄清与小改可直接在对话中完成。
\nRun from the skill directory with Python 3.10+. Use these tools at relevant delivery checkpoints; everyday clarification and small edits can stay in the conversation.\n\n```bash\npython -m pip install -r scripts/requirements.txt\npython scripts/ai_delivery_spec_cli.py version\npython scripts/ai_delivery_spec_cli.py check\npython scripts/ai_delivery_spec_cli.py triage --input examples/minimal-v5/intake.yaml --format json\npython scripts/ai_delivery_spec_cli.py gate --profile prd --prd examples/minimal-v5/requirement-card.md --stage specify\n```\n\n按任务需要使用以下命令；`analysis.md`、`app.html`、`old-app.html` 换为你的文件：
\nUse these as needed; replace `analysis.md`, `app.html` and `old-app.html` with your files:\n\n```bash\npython scripts/ai_delivery_spec_cli.py gate --profile prd --prd analysis.md --stage explore\npython scripts/ai_delivery_spec_cli.py query-domain --search confidence --limit 8\npython scripts/ai_delivery_spec_cli.py query-domain --search 完成率 --limit 8\npython scripts/ai_delivery_spec_cli.py query-domain --domain ai-native --section \"Metric / Indicator Governance\"\npython scripts/ai_delivery_spec_cli.py query-domain --domain medical-hospital-it --section \"Policy / Privacy Constraints\" --source-detail full --language en-US\npython scripts/ai_delivery_spec_cli.py gate --profile prototype --prototype app.html --prototype-baseline old-app.html\n```\n\n- **阶段 / Stage**：分析建议用 `explore` 检查；不传 `--stage` 仍默认 `baseline`。建议暂缓不会自动降低检查阶段或改变需求状态。 / Use `explore` for analysis. Omitting `--stage` retains the `baseline` default; a deferral recommendation does not change the stage or lifecycle.\n- **风险 / Risk**：`triage` 读取声明和部分正文线索，只给建议；未声明 `ai_write_scope` 表示未知，正文自动写回风险仍会独立提示。 / Triage returns advice using declarations and bounded text cues. Missing `ai_write_scope` means unknown; write-back cues are checked separately.\n- **变更 / Impact**：`impact` 接受顶层 `seed_refs: [REQ-A]` 或 `request.seed_refs`，同时给出时必须一致。图上的相关对象先作为候选核实。 / Impact accepts either seed location; both must agree if supplied. Related graph objects remain candidates until their dependencies are verified.\n- **领域 / Domains**：也可用 `--section 指标`。搜索命中是候选线索，不能直接变成项目已批准规则。 / Chinese section aliases are supported. Search hits are leads, not approved project rules.\n\n搜索结果给出扩展词、原文位置及 literal/alias 命中方式；它是有界关键词检索，未命中也可能只是术语表未覆盖。`--source-detail full` 可查看来源 URL、法域及适用范围，原文章节不会由脚本自动翻译。
\nResults show expanded terms, source locations and literal/alias matches. This is bounded keyword retrieval; zero hits may mean a vocabulary gap. `--source-detail full` exposes source URLs, jurisdictions and applicability. The script preserves source passages without automatically translating them.\n\n常用中文词可检索已有台账、报销和里程知识；“活跃”按客户/企业/用户相关短语召回，不等同于所有 `active` 状态。账本也保留 JS 动态声明候选及来源；候选进入盘点不代表已经渲染，门禁保留相应 GAP。UNK 表头未被识别时提示定位问题；显式无效状态、无依据关闭与真实冲突仍分别检查。
\nChinese ledger, reimbursement and mileage queries retrieve existing domain passages. Activity terms target relevant customer, enterprise or user phrases. Interaction ledgers retain JS declaration candidates with their origins; unresolved rendering remains a GAP. Unlocated unknown-status columns are distinguished from explicitly invalid statuses, unsupported closure and conflicting declarations.\n\n
\n\n
\n从 5.4.x 升级｜Upgrading from 5.4.x\n\n5.4.x 产物可继续读取，稳定 ID 和已批准事实保留。`artifact_mode` 使用 `direct/card/prd`；旧 L0–L4 在 PRD 中只作呈现提示，不能覆盖显式模式或自动提高风险、证据要求。专业原型、评审、Truth、handoff 与执行状态工具保留其版本化 Schema，不自动迁移旧产物，也不成为日常任务的默认依赖。\n\n5.4.x artifacts remain readable with stable IDs and approved facts preserved. `artifact_mode` uses `direct/card/prd`; legacy tiers are presentation hints in PRDs. Specialized prototype, review, Truth, handoff and execution-state tools retain their versioned schemas and remain optional. Old artifacts are not automatically migrated.\n\n旧未知项状态 `partial`、`in_progress` 及“部分关闭/部分解决”按 `open` 理解：只要仍有未决部分，就按其依赖范围与阻断阶段处理。同一未知项的相同诊断合并，冲突声明仍保留。其他未识别状态必须明确迁移，不能被当作关闭。
\nLegacy `partial`, `in_progress` and equivalent Chinese partial-resolution labels are treated as `open`. Remaining decisions retain their scope and blocking stage. Repeated identical findings are merged; conflicting declarations remain visible. Other unknown status values require explicit migration and never count as closed.\n\n接入脚本需注意：5.5 triage 使用 `recommendation / artifact_mode / risk_facets / governed`，移除旧 tier/mode 推导结果键；旧 Markdown PRD 入口复用新内核，退出码可能改变；旧 `validate_prd_quality.py --domain-rules` 已退役，领域约束通过显式 custom gate 或领域工具处理。\n\nScript consumers: update to the triage keys above. Old derived tier/mode output keys are removed; legacy Markdown PRD entry points share the new engine, so exit codes may differ. The old `--domain-rules` keyword check is retired; use explicit custom gates or domain tools. See the [changelog](CHANGELOG.md) for version history.\n\n
\n\n
\n检查能证明什么；如何维护｜Evidence and maintenance\n\n正文检查对状态权威、指标口径、恢复路径、空值/陈旧、权限边界与变更传播做有界抽查，返回带位置的待核实 GAP。填写评阅 pass 不能覆盖正文疑点。`PASS` 只表示实际执行的确定性检查未发现对应阻断或缺口；业务语义、来源授权、浏览器交互、真实实现和客户验收仍需各自的证据。\n\nText checks sample known ambiguity patterns in state authority, metrics, recovery, null/stale values, permissions and change propagation. Findings are located GAPs for review; a declared review pass cannot suppress them. A deterministic `PASS` covers only the checks executed. Business meaning, source authority, browser behavior, implementation and customer acceptance require their own evidence.\n\n对“补考成绩取最新”等取值政策，依据未显式出现时只给 WARN 提示；附近同时存在未决声明才给 GAP。提示不能鉴定授权真伪，也不要求为每个界面默认值补决策表。
\nFor selected policies such as using the latest retake score, an absent explicit basis produces a WARN advisory; a nearby unresolved decision produces a GAP. This does not authenticate authority or require decision tables for ordinary UI defaults.\n\n本 Skill 管需求与其产物，连接工程反馈；排期、编码、部署与运营由相应工作流负责。公共反馈与示例请使用脱敏材料。
\nThe skill manages requirements and their artifacts, incorporating engineering feedback. Scheduling, coding, deployment and operations belong to the relevant workflows. Use sanitized material in public examples and feedback.\n\n维护者可在**完整源码仓库**运行以下命令；构建发布包要求干净 Git 来源。运行包不包含 `maintainer/`，其中 `check` 只检查实际携带的文件与契约。
\nMaintainers can run these commands in the **full source repository**. Release packaging requires a clean Git source. Runtime packages exclude `maintainer/`; their `check` covers the files and contracts actually shipped.\n\n```bash\npython -m pip install \"pytest>=8,<9\"\npython scripts/ai_delivery_spec_cli.py check --profile release\npython maintainer/tools/build_runtime_package.py --release --check --output dist/ai-delivery-spec-5.5.2.zip\n```\n\n兼容验证入口 `scripts/validators/validate_prd_quality.py`、`scripts/validators/validate_unified_prd.py` 可手动执行，复用当前内核。 / These legacy validators remain available for manual execution.\n\n
\n\n\n\n## 一起把需求做得更清楚｜Join the community\n\n欢迎**中文或英文**提问、反馈真实使用问题、分享脱敏案例、完善翻译或贡献领域知识。复现材料、预期与实际差异，会帮助我们判断该修模型指引、工具还是示例。\n\n**Chinese and English** questions, bug reports, sanitized examples, translations and domain contributions are welcome. Reproduction steps and expected-versus-observed behavior help identify what needs to change.\n\n- **[GitHub Issues：反馈与交流 / Feedback and questions](https://github.com/franklinxkk/ai-delivery-spec/issues)**\n- **[ClawHub：社区安装入口 / Community listing](https://clawhub.ai/franklinxkk/skills/ai-delivery-spec)** · **[SkillHub：中文社区入口 / Chinese community listing](https://skillhub.cn/skills/user_12c92261/ai-delivery-spec)**\n- **[参与贡献 / Contributing](https://github.com/franklinxkk/ai-delivery-spec/blob/main/.github/CONTRIBUTING.md)** · **[版本记录 / Changelog](CHANGELOG.md)**\n\n如果它帮助你澄清了一条需求、发现一次关键分歧，欢迎给项目一个 **[Star ⭐](https://github.com/franklinxkk/ai-delivery-spec)**，让更多产研同伴找到它。
\nIf it helped clarify a requirement or expose a critical ambiguity, a **[Star ⭐](https://github.com/franklinxkk/ai-delivery-spec)** helps more product teams discover it.\n\n[Apache 2.0](LICENSE) · [返回顶部 / Back to top](#ai-delivery-spec-552)\n\nFile v5.5.2:_meta.json\n\n{\n \"ownerId\": \"kn7789n9h3n4chvd2jehwxy2jh88w52h\",\n \"slug\": \"ai-delivery-spec\",\n \"version\": \"5.5.2\",\n \"publishedAt\": 1790726245062\n}\n\nFile v5.5.2:references/change-acceptance.md\n\n# 变更、交接反馈与验收\n\n## 从一个有效决定开始\n\n记录来源、授权、基线、前后变化与影响种子；继承已有决定。写入前核对版本，按主题合并并发变化，不以最后写入者胜出。\n\n一条规则只改一个权威定义。页面、流程、原型说明、AC 与交接引用同步。显示序号变化不等于业务身份变化；需要替换身份时保存旧 ID 与替代关系，不改写历史。\n\n## 找消费者并保留依据\n\n从规则找到写入者、读取者、指标、操作入口、事件/通知、旧数据、进行中流程、验收与接收者。每项说明依赖链；区分直接、传递、仅回归和有理由的无影响。\n\n推测消费者记为候选，核实后更新同项及依据，排除保留理由。核实依赖不代表批准修改；高影响候选未核实就限制相应完成声明。\n\n`impact` 仅沿已登记关系找候选，暴露缺种子、未解析关系和深度截断；命令见 [工具](stages.md)，输入见 [变更模板](templates/change-request-template.yaml)。复杂审计才用 change-package Schema；小改写在当前产物。\n\n已有对象如何处理须依授权决定；不能默认全部迁移或沿用。记录哪些证据仍适用、哪些失效以及谁接收新版本。工程反馈回链为规格缺陷、实现缺陷、来源失效或新增范围，不把它们混成同一种返工。\n\n## 证据按主张、范围、版本成立\n\n| 证据 | 能证明 | 不能替代 |\n|---|---|---|\n| 静态检查 | 指定声明、引用、结构及可确定约束 | 自然语言业务真值和实际行为 |\n| 语义评阅 | 指定对象的来源核对、反例及处置 | 运行验证；两模型一致也不是真值 |\n| 浏览器原型 | 指定环境的操作、可见/Mock结果 | 真实接口、持久、并发与生产权限 |\n| 真实系统 | 指定版本环境中的实际对象与结果 | 客户批准规则或验收签署 |\n| 业务/客户确认 | 有权人在对应范围作出的决定或接受 | 实现已完成、范围外能力或上线 |\n\n关键主张绑定范围、版本、证据类型、环境/责任人、结果与位置。验证后修改重验受影响范围；未执行、缺来源、不适用分开。\n\n断言数、唯一 AC 数和已验证行为分开；编号注释不算覆盖。按同一输入核对来源→PRD→原型实际操作→实现结果，检查对象/字段含义、状态、权限、口径、拒绝和副作用，允许有明确映射的不同命名。AC 前提须能独立造数或重置；删除对象使旧 AC 失效。声明可配置的规则，用改值前后样本验证各消费者响应及生效时点，不把所有阈值升级成配置平台。\n\n复用 ARUN/EVD；真实验收保留 acceptance_ref、requirement_refs、result、actual_result、evidence_refs，见 [验收模板](templates/acceptance-run-template.yaml)。本地证据可解析且不越出记录目录；远程链接或字符串不证明执行。\n\n## 关闭\n\n实际验证覆盖指定主路径及高价值拒绝/恢复；验收接受需相应有权签署。残留问题明确范围、责任及复议条件；有条件批准到期重新处理，不用条件措辞掩盖未完成的关键规则。\n\n需求通过验收不等于已经发布。外部上线、运营和结果反馈只记录引用，必要时作为新来源或 CHG 返回；不由需求状态自动驱动生产操作。\n\nFile v5.5.2:references/context.md\n\n# 大任务、上下文与交接\n\n材料多或跨会话时读取；摘要用于导航，不替代原始依据。\n\n## 切片与恢复\n\n登记来源、版本、授权和范围，按角色完整路径或业务切片读取，保留权限、状态、指标、恢复与验收。存量走查列明受影响页面、页签、弹层、角色及未访问范围；抽样须注明，不能以首页代替全系统。\n\n长任务在现有检查点记录目标、来源/版本、产物、有效决定、未决依赖和下一步；小改不建。恢复先核实文件及版本，限定冲突范围再合并。压缩保留未完成范围，不把进展包装成完成。\n\n确有机器消费者才使用多文件 Truth；按模块生成、校验引用并重验。大型 PRD 同样切片编写，事实只维护一处。\n\n## 按需取领域与结构化资料\n\n依据项目及有权来源确定领域；通用词不自动决定领域。知识包是候选参考，其成熟度不是验收证据。\n\n```powershell\npython scripts/query_domain.py --domain oa\npython scripts/query_product_truth.py --help\npython scripts/plan_context.py --help\n```\n\n复杂项目可用 context-plan Schema/plan_context 估算切片；预算是建议，不强制新增计划或全量审查。\n\n## 给研发、测试与 Coding Agent 的交接\n\n交接引用目标、规则、输入输出、数据权威、权限、失败恢复、验收和禁止推断项的原位置及版本。业务语义依赖的来源须可获得，或指出缺谁提供的什么信息；Mock 数据不能假装已接通来源。技术方案由研发负责。\n\n各输出注明源版本及待同步范围，手工投影不覆盖基线；新决定使旧证据失效时标明范围。多人接力明确边界和依赖负责人，不虚构责任人。\n\n需要长期机器执行时使用 [handoff 模板](templates/agent-handoff-manifest-template.yaml)；\n需要强追溯快照时使用 execution-state 工具。这些是明确选择的高级合同，不能反向变成日常 PRD 的必填项。\n\n```powershell\npython scripts/manage_execution_state.py --help\npython scripts/ai_delivery_spec_cli.py gate --profile handoff --prd PRD.md --prototype app.html --manifest handoff.yaml\n```\n\n执行边界使用环境、权限和凭据引用，不写凭据本身。需求批准不代替外部操作授权。接收方反馈区分规格缺失、来源不可用、已登记未知、投影漂移、实现偏差、技术选择和环境问题，回写对应层；不能把研发忽略已有规则也算成需求遗漏。\n\nFile v5.5.2:references/discover.md\n\n# 判断、探索与澄清\n\n先辨认用户给的是问题、目标、候选方案、批准变更还是观察。对明确授权的小改继承决定；对模糊机会，找当前真正需要判断的事项。\n\n## 从证据推进\n\n观察发生在谁、什么场景、什么对象上；区分购买者、管理者与实际使用者。把可能原因当作假设，找能够改变方案选择的证据。比较有实质差异的选择以及维持现状的代价；在形成下一决定所需信息足够时停止。\n\n诊断关系是“观察→可能原因→会改变选择的证据→最小验证→推荐决定”。直接写入现有 problem brief 或 solution sketch，不生成一份额外诊断报告。可复用 [problem-brief-template.md](templates/problem-brief-template.md) 和 [solution-sketch-template.md](templates/solution-sketch-template.md)，按当前目标裁剪。\n\n战略、系统、行为和反方视角是可选思考角度，不是四轮问卷。只描述有依据的动机与激励，不给人做心理诊断。建议扩大方案前检查是否有更小、可逆且能验证判断的选择。\n\n## 决定与未知\n\n先查直接相关的存量能力、规则和开放问题；已有答案且适用就继承。优先问会改变范围、权威、业务结果或验收的主问题；少量独立小项可合并一轮，不依赖尚未得到的答案。其他工作继续，轮数不证明充分。\n\n澄清只给必要上下文与当前问题；选项有依据时给推荐及理由，不将推荐当决定。回答就地记入原产物的决定/未知来源，不另交问答日志或未来问卷。\n\n用户最新有效决定在授权主题内覆盖旧内容。未合入正式文档不意味着需要重新批准。合同、法规或系统权威是否适用须核实，不能按文件新旧或长度自动排序。\n\n未知写明影响、何时必须决定、谁有权决定及当前可继续的边界。无人可确认时记录待指定，不虚构负责人。风险接受需要具名授权及范围/期限，不能由模型代签。\n\n退路不得选定未决政策。例如补考取值未定，暂停依赖该选择的结论或保留待决分支；不能将“取最新”标成保守默认。晚到数据、冻结、权限例外和历史迁移同理。\n\n## 暂缓、不做与收敛\n\n建议暂缓、不做、缩范围记在当前产物或 DEC 中，标记 proposed；明确获权后才 confirmed。记录理由、证据、适用范围及复议条件。无正式 REQ 时不为记录建议强制准入。\n\n分析任务可以在给出有依据的建议后完成；需求是否取消、外部承诺是否撤回、系统是否停止是另外的授权与执行。已有用户决定不重复问。状态细则见 [lifecycle.md](lifecycle.md)。\n\n停止依据当前目标所需的决定，不是先关闭未来所有 P0 再允许得出“暂缓”。若当前成功就是确认缺少价值证据，应交付最小验证、责任/复议条件及未定范围，保留实施缺口；检查时明确 `--stage frame` 或 `--stage explore`。\n\n当前决定足够明确后，自动回到用户要的 PRD/原型/变更交付。用户要求完整结果时不把澄清清单当终点。\n\nFile v5.5.2:references/domain-coverage.yaml\n\nschema_version: 5.1.1\n# Retrieval aliases only; a match never establishes project applicability.\nsearch_aliases:\n 台账: [ledger]\n 报销: [reimbursement, expense, expense claim, expense workflow]\n 里程: [mileage]\n 里程碑: [milestone]\n 活跃: [active customer, active enterprise, active user]\n 活跃客户: [active customer]\n 活跃企业: [active enterprise]\n 活跃用户: [active user]\n 合同: [contract]\n 商机: [opportunity, opportunities]\n 客户: [customer]\n 支付: [payment]\n 回款: [payment, collection]\n 审核: [approval, review]\n 审批: [approval]\n 完成率: [completion rate, completion]\n 置信度: [confidence]\n 补考: [retake, resit]\n 成绩: [grade, score]\n 高校: [university, college, campus]\n 学籍: [student status, enrollment]\n 教育数据: [education data]\n 数据分类分级: [data classification, classification, grading]\n 交通: [transport, traffic]\n 道路运输: [road transport]\n 继续教育: [continuing education]\n 培训档案: [training record, training, archive]\n 安全考核: [safety assessment, safety examination]\n 医院: [hospital]\n 电子病历: [electronic medical record, EMR]\n 病历: [medical record, EMR]\n 医疗数据: [health data, medical data]\n 医疗: [medical, clinical]\n 互认: [mutual recognition]\n 权限: [permission, access control]\n 审计: [audit]\n 隐私: [privacy]\n 个人信息: [personal information]\n 流程: [workflow]\n 办公: [office, OA]\n 报表: [report]\n 指标: [metric, indicator]\n 数据权威: [authority, source of truth]\n 分母: [denominator]\n 血缘: [lineage]\n 人工确认: [human confirmation, human approval, human gate, human-gate]\n 工具授权: [tool authorization, tool permission, tool credentials]\n 视频: [video]\n 转写: [transcription, transcript]\n 时间码: [timecode, time code, timestamp]\ngenerated_from: domain-pack metadata and evaluation evidence\nevaluation_index: maintainer/evals/domain-fixtures.yaml\nclaim_policy:\n rule: Domain assurance has explicit sourced, contract, behavioral, expert and audit levels.\n practice_rule: Production practice proves real delivery use but does not promote reusable-pack maturity.\n contract_tested_requires:\n - sourced knowledge\n - passed deterministic contract evaluation\n behavior_validated_requires:\n - contract-tested maturity\n - passed fresh-agent behavioral evaluation\n expert_reviewed_requires:\n - behavior-validated maturity\n - accountable expert review\n - project-sampled scenarios\n audited_requires:\n - expert-reviewed requirements\n - independent audit evidence\n simulated_review_is_expert_review: false\n\ndomains:\n - domain_id: traffic\n name: Traffic Safety and Road Transport SaaS\n knowledge_file: references/domains/domain-traffic.md\n applies_when: [road transport, traffic safety, regulator supervision, transport enterprise]\n does_not_apply_when: [generic workflow with no transport object]\n coverage:\n knowledge: sourced\n scenarios: project_sampled\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: production_practiced\n evidence:\n - id: EV-DOM-TRAFFIC-PRACTICE\n kind: owner_attestation\n location: maintainer/evals/evidence/production-practice-attestation-2026-07-12.yaml\n status: passed\n owner: project and skill maintainer\n - id: EV-DOM-TRAFFIC-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-TRAFFIC-INSPECTION-2025\n status: current\n owner: domain maintainer\n - id: EV-DOM-TRAFFIC-KNOWLEDGE\n kind: project_sample\n location: maintainer/examples/traffic-regulatory-change-v5\n status: current\n owner: domain maintainer\n - id: EV-DOM-TRAFFIC-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-TRAFFIC-RISK-CLOSE\n status: passed\n owner: domain maintainer\n known_gaps:\n - National source coverage was refreshed through 2026-08-10, including separate current passenger and dangerous-goods safety-management baselines; provincial/local implementation, project procurement and customer source-system rules remain project-specific.\n - Nine-industry depth has not been behaviorally evaluated.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\n - domain_id: crm\n name: CRM and Customer Response\n knowledge_file: references/domains/domain-crm.md\n applies_when: [lead, opportunity, customer, contract, payment, ticket, renewal]\n does_not_apply_when: [generic contact list without customer lifecycle]\n coverage:\n knowledge: sourced\n scenarios: mocked\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: production_practiced\n evidence:\n - id: EV-DOM-CRM-PRACTICE\n kind: owner_attestation\n location: maintainer/evals/evidence/production-practice-attestation-2026-07-12.yaml\n status: passed\n owner: project and skill maintainer\n - id: EV-DOM-CRM-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-CRM-COMPLAINT-10002\n status: current\n owner: domain maintainer\n - id: EV-DOM-CRM-KNOWLEDGE\n kind: mock_scenario\n location: maintainer/evals/domain-fixtures.yaml#FX-CRM-LEAD-REVENUE\n status: current\n owner: domain maintainer\n - id: EV-DOM-CRM-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-CRM-LEAD-REVENUE\n status: passed\n owner: domain maintainer\n known_gaps:\n - CPQ, complex account hierarchy, customer success, and partner conflict need deeper evaluation.\n - Current fixtures are evaluation inputs, not production evidence; Microsoft, Salesforce and HubSpot sources are vendor product patterns, not independent outcome proof.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\n - domain_id: education-it\n name: Education IT\n knowledge_file: references/domains/domain-education-it.md\n applies_when: [academic affairs, student affairs, course, training, learning, assessment]\n does_not_apply_when: [generic content publishing with no learner lifecycle]\n coverage:\n knowledge: sourced\n scenarios: project_sampled\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: production_practiced\n evidence:\n - id: EV-DOM-EDU-PRACTICE\n kind: owner_attestation\n location: maintainer/evals/evidence/production-practice-attestation-2026-07-12.yaml\n status: passed\n owner: project and skill maintainer\n - id: EV-DOM-EDU-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-EDU-CLASSIFICATION-0661\n status: current\n owner: domain maintainer\n - id: EV-DOM-EDU-KNOWLEDGE\n kind: project_sample\n location: maintainer/examples/publishing-learning-v5\n status: current\n owner: domain maintainer\n - id: EV-DOM-EDU-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-EDU-COURSE-VERSION\n status: passed\n owner: domain maintainer\n known_gaps:\n - Current knowledge is higher-education heavy.\n - Vocational education, enterprise training, one-book-one-code, channel authorization, and content rights require deeper coverage.\n - The exact national term “AI教育三进战略” was not verified from a primary source and must remain an unknown until its issuer/document is supplied.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\n - domain_id: oa\n name: Collaborative Office and OA\n knowledge_file: references/domains/domain-oa.md\n applies_when: [approval, official document, meeting, supervision, office knowledge]\n does_not_apply_when: [single approval action with no OA object or lifecycle]\n coverage:\n knowledge: sourced\n scenarios: mocked\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: knowledge_only\n evidence:\n - id: EV-DOM-OA-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-OA-ESIGN\n status: current\n owner: domain maintainer\n - id: EV-DOM-OA-OFFICIAL-DOC\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-OA-OFFICIAL-DOC\n status: current\n owner: domain maintainer\n - id: EV-DOM-OA-ARCHIVE\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-OA-ARCHIVE-39362\n status: current\n owner: domain maintainer\n - id: EV-DOM-OA-VENDOR-PATTERNS\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-OA-WEAVER-WHITEPAPER\n status: current\n owner: domain maintainer\n - id: EV-DOM-OA-OPEN-COMPONENTS\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-OA-DINGTALK-OSS\n status: current\n owner: domain maintainer\n - id: EV-DOM-OA-KNOWLEDGE\n kind: mock_scenario\n location: maintainer/evals/domain-fixtures.yaml#FX-OA-APPROVAL-RECALL\n status: current\n owner: domain maintainer\n - id: EV-DOM-OA-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-OA-APPROVAL-RECALL\n status: passed\n owner: domain maintainer\n known_gaps:\n - Vendor whitepapers and cases are product/scenario evidence, not independent outcome proof or project truth.\n - DingTalk and Feishu public repositories validate exact SDK/demo/integration components only, not core-product open-source status.\n - OA pack still needs fresh-agent behavioral runs and accountable OA practitioner review before behavior_validated or expert_reviewed promotion.\n - The 2026 archive standard project guide is change-watch evidence, not a published standard.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\n - domain_id: medical-hospital-it\n name: Medical and Hospital IT\n knowledge_file: references/domains/domain-medical-hospital-it.md\n applies_when: [patient, encounter, medical record, order, report, clinical workflow]\n does_not_apply_when: [generic appointment or CRM flow without clinical decisions]\n coverage:\n knowledge: sourced\n scenarios: mocked\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: knowledge_only\n evidence:\n - id: EV-DOM-MEDICAL-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-MEDICAL-EMR-2025\n status: current\n owner: domain maintainer\n - id: EV-DOM-MEDICAL-KNOWLEDGE\n kind: mock_scenario\n location: maintainer/evals/domain-fixtures.yaml#FX-MEDICAL-ORDER-CANCEL\n status: current\n owner: domain maintainer\n - id: EV-DOM-MEDICAL-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-MEDICAL-ORDER-CANCEL\n status: passed\n owner: domain maintainer\n known_gaps:\n - High-risk clinical, legal, interoperability, privacy, and local-policy assertions require accountable review.\n - No production claim is allowed from simulated scenarios.\n - A consolidated “数智医院” evaluation transition has not been verified in a publicly accessible NHC replacement standard; preserve it as change watch only.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\n - domain_id: data-product\n name: AI and Data Product System\n knowledge_file: references/domains/domain-data-mart.md\n applies_when: [data registration, authorized operation, trusted data space, data product, high-quality dataset, data labeling, training data, evaluation data, preference data, data source, metric, lineage, semantic layer, BI, ChatBI, Data Agent]\n does_not_apply_when: [simple transactional form with no governed data product]\n coverage:\n knowledge: sourced\n scenarios: mocked\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: production_practiced\n evidence:\n - id: EV-DOM-DATA-PRACTICE\n kind: owner_attestation\n location: maintainer/evals/evidence/production-practice-attestation-2026-07-12.yaml\n status: passed\n owner: project and skill maintainer\n - id: EV-DOM-DATA-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-DATA-PROPERTY-REGISTER-2026\n status: current\n owner: domain maintainer\n - id: EV-DOM-DATA-KNOWLEDGE\n kind: mock_scenario\n location: references/domains/domain-data-mart.md\n status: current\n owner: domain maintainer\n - id: EV-DOM-DATA-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-DATA-METRIC-VERSION\n status: passed\n owner: domain maintainer\n known_gaps:\n - Formal provenance, digital-contract and OpenLineage mappings need executable examples.\n - Regional implementation details, sector rules, price/settlement, accounting judgment and cross-border applicability remain project evidence.\n - Fresh-agent behavior, data-domain expert review and production correctness are not proven by the expanded knowledge or static contracts.\n - Official platform, typical-case and sandbox evidence describes implementation patterns, not universal architecture or outcome proof.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\n - domain_id: ai-native\n name: AI Native and Agentic Systems\n knowledge_file: references/domains/domain-ai-native.md\n applies_when: [agent, tool calling, model decision, AI writeback, memory, evaluation]\n does_not_apply_when: [AI is only used to write documents or code]\n coverage:\n knowledge: sourced\n scenarios: mocked\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: production_practiced\n evidence:\n - id: EV-DOM-AI-PRACTICE\n kind: owner_attestation\n location: maintainer/evals/evidence/production-practice-attestation-2026-07-12.yaml\n status: passed\n owner: project and skill maintainer\n - id: EV-DOM-AI-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-AI-ISO-42001\n status: current\n owner: domain maintainer\n - id: EV-DOM-AI-KNOWLEDGE\n kind: mock_scenario\n location: references/domains/domain-ai-native.md\n status: current\n owner: domain maintainer\n - id: EV-DOM-AI-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-AI-TOOL-DENIAL\n status: passed\n owner: domain maintainer\n known_gaps:\n - Agent delegation, identity, memory governance, prompt injection, red-team, GB/Z 185 profiles and MCP 2026-07-28 migration need executable evaluations.\n - NIST initiatives and China agent-security standard projects are change-watch sources, not final binding standards.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\n - domain_id: media-knowledge\n name: Media Assets and Multimodal Knowledge\n knowledge_file: references/domains/domain-media-knowledge.md\n applies_when: [media asset, video archive, audio archive, live stream, multimodal knowledge, video search, moment retrieval, media RAG, AI clipping]\n does_not_apply_when: [decorative media upload with no managed lifecycle or semantic retrieval]\n coverage:\n knowledge: sourced\n scenarios: mocked\n contract_eval: passed\n behavioral_eval: not_run\n expert_review: not_reviewed\n maturity: contract_tested\n practice_status: knowledge_only\n evidence:\n - id: EV-DOM-MEDIA-SOURCES\n kind: official_source\n location: references/domains/domain-sources.yaml#KS-MEDIA-IPTC-VMH-1.7\n status: current\n owner: domain maintainer\n - id: EV-DOM-MEDIA-KNOWLEDGE\n kind: mock_scenario\n location: references/domains/domain-media-knowledge.md\n status: current\n owner: domain maintainer\n - id: EV-DOM-MEDIA-CONTRACT\n kind: contract_eval\n location: maintainer/evals/domain-fixtures.yaml#FX-MEDIA-MOMENT-AUTHORIZATION\n status: passed\n owner: domain maintainer\n known_gaps:\n - Fresh-agent behavior and accountable media-archivist, rights and privacy expert review have not run.\n - Project contracts, source systems, jurisdictions, languages, media corpus and downstream use remain project evidence.\n - Vendor product patterns do not prove project recall, precision, localization, latency, cost, safety or production suitability.\n - Static contracts do not prove preservation fidelity, legal rights, identity, authenticity or truth.\n last_verified_at: \"2026-08-29T00:00:00+08:00\"\n maintainer: ai-delivery-spec maintainers\n production_claim: prohibited\n\nfallback:\n mode: project_domain_capsule\n rule: Missing a dedicated domain pack never blocks generic delivery.\n required_outputs:\n - vocabulary\n - entities\n - state_machines\n - workflows\n - policies\n - source_register\n - unknowns\n - scenario_fixtures\n assertion_policy:\n verified: attributable to an authoritative or accountable source\n inferred: plausible but awaiting owner confirmation\n unknown: missing information changes scope, safety, compliance, data, or acceptance\n\nFile v5.5.2:references/domains/domain-ai-native.md\n\n# Domain: AI Native Product And Agentic Systems\n\nSource authority and freshness metadata: `references/domains/domain-sources.yaml`.\nCoverage and maturity: `references/domain-coverage.yaml`.\n\nUse this replaceable domain module for AI-native products, agent workflows,\nLLM-powered applications, multi-agent systems, copilots, AI assistants, RAG\nproducts, tool-using agents, evaluation platforms, prompt/model operations, and\nhuman-in-the-loop automation. A replacement `domain-*.md` must preserve the\nsame 15 section headings used here and in `domain-module-template.md`.\n\n## Contents\n\n- Domain Purpose\n- First-Principles Domain Lens\n- Vocabulary\n- Aggregates and Entities\n- Domain Events\n- State Machines\n- Metric / Indicator Governance\n- AI Context Sources\n- Content / Knowledge Assets\n- Core Workflows\n- Role Path Patterns\n- UI / Mobile Patterns\n- Policy / Privacy Constraints\n- Domain Test Scenarios\n- Evaluation Profile\n- Acceptance Checklist\n\n## Domain Purpose\n\n- Business outcome: deliver trustworthy AI-assisted or AI-core outcomes with\n clear user value, accountable human gates, measurable quality, and safe\n rollback.\n- Primary users: end user, expert reviewer, operations owner, AI engineer,\n product manager, risk/compliance reviewer, admin.\n- Sensitive areas: private context, prompts, retrieved documents, model output,\n tool credentials, writeback actions, evaluation data, and user feedback.\n- AI may optimize: intent understanding, retrieval, summarization,\n recommendation, reasoning support, drafting, routing, and workflow assistance.\n- AI must not autonomously decide: high-impact legal, financial, medical,\n safety, enforcement, account closure, payment, or irreversible business\n actions without explicit authorized human accountability.\n\n### Current Baseline and Change Watch (verified 2026-08-29)\n\n| Evidence class | Current signal | Requirement effect |\n|---|---|---|\n| design_guidance | `KS-AI-AGENT-IMPLEMENTATION-2026` adds current agent policy direction; `KS-AI-GBZ185-SERIES-2026` covers agent architecture, identity, description, discovery, interaction and tool calling; `KS-AI-MCP-2026-07-28` resets MCP around a stateless core and formal extensions | declare policy applicability, protocol/profile versions, identity/authority, capability metadata, discovery, auth, tool side effects, compatibility and migration tests instead of saying only “supports agents/MCP” |\n| binding_baseline | existing privacy, data-security, AI-content-labeling and anthropomorphic-interaction rules remain applicable by product scenario | resolve whether content, interaction, personal data and high-impact writeback rules apply; keep accountable human gates and audit evidence |\n| change_watch | `KS-AI-NIST-AGENT-STANDARDS-2026` and `KS-AI-AGENT-SECURITY-DRAFT-2026` signal stronger identity, permission, tool, data, log and shutdown controls | use them for threat modeling and re-verification only; initiatives, consultations and draft projects cannot create a compliance hard stop |\n| compatibility guard | MCP 2026-07-28 deprecates legacy lifecycle/SSE features with a migration window | never silently break older clients; record supported versions, deprecated paths, fallback and coexistence evidence |\n\n### AI Transformation Horizon\n\n| Axis | From -> toward | Non-negotiable product guard |\n|---|---|---|\n| product unit | isolated chatbot/agent -> governed agent capability deployed across workflows | capability identity, owner, purpose, tools, data scope, version and retirement are explicit |\n| architecture | prompt plus tools -> context plane + identity/authority + registry/discovery + runtime + memory + evaluation + human control | protocol support names exact version/profile; tool side effects and delegation chains remain inspectable |\n| operation | foreground request -> durable/event-driven/multi-agent execution | checkpoint, budget, cancellation, retry, conflict resolution, incident handling and human takeover are designed |\n| value loop | fluent output -> verified state change, accepted artifact, cycle-time/quality improvement and failure evidence | optimize business outcome within safety limits; do not use task volume or model confidence as proof |\n| ecosystem | proprietary plugins -> portable skills/tools/agents with declared trust boundaries | preserve export, interoperability, source authority and least privilege; open protocol does not equal trusted component |\n\n## First-Principles Domain Lens\n\nFirst-Principles Product Logic: AI-native product judgment must start from work physics, not tool hype. Claude,\nCodex, Trae, OpenClaw/Lobster-class tools, WorkBuddy-class internal assistants,\nand new agent platforms are delivery surfaces. They are not the product logic.\n\nUse these first principles before selecting features:\n\n| Principle | Product Question | Value Signal |\n|---|---|---|\n| Work outcome first | What role, decision, artifact, or state transition changes? | shorter cycle time, fewer defects, higher throughput, better decision evidence |\n| Context before model | What context must be correct, fresh, permitted, and cited? | lower hallucination, fewer reviewer edits, reusable knowledge assets |\n| Workflow before chat | Which repeated workflow can be compressed or redesigned? | fewer handoffs, less waiting, clearer owner/action closure |\n| Reversibility determines autonomy | If AI is wrong, can the action be reversed cheaply and safely? | safe automation scope and human-gate placement |\n| Evidence beats fluency | Can the output be traced to source, rule, eval, or reviewer decision? | acceptance confidence and auditability |\n| Human accountability remains | Who signs off on consequential decisions? | deployable governance instead of demo-only automation |\n| Eval is product telemetry | What golden cases, adversarial cases, and live metrics prove value? | launch readiness and regression control |\n| Ontology turns data into action | What business objects, links, actions, and rules must AI understand? | agents can reason over operations, not just documents |\n\nDurable AI-native trend map:\n\n| Trend Layer | Stable Capability | Product Risk If Ignored |\n|---|---|---|\n| Coding agents | spec -> plan -> tasks -> code -> tests loop | fast code with wrong product behavior |\n| Enterprise work agents | inbox/task/workflow orchestration across tools | fragmented assistants that do not close work |\n| Knowledge engineering | curated corpus, metadata, lifecycle, retrieval, feedback | confident answers over stale or unauthorized knowledge |\n| Ontology / semantic layer | object/link/action/rule model over business reality | agents cannot safely write back or explain decisions |\n| Tool protocol layer | MCP/API/tool registry, auth, schema, side-effect control | unsafe tool calls and hidden integration drift |\n| Evaluation and observability | golden cases, traces, feedback, rollout/rollback | demo success but production regression |\n| Human-in-the-loop ops | review queues, escalation, override, incident response | no accountable owner for AI mistakes |\n\n## Vocabulary\n\n| Term | Meaning | Source Of Truth |\n|---|---|---|\n| AI Feature | AI-supporting capability with a valid manual/deterministic path | Product Truth + runtime contract |\n| AI-Core Workflow | primary user outcome depends on model/tool orchestration | AI runtime contract |\n| Agent | bounded AI actor with goal, tools, memory/context, eval, owner | agent registry |\n| Tool | callable system/API/function with permission and side-effect scope | tool registry |\n| Prompt | versioned instruction/context template | prompt registry |\n| RAG Context | retrieved document, record, vector result, or knowledge snippet | retrieval/index service |\n| Human Gate | required review/approval before publish, send, write, or close | workflow policy |\n| Golden Case | versioned evaluation example with expected behavior | eval set |\n| Anchor Case | stable calibration case used to detect model/prompt/tool drift before normal eval | eval set |\n| Refusal | safe response when request is unsupported, unsafe, or unauthorized | policy/rule set |\n\n## Aggregates and Entities\n\n| Aggregate | Owns | Notes |\n|---|---|---|\n| AIUseCase | user job, scope, risk level, success metric | product-facing intent |\n| AgentDefinition | role, goal, allowed tools, write scope, memory policy | versioned and reviewable |\n| ToolDefinition | schema, auth owner, side effects, rate limits, rollback | blocks unsafe tool mixing |\n| PromptVersion | template, variables, examples, owner, status | draft -> tested -> active |\n| ContextAsset | document/table/API/vector index, freshness, classification | permission inherited by user |\n| EvalSet | anchor cases, golden cases, adversarial cases, thresholds, reviewer | required for AI-core release |\n| RunTrace | input hash, context refs, model, tools, output, feedback | observability and audit |\n| HumanReview | reviewer, decision, reason, edited output, audit | closes accountability loop |\n\n## Domain Events\n\n```yaml\nevents:\n AgentRunStarted:\n payload: { run_id, agent_id, user_id, prompt_version, model_version }\n ContextRetrieved:\n payload: { run_id, context_refs, freshness, permission_scope }\n ToolCallRequested:\n payload: { run_id, tool_id, side_effect_scope, approval_required }\n HumanReviewRequested:\n payload: { run_id, reviewer_role, reason, risk_level }\n AIOutputPublished:\n payload: { run_id, output_id, reviewer_id, published_at }\n AIWritebackBlocked:\n payload: { run_id, target_object, reason, policy_version }\n EvalRegressionDetected:\n payload: { eval_set_id, agent_id, threshold, failed_cases }\n ChildAgentProgressObserved:\n payload: { parent_run_id, child_run_id, lifecycle_state, last_progress_at, progress_evidence_ref }\n DuplicateExecutionBlocked:\n payload: { parent_run_id, child_run_id, task_ref, reason, fallback_owner }\n```\n\n## State Machines\n\n```text\nAIUseCase: idea -> shaped -> specified -> evaluated -> launched -> monitored -> retired\nAgentDefinition: draft -> test_ready -> approved -> active -> deprecated | suspended\nPromptVersion: draft -> candidate -> evaluated -> active -> rolled_back | archived\nAgentRun: created -> retrieving_context -> reasoning -> tool_pending -> human_review -> completed\nAgentRun: reasoning | tool_pending -> refused | failed | fallback\nChildAgent: queued -> running -> waiting_on_tool | progress_visible -> completed | failed | cancelled\nHumanReview: pending -> approved | edited_and_approved | rejected | escalated\n```\n\nState consistency:\n\n| Concept | Source Of Truth | Consistency Need |\n|---|---|---|\n| write scope | AgentDefinition + policy | strong before tool execution |\n| context permission | user scope + asset classification | strong before retrieval and answer |\n| prompt/model version | RunTrace | immutable after run starts |\n| review decision | HumanReview | auditable before publish/writeback |\n| eval threshold | EvalSet | versioned before launch gate |\n\n## Metric / Indicator Governance\n\n| Metric | Caliber | Owner |\n|---|---|---|\n| task_success_rate | reviewed runs that complete user goal without material correction | product + QA |\n| human_edit_rate | approved outputs requiring user/reviewer edits | product ops |\n| refusal_correctness | unsafe/unsupported/unauthorized requests correctly refused | risk owner |\n| tool_call_error_rate | failed or blocked tool calls / total tool calls | engineering |\n| grounded_answer_rate | answers with valid context citations / answer runs | AI engineer |\n| eval_p0_pass_rate | P0 golden cases passed / P0 cases | release owner |\n| anchor_case_pass_rate | anchor cases passed before normal eval / anchor cases | AI engineer + QA |\n| rollback_time | time from regression alert to safe fallback | operations |\n\nRules:\n\n- Every launch claim needs an eval set, baseline, threshold, sample size, and\n owner.\n- Any model, prompt, retrieval, tool, or schema change must run Anchor-Cases\n before normal golden/adversarial evaluation. Anchor failure means the eval\n baseline is not comparable and must stop release as `REVIEW_COMPLETE_WITH_GAPS`.\n- AI-core modules require P0 smoke, P1 regression, and adversarial cases.\n- Production monitoring must connect run traces to user feedback and incidents.\n- Metrics must segment by model/prompt/tool/context version.\n\n## AI Context Sources\n\n| Context | Source | Freshness | Permission / Reliability Rule |\n|---|---|---|---|\n| user profile and task state | product database | real-time | inherit user/tenant/data scope |\n| knowledge documents | approved knowledge base | versioned | cite document/version; do not use expired policy |\n| structured business data | API/data warehouse/semantic layer | real-time/batch | row/column permissions before aggregation |\n| conversation memory | session or durable memory store | session/versioned | user-controlled retention and delete/export |\n| tool outputs | tool/API result | per call | include tool version and error handling |\n| eval cases | evaluation repository | versioned | do not train or leak private cases without approval |\n\n## Content / Knowledge Assets\n\n| Asset | Minimum Metadata | Governance |\n|---|---|---|\n| prompt library | owner, variables, examples, version, risk class | review before active |\n| tool registry | schema, auth owner, write scope, rollback, rate limit | security review |\n| knowledge base | owner, source, effective date, classification, freshness | lifecycle review |\n| eval set | anchor/golden/adversarial case type, expected behavior, threshold, reviewer, version | release gate |\n| refusal policy | forbidden request class, response boundary, escalation | risk/compliance owner |\n| fallback playbook | trigger, user message, manual path, owner, recovery | operations owner |\n\n## Core Workflows\n\n1. Opportunity shaping: user pain -> AI centrality -> manual fallback ->\n success metric -> risk class -> PRD profile.\n2. Agent specification: goal -> allowed context -> allowed tools -> write scope\n -> human gate -> eval set -> trace fields.\n3. Runtime answer: user request -> permission check -> context retrieval ->\n model/tool execution -> citation/refusal -> human gate when needed -> output.\n4. Tool writeback: proposed action -> policy check -> reviewer approval ->\n idempotent tool call -> audit/event -> rollback path.\n5. Evaluation loop: anchor calibration -> golden/adversarial run -> threshold\n check -> regression triage -> prompt/model/tool update -> release decision.\n6. Operations loop: trace/feedback/incident -> classify failure -> fallback or\n rollback -> fix -> post-launch review.\n7. Delegation loop: freeze task/authority/budget -> start child -> observe explicit\n liveness/progress -> wait or cancel -> accept one terminal result -> prevent\n parent duplicate execution -> merge evidence into the parent trace.\n\n## Role Path Patterns\n\n| Role | Entry | Core Actions | Forbidden Actions | Exit |\n|---|---|---|---|---|\n| end user | chat/copilot/task UI | ask, refine, accept, reject, give feedback | bypass permission or hidden data | completed task or fallback |\n| expert reviewer | review queue | inspect evidence, edit, approve/reject | approve without evidence for high-risk output | accountable decision |\n| AI engineer | eval/runtime console | tune prompt/model/tool, run evals | change active prompt without release evidence | versioned candidate |\n| product manager | PRD/eval dashboard | define outcome, scope, gate, metrics | claim quality without eval evidence | launch decision |\n| operations owner | incident console | monitor, rollback, suspend agent | silently ignore regressions | restored service |\n| admin/risk | policy/tool registry | approve tools, scopes, retention | grant broad write scopes without audit | safe active policy |\n\n## UI / Mobile Patterns\n\n- Show AI role, confidence/evidence boundary, citations, and freshness when it\n affects user trust.\n- Separate \"draft/recommendation\" from \"published/written\" states visually.\n- High-impact actions require review drawer, diff, reason, and explicit approve\n button.\n- Provide a deterministic/manual path for AI-supporting features.\n- Show source gaps, permission denials, stale context, and refusal reasons\n without exposing sensitive hidden data.\n- For mobile, keep review/approve actions short, auditable, and reversible\n where business policy allows.\n\n## Policy / Privacy Constraints\n\n- AI context retrieval inherits user, tenant, organization, row, column, and\n field-level permissions.\n- Tool calls must declare read/write scope, idempotency, side effects, timeout,\n rollback, and audit owner.\n- Model output cannot be treated as authoritative when source evidence is\n missing, stale, unauthorized, or contradicted.\n- Prompt, context, trace, and feedback retention must follow privacy and data\n deletion/export policy.\n- Autonomous writeback is forbidden for high-impact actions unless the PRD\n explicitly defines policy, human gate, eval, rollback, and on-call ownership.\n- Test agents must use shadow data or mocked tools for write paths.\n\n## Domain Test Scenarios\n\n| Scenario | Role | Expected Result |\n|---|---|---|\n| grounded answer with citations | end user | answer cites permitted context refs and freshness |\n| unauthorized data request | end user | refusal without leaking hidden fields |\n| tool write requires approval | expert reviewer | writeback waits for approval and logs decision |\n| stale knowledge base | end user | answer warns stale source or routes to manual path |\n| eval regression before launch | AI engineer | release blocked and failed cases listed |\n| anchor case drift after model upgrade | AI engineer | normal eval is paused until baseline is recalibrated |\n| fallback on model outage | operations owner | deterministic/manual fallback within SLA |\n| prompt rollback | operations owner | active prompt returns to last passing version |\n| adversarial prompt injection | risk reviewer | tool call blocked and event logged |\n| delegated tool denial | risk reviewer | a denied tool/action remains denied through agent delegation, memory and prompt content; denial is audited |\n| repeated propose-reject loop | accountable user | loop stops at configured limit and hands control to a human/manual path |\n| healthy long-running child is quiet | parent agent | explicit liveness distinguishes running from stalled; parent does not duplicate the task or token spend |\n| cross-model regression run | QA / AI engineer | evidence records provider, runtime model ID, client version, settings and date; an unverified marketing version is rejected |\n\n## Cross-Domain Requirement Patterns\n\n- `PAT-VERSION-COMPATIBILITY-001`: model, prompt, tool schema, retrieval corpus and policy versions bind every consequential run and remain replayable.\n- `PAT-LONG-RUNNING-JOB-001`: indexing, evaluation, batch inference and multi-agent jobs expose bounded progress, retry, cancellation and human takeover.\n- `PAT-METRIC-CALIBER-001`: quality, safety, cost and latency metrics declare datasets, thresholds, aggregation windows, versions and accountable interpretation.\n\n## Evaluation Profile\n\nDomain knowledge is not execution evidence. Register coverage and maturity in\n`references/domain-coverage.yaml`; keep behavioral scenarios and run evidence\noutside this knowledge file.\n\nBefore raising maturity, independently evaluate:\n\n- one primary happy path;\n- one validation or exception path;\n- one permission/privacy path;\n- one lifecycle transition;\n- one coding-agent no-guess handoff path;\n- applicable migration, integration-failure, AI, and high-risk human-gate paths.\n\nRecord executor, input, environment, timestamp, result, and evidence location.\nMocked matrices and simulated reviewers cannot satisfy expert review or audit.\n\n## Acceptance Checklist\n\n- [ ] All 15 domain module sections are present.\n- [ ] AI centrality is classified per module: AI-incidental, AI-supporting, or AI-core.\n- [ ] Agent/tool/write scopes and human gates are explicit.\n- [ ] Context sources have freshness, permission, citation, and reliability rules.\n- [ ] Eval sets, thresholds, baselines, and owners exist before AI-core launch.\n- [ ] Anchor-Cases run before golden/adversarial eval after model, prompt,\n retrieval, tool, schema, or context changes.\n- [ ] Refusal, fallback, rollback, incident, and post-launch review paths are defined.\n- [ ] High-impact decisions keep authorized human accountability.\n- [ ] Delegated work has explicit lifecycle, liveness, cancellation, duplicate-execution prevention and one terminal result owner.\n- [ ] Cross-model evidence records the runtime provider/model/client/settings instead of relying on an assumed model name.\n- [ ] Coding-agent handoff includes `ai_contract_lite` or `ai_runtime_contract`\n only at the necessary depth.\n\nFile v5.5.2:references/domains/domain-crm.md\n\n# Domain: CRM / Customer Response\n\nSource authority and freshness metadata: `references/domains/domain-sources.yaml`.\nCoverage and maturity: `references/domain-coverage.yaml`.\n\nUse this file for CRM, sales pipeline, customer service, partner service, customer success, contract/payment follow-up, and product-feedback-loop scenarios.\n\nThis file validates that the domain module contract can migrate beyond traffic safety.\n\n## Contents\n\n- Domain Purpose\n- First-Principles Domain Lens\n- Vocabulary\n- Aggregates and Entities\n- Domain Events\n- State Machines\n- Metric / Indicator Governance\n- AI Context Sources\n- Content / Knowledge Assets\n- Core Workflows\n- Role Path Patterns\n- UI / Mobile Patterns\n- Policy / Privacy Constraints\n- Domain Test Scenarios\n- Evaluation Profile\n- Multi-Module PRD Quality Gate\n- Acceptance Checklist\n\n## Domain Purpose\n\n- Business outcome: prevent missed opportunities, delayed customer responses, unresolved issues, and invisible product feedback.\n- Primary users: boss/sponsor, sales, customer service, product/R&D, finance, partner/channel manager.\n- Sensitive areas: customer contacts, contract amount, payment, customer complaints, partner relationship.\n- AI may optimize: lead classification, ticket routing, customer summary, follow-up suggestion, churn/renewal hint.\n- AI must not decide automatically: contract approval, payment confirmation, customer punishment, final sales commitment.\n- Capability scope may include marketing lead acquisition, SFA pipeline,\n Customer 360, CPQ/quotation, contract/payment, service tickets, customer\n success/renewal, partner/channel operations, and product-feedback loops.\n- Product Truth must lock module/view/region/action ownership when CRM spans\n lead, opportunity, customer, ticket, contract, payment, multiple roles, or a\n cross-module lifecycle.\n- Keep shared CRM fields canonical as `FLD-*`; projections show only relevant\n scenario, state, rule, permission, exception, and acceptance differences.\n\n### Current Baseline and Change Watch (verified 2026-08-29)\n\n| Evidence class | Current signal | Requirement effect |\n|---|---|---|\n| design_guidance | `KS-CRM-COMPLAINT-10002` | complaint intake, response, escalation, closure and learning need accountable ownership and evidence when this standard is adopted |\n| product_pattern | `KS-CRM-MICROSOFT-ACTION-2026`, `KS-CRM-SALESFORCE-CONTACT-CENTER-2026`, `KS-CRM-HUBSPOT-SMART-CRM` | current leaders are shifting from passive records to action-centric CRM with unified context, governed agents and explicit AI-to-human handoff |\n| change_watch | vendor roadmaps and AI claims change faster than customer lifecycle physics | never copy a vendor module list or claim “best practice” without project evidence; customer truth, state ownership and human commitments remain primary |\n\n### AI Transformation Horizon\n\n| Axis | From -> toward | Non-negotiable product guard |\n|---|---|---|\n| CRM role | passive system of record -> customer/revenue/service system of action | every AI action resolves to a customer object, lifecycle state, owner, deadline and observable result |\n| context | scattered notes -> permissioned longitudinal customer memory | distinguish recorded facts, inferred intent, stale data and external signals; preserve source links |\n| execution | next-best-action text -> bounded research, follow-up drafting, routing, renewal/service playbooks and exception escalation | discount, contract, payment, customer commitment and final issue closure retain accountable human gates |\n| learning loop | activity count -> conversion, response, retention, service resolution and override evidence | do not optimize spammy outreach or hide agent failure behind generated activity |\n\n## First-Principles Domain Lens\n\nFirst-Principles CRM Product Logic: CRM product judgment starts from customer\noperating response, not from a list of tables or fashionable AI features.\n\n| Lens | CRM Question | Acceptance Signal |\n|---|---|---|\n| Revenue lifecycle | Which customer or opportunity state moves toward signed revenue, renewal, or explicit loss? | lead/opportunity/customer/contract state changes and owner are traceable |\n| Response accountability | Who must respond, by when, and what closes the response? | SLA task, owner, close guard, and audit are explicit |\n| Customer truth | Which module is the source of truth for customer, contact, contract, ticket, and follow-up facts? | Customer 360 can explain every aggregate with source links |\n| Exception closure | Which stuck, overdue, rejected, lost, or escalated case must be visible to a manager? | alert/response task links to the business object and real closure action |\n| Boundary of automation | Which sales, finance, service, or roadmap actions cannot be automatic? | payment, contract approval, final commitment, and ticket closure keep human gates |\n| Domain breadth | Which CRM family is in scope: SFA, service, customer success, CPQ, campaign, channel, data/BI, AI assistant? | included/deferred modules and external systems are explicit |\n\n## Vocabulary\n\n| Term | Meaning | Source of Truth |\n|---|---|---|\n| Lead | potential customer/opportunity signal | CRM |\n| Opportunity | qualified sales chance with stage/value | CRM |\n| Customer 360 | customer profile, contacts, interactions, contracts, tickets | CRM |\n| Ticket | customer issue or request requiring handling | service module |\n| ResponseTask | SLA-driven reminder/escalation task | workflow module |\n| Partner | channel/agent who introduces or serves customers | partner module |\n| Campaign | marketing activity, channel, audience, and conversion source | marketing module |\n| Quote / CPQ | product/package/discount proposal before contract | quotation module |\n| Renewal / Churn Risk | customer success lifecycle signal | customer success module |\n| Demand | product request created from service or sales feedback | product/R&D module |\n\n## Aggregates and Entities\n\n| Aggregate | Owns | Key States | Notes |\n|---|---|---|---|\n| Lead | source, owner, first response, qualification | new / assigned / contacted / converted / invalid |\n| Opportunity | stage, value, owner, next action | qualified / proposal / poc / negotiation / won / lost |\n| Customer | profile, contacts, tags, owner, lifecycle | prospect / active / at_risk / churned |\n| Contract | amount, payment plan, invoices | draft / signed / partially_paid / paid / overdue |\n| Ticket | type, owner, SLA, result | new / assigned / processing / pending_customer / escalated / closed |\n| ResponseTask | object ref, SLA, owner, priority | pending / responded / overdue / escalated / closed |\n| Campaign | audience, channel, cost, leads, conversion | draft / active / completed / archived |\n| Quote | product package, price, discount, approval | draft / submitted / approved / rejected / expired |\n| RenewalPlan | account, renewal date, risk, owner, actions | planned / at_risk / negotiating / renewed / churned |\n| Demand | source ticket/customer, value, review, roadmap link | new / reviewed / scheduled / released / rejected |\n\n## Domain Events\n\n```yaml\nevents:\n LeadAssigned:\n payload: { lead_id, owner_id, assigned_by, assigned_at }\n LeadFirstResponded:\n payload: { lead_id, response_time, followup_id }\n OpportunityWon:\n payload: { opportunity_id, customer_id, contract_id, amount }\n TicketCreated:\n payload: { ticket_id, customer_id, type, creator_id }\n TicketEscalated:\n payload: { ticket_id, reason, owner_role, response_task_id }\n PaymentRegistered:\n payload: { contract_id, payment_id, amount, operator_id }\n RenewalRiskDetected:\n payload: { customer_id, risk_level, evidence_refs, owner_id }\n DemandCreatedFromTicket:\n payload: { demand_id, ticket_id, customer_id, creator_id }\n QuoteApproved:\n payload: { quote_id, opportunity_id, approver_id, approved_at }\n```\n\n## State Machines\n\n```text\nLead: new -> assigned -> contacted -> converted | invalid\nOpportunity: qualified -> proposal -> poc -> negotiation -> won | lost\nTicket: new -> assigned -> processing -> pending_customer -> closed\nTicket: processing | pending_customer -> escalated -> processing\nResponseTask: pending -> responded -> closed\nResponseTask: pending -> overdue -> escalated -> closed\nQuote: draft -> submitted -> approved | rejected | expired\nRenewalPlan: planned -> at_risk -> negotiating -> renewed | churned\nDemand: new -> reviewed -> scheduled -> released | rejected\n```\n\n## Metric / Indicator Governance\n\n| Metric | Caliber | Source | Owner |\n|---|---|---|---|\n| first_response_timeout_count | high-intent leads not responded within SLA | Lead + ResponseTask | sales ops |\n| overdue_ticket_count | tickets beyond SLA | Ticket | service manager |\n| stalled_opportunity_count | opportunities with no update beyond threshold | Opportunity | sales manager |\n| partner_feedback_delay | partner leads without feedback | Partner + Lead | channel manager |\n| campaign_conversion_rate | qualified leads / campaign leads by channel and period | Campaign + Lead | marketing ops |\n| win_rate_by_stage | won opportunities / stage-entered opportunities | Opportunity | sales ops |\n| payment_overdue_amount | unpaid amount past payment-plan date | Contract + Payment | finance |\n| renewal_risk_count | active customers with renewal risk above threshold | Customer + RenewalPlan | customer success |\n| ticket_to_demand_rate | product-related tickets converted to demands | Ticket + Demand | product ops |\n\n## AI Context Sources\n\n| Context | Source | Freshness | Permission Scope |\n|---|---|---|---|\n| lead source and text | CRM lead records | real-time | assigned org/role |\n| customer interactions | followup/ticket records | real-time | customer owner + authorized roles |\n| contract/payment | finance module | real-time | masked by role |\n| product issue history | ticket/demand module | daily/real-time | service/product roles |\n| marketing and channel source | campaign/partner platform | daily/real-time | marketing/channel roles |\n| quote/discount history | quotation/contract module | real-time | sales manager/finance/legal scope |\n| renewal and usage signals | customer success/product telemetry | daily | customer owner + CS role |\n\n## Content / Knowledge Assets\n\n| Asset | Minimum Metadata | Governance |\n|---|---|---|\n| sales playbook | stage, target customer, owner, version | approved by sales leadership |\n| response/SLA policy | object type, priority, deadline, escalation | versioned and auditable |\n| product FAQ/solution | product/version, audience, effective date | product review before publishing |\n| contract/payment policy | contract type, approval role, financial boundary | finance/legal authority |\n| service templates | ticket type, response text, attachment requirements | service owner and revision history |\n| sales stage playbook | stage, exit criteria, required evidence, next action | sales leadership review |\n| quote/discount policy | product package, discount boundary, approval role | finance/legal authority |\n| renewal/churn playbook | risk signal, intervention action, owner, cadence | customer success owner |\n\n## Core Workflows\n\n| Workflow | Trigger | End State |\n|---|---|---|\n| lead response | lead imported/created | contacted or invalid |\n| opportunity conversion | lead qualified | opportunity won/lost |\n| customer issue handling | customer feedback | ticket closed or demand created |\n| product feedback loop | repeated tickets | demand evaluated/scheduled |\n| partner service | partner introduces lead | feedback sent to partner |\n| marketing-to-sales handoff | campaign lead qualified | lead assigned and SLA tracked |\n| quote-to-contract | opportunity price proposal approved | contract draft or signed |\n| renewal/churn intervention | renewal risk signal appears | renewal plan closed as renewed/churned |\n\n## Role Path Patterns\n\n| Role | Entry | Core Actions | Forbidden Actions | Exit |\n|---|---|---|---|---|\n| boss | dashboard | inspect exception, assign owner, view closure | edit raw financial records | exception reduced |\n| sales | my leads/customers | respond, follow up, convert, create ticket | view other sales sensitive data | next action recorded |\n| customer service | service queue | create ticket, assign, follow, close | confirm payment | ticket closure |\n| product/R&D | ticket/demand queue | accept, resolve, adopt demand | change contract/payment | demand roadmap |\n| finance | contract/payment | register payment, view overdue | edit customer issue | payment updated |\n| marketing ops | campaign dashboard | import leads, inspect conversion | change opportunity stage | leads handed off |\n| customer success | customer success workspace | inspect health, plan renewal, create task | approve discount/payment | renewal outcome |\n| channel manager | partner workspace | manage partner leads, feedback, settlement evidence | see unrelated partner data | partner SLA closed |\n\n## UI / Mobile Patterns\n\n- dashboard: exception metrics with drilldown;\n- response center: SLA task queue;\n- customer 360: contacts, followups, opportunities, contracts, tickets;\n- customer success workspace: health, usage, renewal, churn risk, intervention tasks;\n- campaign and channel views: lead source, conversion funnel, attribution, partner feedback;\n- quote/contract/payment flow: quote approval, contract state, payment plan, overdue follow-up;\n- mobile sales: customer lookup, followup, ticket submission, weak-network draft;\n- boss path: exception -> object detail -> owner action -> status verification.\n- Product Truth view examples:\n - `M01-V01`: boss operating cockpit / exception dashboard;\n - `M02-V01`: lead pool and lead response queue;\n - `M03-V01`: opportunity pipeline;\n - `M04-V01`: customer 360;\n - `M05-V01`: ticket and demand loop;\n - `M06-V01`: contract/payment follow-up;\n - `M02-V01-mobile`: sales mobile lead/customer follow-up when the mobile path\n has distinct navigation, permission, or offline behavior.\n- Coding-agent handoff uses `delivery/truth/product-truth.yaml`, approved\n changes, projections, prototype, acceptance, agents, evidence, and manifest.\n\n## Policy / Privacy Constraints\n\n- customer contacts and contract amounts are masked by role;\n- export requires audit;\n- cross-tenant/customer access requires explicit authorization;\n- AI suggestions must show source and confidence;\n- AI cannot confirm payment, approve contract, or close customer issue without human confirmation.\n- AI cannot approve quote discounts, decide churn status, modify payment amount,\n or publish product roadmap commitments without accountable human approval.\n\n## Domain Test Scenarios\n\n| Scenario | Role | Preconditions | Steps | Expected Domain Result |\n|---|---|---|---|---|\n| high-intent lead not responded | boss/sales | lead overdue | assign -> first respond | LeadFirstResponded, ResponseTask closed |\n| customer rejects ticket result | customer service | ticket pending_customer | escalate | TicketEscalated, urgent ResponseTask created |\n| sales mobile weak network followup | sales | customer visible | write followup offline | draft preserved, no data loss |\n| partner lead feedback | channel manager | partner lead converted | send feedback | PartnerFeedbackRecorded |\n| quote approval boundary | sales manager/finance | quote exceeds discount threshold | submit -> review -> approve/reject | QuoteApproved or rejection audit |\n| renewal risk intervention | customer success | customer is at_risk | create plan -> follow up -> close | RenewalPlan ends renewed/churned |\n| ticket to demand loop | service/product | repeated product issue | convert ticket to demand | DemandCreatedFromTicket traceable |\n| lead to collected revenue | sales/finance | qualified lead exists | convert -> win opportunity -> sign contract -> invoice -> register payment | lead, opportunity, customer, contract, invoice, and payment retain source IDs, state owners, permissions, and audit |\n\n## Cross-Domain Requirement Patterns\n\n- `PAT-CROSS-MODULE-CONVERSION-001`: ticket→demand, lead→opportunity/customer and opportunity→contract preserve stable source/target references and a reachable next owner action.\n- `PAT-PERMISSION-001`: navigation visibility, object permission, row scope and field masking are separate contracts; a visible menu never expands data access.\n- `PAT-VERSION-COMPATIBILITY-001`: sales stages, SLA, quote policy and approval configuration preserve in-flight and historical interpretation.\n- `PAT-METRIC-CALIBER-001`: funnel, win rate, response SLA, revenue and collection metrics share one state, time and dedup caliber across cockpit/export/API.\n\n## Evaluation Profile\n\nDomain knowledge is not execution evidence. Register coverage and maturity in\n`references/domain-coverage.yaml`; keep behavioral scenarios and run evidence\noutside this knowledge file.\n\nBefore raising maturity, independently evaluate:\n\n- one primary happy path;\n- one validation or exception path;\n- one permission/privacy path;\n- one lifecycle transition;\n- one coding-agent no-guess handoff path;\n- applicable migration, integration-failure, AI, and high-risk human-gate paths.\n\nRecord executor, input, environment, timestamp, result, and evidence location.\nMocked matrices and simulated reviewers cannot satisfy expert review or audit.\n\n## Acceptance Checklist\n\n- [ ] Lead, opportunity, customer, ticket, contract/payment, response task states are defined.\n- [ ] Marketing/campaign, quote/CPQ, renewal/customer success, partner/channel,\n and product-demand loops are included or explicitly out of scope.\n- [ ] Every SLA exception has owner, action, and closure rule.\n- [ ] Customer 360 shows business data, not only metadata.\n- [ ] Role/data isolation is explicit.\n- [ ] AI suggestions remain suggestions unless explicitly human-approved.\n- [ ] If CRM spans multiple modules or roles, Product Truth locks shared\n objects, state owners, views, actions, events, permissions, and flows.\n- [ ] Common CRM fields use canonical `FLD-*` definitions; projections do not\n duplicate or redefine them.\n- [ ] Coding-agent delivery package paths are explicit when implementation by\n Claude Code, Cursor, Codex, or Copilot is expected.\n\nFile v5.5.2:references/domains/domain-data-mart.md\n\n# AI + Data Product System Domain Module\n\nSource authority and freshness metadata: `references/domains/domain-sources.yaml`.\nCoverage and maturity: `references/domain-coverage.yaml`.\n\nUse this replaceable domain module for data-resource inventory and registration, data products and services, public-data authorization operations, trusted data spaces, AI-ready and high-quality datasets, data labeling, training/evaluation/preference data, AI+Data platforms, warehouses/lakehouses, governance, catalog, lineage, semantic/ontology products, BI, ChatBI, Data Agents, reporting, retrieval, and operational data applications. A replacement domain module must preserve the same section headings used here and in the maintainer domain template.\n\nThis module is intentionally broader than a data mart checklist. It treats data products as an end-to-end system: source acquisition -> processing -> governance -> storage/retrieval -> semantic/ontology layer -> analytics/BI -> AI agents -> action/decision loop -> operations/evaluation.\n\n## Contents\n\n- Domain Purpose\n- First-Principles Domain Lens\n- Vocabulary\n- Aggregates and Entities\n- Domain Events\n- State Machines\n- Metric / Indicator Governance\n- AI Context Sources\n- Content / Knowledge Assets\n- Core Workflows\n- Role Path Patterns\n- UI / Mobile Patterns\n- Policy / Privacy Constraints\n- Domain Test Scenarios\n- Evaluation Profile\n- Acceptance Checklist\n\n## Domain Purpose\n\nAI+Data PRDs must specify how raw, distributed, and often untrusted data becomes governed, explainable, searchable, analyzable, and actionable. The product contract is not just a dataset, report, or agent conversation. It is the full operating chain:\n\n```text\nscenario demand -> source/right/use proof -> acquire/ingest -> clean/label/standardize\n-> govern/catalog/lineage -> register/certify -> productize and price\n-> authorize/digital contract/trusted use -> settle/evaluate\n-> BI/API/AI training or action -> application/model feedback -> iterate or retire\n```\n\nUse this module when any of these capabilities are in scope:\n\n- multi-source connectors, file upload, API sync, CDC, streaming, batch jobs, or manual fill-in;\n- data cleaning, transformation, entity resolution, standardization, deduplication, quality checks, or exception handling;\n- metadata, catalog, lineage, RBAC/ABAC, row/column permissions, privacy classification, approval, audit, or data asset lifecycle;\n- data lake, warehouse, lakehouse, ODS/DWD/DWS/ADS, OLAP cube, search index, vector index, cache, or retrieval service;\n- semantic model, metric layer, dimension dictionary, business glossary, ontology object/link/action types, or data product API;\n- dashboard, self-service analysis, report template, filling task, export, embedded analytics, or operational cockpit;\n- Data Agent, ChatBI, NL2SQL, NL2Metrics, insight generation, anomaly attribution, report writing, or action agent;\n- public-data resource registration, authorized operation, product/service disclosure, pricing, settlement, supervision, or exit;\n- data-property registration, source/right evidence, transfer/change/renewal/cancellation, objection, credential, or cross-platform recognition;\n- dataset collection, cleaning, deduplication, labeling, de-identification, train/validation/test split, SFT, preference/DPO/RLHF, model evaluation, contamination control, or application-feedback iteration.\n\n### Evidence-Bounded 2026-2028 Planning Horizon\n\nTreat these as planning directions supported by current policy and standards, not\nas mandatory features for every project.\n\n| Direction | Likely Product Shift | Requirement Guard |\n|---|---|---|\n| nationwide registration and interoperability | resource/property/product credentials become reusable trust evidence | registration type, applicant, source proof, review, objection, validity, change and cancellation remain distinct |\n| trusted circulation infrastructure | one-off file exchange shifts toward governed use inside data spaces | participant identity, purpose, digital contract, usage control, audit, revocation and exit are explicit |\n| public-data market operations | catalog publication shifts toward scenario-based authorized products/services | implementing/operating roles, authorization scope, cost/revenue, price, disclosure, supervision and no-raw-data-market rule are explicit |\n| high-quality industry datasets | data governance extends from BI correctness to AI fitness | intended model/scenario, data rights, coverage, labeling, split, quality, bias, contamination and benchmark are versioned |\n| model-data-scene flywheel | dataset delivery becomes continuous supply and evaluation | model/evaluation/app feedback creates traceable dataset changes without polluting held-out tests |\n| asset and value management | registration, accounting, valuation and financing become connected but separate workflows | no credential or appraisal automatically becomes accounting recognition, ownership judgment, or guaranteed value |\n\n### Current Baseline and Change Watch (verified 2026-08-29)\n\n| Evidence class | Current signal | Requirement effect |\n|---|---|---|\n| binding_baseline | `KS-DATA-CIRCULATION-SERVICES-2026` extends the current data-resource, trusted-space, authorized-operation and high-quality-dataset policy chain | identify provider/operator/user roles, data/right/purpose proof, product/service disclosure, safety, evaluation and exit before designing circulation or agent services |\n| product_pattern | `KS-DATASET-PLATFORM-2026`, `KS-DATA-SECURITY-CASES-2026`, `KS-DATA-COMPLIANCE-SANDBOX-2026` | learn dataset cards, participant identity, digital contracts, usage control, audit, bounded experiments and exit; cases do not prove a universal architecture |\n| change_watch | sector rules, regional exchange mechanics, price/settlement, accounting and cross-border scope remain project-specific | register jurisdiction and accountable-source gaps; never let a dataset card, sandbox or registration credential silently prove rights, quality, value or lawful use |\n\n### AI Transformation Horizon\n\n| Axis | From -> toward | Non-negotiable product guard |\n|---|---|---|\n| platform role | warehouse/BI backend -> governed context, semantic and action substrate for people and agents | source/right/purpose, lineage, quality, freshness and permission travel with every consumption path |\n| interface | dashboard/query builder -> conversational analysis plus inspectable semantic/query plan | resolve metrics/objects first; cite source/caliber and keep deterministic query/export paths |\n| operation | manual catalog/quality work -> bounded data engineering, governance and quality agents | production writes, grants, schema/metric changes and destructive repair keep review, audit and rollback |\n| supply loop | one-off dataset delivery -> model/application feedback driving versioned data products | protect evaluation isolation; no automatic ingestion of conversations, feedback or synthetic derivatives |\n| value | data volume/asset labels -> measurable decision, model, public-service or commercial outcomes | registration, valuation, accounting, rights and fitness remain separate accountable claims |\n\n## First-Principles Domain Lens\n\nAI+Data product judgment starts from trusted decision flow, not from a report,\nwarehouse, or chatbot label.\n\n| Lens | Data Product Question | Acceptance Signal |\n|---|---|---|\n| Data supply chain | Which source becomes trusted data, at what freshness and quality? | source, sync, quality, lineage, and owner are explicit |\n| Semantic truth | Which metrics, dimensions, entities, and ontology actions users rely on? | semantic/ontology version and permission scope are governed |\n| Decision loop | What analysis, fill-in, alert, or action changes after insight? | BI/ChatBI/Data Agent output links to workflow or accountable decision |\n| Reversibility | Which data or ontology writes can be corrected safely? | human gate, audit, rollback, and recalculation scope are defined |\n| Evidence | Can every answer, metric, dataset and generated report cite source/caliber? | citations, query trace, dataset card and quality status are visible |\n| Rights and purpose | Who may process which data for which purpose and period? | source/right evidence, lawful basis, authorization scope, use controls and expiry are linked |\n| Market operation | What product/service is delivered without releasing uncontrolled raw data? | product contract, service level, price/settlement, disclosure, supervision and exit are explicit |\n| AI fitness | Does the dataset improve the intended model and scenario rather than merely look clean? | task coverage, label agreement, split isolation, benchmark result and model feedback are versioned |\n| Value proof | What measurable public, operational or commercial outcome justifies continued supply? | beneficiary, baseline, outcome metric, cost/revenue and attribution limits are declared |\n\n## Vocabulary\n\n| Term | Meaning | PRD Requirement |\n|---|---|---|\n| 数据源 / Data source | database, file, API, stream, SaaS app, manual fill-in, external data | ownership, credential, schema, cadence, SLA, backfill, permission |\n| 接入 / Ingestion | batch, CDC, stream, file import, API pull/push, manual upload | connector, schedule, idempotency, retry, failure quarantine |\n| 清洗 / Cleaning | normalize, type cast, dedupe, map dictionary, repair, standardize | rule, exception queue, owner, audit, sample evidence |\n| 治理 / Governance | catalog, metadata, lineage, quality, standard, security, lifecycle | owner, policy, approval, lineage, quality score, audit |\n| 存储 / Storage | lake, warehouse, lakehouse, mart, OLAP, cache | model layer, partition, retention, cost, refresh |\n| 检索 / Retrieval | SQL, OLAP, search, vector search, API, materialized view | latency, freshness, permission, ranking, citations |\n| 语义层 / Semantic layer | business mapping over tables, fields, metrics, dimensions, synonyms | model version, field descriptions, relations, approved query scope |\n| 本体 / Ontology | object types, properties, links, actions, rules, operational semantics | object/link/action contract, write guard, policy, audit |\n| 指标 / Metric | stable measurable business concept | layer, formula, unit, period, source, owner, quality rule |\n| 维度 / Dimension | grouping, filter, drill, hierarchy, permission scope | dictionary, value source, hierarchy, scope rule |\n| ChatBI | conversational analytics over governed semantic data | semantic context, permission, answer citation, query trace, fallback |\n| Data Agent | agent that reads, reasons, analyzes, or acts over governed data | tools, allowed sources, write scope, human gate, evaluation |\n| 数据资源登记 / Resource registration | public-data resource/product/service inventory and disclosure record | registration object, provider, catalog, update, review, credential and change |\n| 数据产权登记 / Property registration | standardized evidence for data-resource or data-product rights and changes | applicant, source/right proof, initial/transfer/change/renewal/cancellation, objection and validity |\n| 授权运营 / Authorized operation | approved operation of scoped public-data resources through an implementing and operating arrangement | plan, decision, agreement, resource/product scope, term, disclosure, supervision and exit |\n| 数据产品与服务 / Data product or service | controlled data result, API, model-ready dataset, verification or computing service delivered for a purpose | inputs, transformations, quality, consumer, usage rule, SLA, price and liability |\n| 数字合约 / Digital contract | machine-enforceable purpose, permission, obligation and usage-control agreement | parties, data/product, purpose, actions, period, count, location, revocation and evidence |\n| 高质量数据集 / High-quality dataset | scenario-oriented, governed dataset that can measurably support model development or evaluation | intended use, provenance, rights, coverage, labels, splits, quality, benchmark and version |\n| 训练/验证/测试数据 | separated data partitions for fitting, tuning and unbiased evaluation | split rule, isolation, leakage/contamination check and release lineage |\n| 偏好数据 / Preference data | ranked/comparison/critique data used for preference optimization or reward learning | task/source, annotator policy, agreement, safety taxonomy, model/version and bias review |\n| 数据卡 / Dataset card | human- and machine-readable evidence for one dataset version | owner, license/rights, collection, processing, population, gaps, risks, metrics and changes |\n| 数据资源会计 / Data-resource accounting | accounting treatment based on applicable standards and economic substance | purpose, cost basis, recognition decision, useful life, impairment and disclosure; separate from registration/valuation |\n| sys column | system-extracted report/fill field | source, refresh, lock, quality rule |\n| ext column | human-filled report/fill field | control, validation, submit/review state |\n\n## Aggregates and Entities\n\n| Layer | Aggregates / Entities | Key Invariants |\n|---|---|---|\n| Source and ingestion | DataSource, Connector, SyncJob, StreamTopic, FileImport, Credential, BackfillTask | every ingest path has owner, auth, schema, cadence, idempotency, retry, and failure quarantine |\n| Processing and quality | Pipeline, TransformStep, CleanRule, QualityRule, ExceptionRecord, MasterDataMap | raw data is never silently overwritten; fixes are versioned and auditable |\n| Governance and catalog | DataAsset, Metadata, BusinessTerm, LineageEdge, Policy, Approval, AuditLog | each trusted asset has owner, classification, lineage, access policy, and lifecycle state |\n| Rights and registration | SourceEvidence, RightEvidence, RegistrationApplication, RegistrationReview, Objection, Credential, RegistrationChange | registration preserves applicant, object, evidence, reviewer, validity and every change without implying accounting recognition |\n| Market and trusted circulation | AuthorizationPlan, OperatingAgreement, DataProduct, DataService, Participant, DigitalContract, UsagePolicy, Settlement, Disclosure, ExitRecord | every use is bounded by participant, product, purpose, action, term, usage evidence, revocation and liability |\n| AI data supply | Dataset, DatasetVersion, Sample, LabelSchema, AnnotationTask, Annotator, Split, DatasetCard, Benchmark, ModelFeedback | every sample/version preserves provenance and right basis; held-out evaluation is isolated from training and feedback |\n| Storage and retrieval | LakeTable, WarehouseTable, MartTable, OLAPCube, SearchIndex, VectorIndex, Cache | consumers know freshness, retention, query limits, and permission scope |\n| Semantic and ontology | SemanticModel, Dataset, Metric, Dimension, ObjectType, LinkType, ActionType, Rule | business questions resolve through approved semantics before raw SQL or tool calls |\n| Analytics and reporting | Dashboard, Chart, AnalysisSession, ReportTemplate, ReportTask, FillSubmission, ExportJob | visible numbers use declared metric, dimension, period, scope, and caliber version |\n| AI agent runtime | Agent, Tool, Prompt, EvalSet, Trace, Conversation, Insight, Recommendation, WritebackAction | AI reads and acts only within governed tools, permissions, confidence, and human gates |\n| Operations | SLA, FreshnessCheck, CostBudget, Incident, RollbackPlan, UsageMetric | production trust is monitored with freshness, quality, latency, cost, adoption, and failure metrics |\n\n## Domain Events\n\n| Event | Trigger | Consumers |\n|---|---|---|\n| DataSourceConnected | connector or credential is approved | catalog, pipeline, permission review |\n| SchemaChanged | source table/API/file schema changes | lineage, semantic model, tests, affected dashboards/agents |\n| IngestionFailed | sync job fails or exceeds retry policy | ops, data owner, downstream freshness warnings |\n| QualityRuleFailed | quality check violates threshold | exception queue, dashboard warning, release gate |\n| DataAssetPublished | curated data asset becomes trusted | semantic model, BI, ChatBI, APIs |\n| LineageUpdated | transform/source dependency changes | governance, impact analysis, audit |\n| MetricCaliberChanged | metric business or technical definition changes | templates, dashboards, exports, agents, history markers |\n| OntologyActionSubmitted | user/agent writes object, link, or workflow state | audit, policy guard, downstream operational app |\n| SemanticModelPublished | dataset/metric/dimension/relationship model changes | ChatBI, self-service analysis, report builder |\n| ChatBIQuestionAnswered | natural-language answer is generated | trace, eval, feedback, usage analytics |\n| AgentInsightGenerated | agent produces insight/recommendation | human review, dashboard, task/workflow |\n| AgentWritebackBlocked | agent tries disallowed write or low-confidence action | safety audit, fallback, human review |\n| ReportAggregationCompleted | report/fill data is merged and calculated | dashboard, export, quality checks |\n| DataProductRetired | asset/model/report/agent is deprecated | consumers, migration plan, retention/deletion |\n| RegistrationCredentialIssued | a registration application passes review and objection handling | catalog, credential registry, applicant, authorized consumers |\n| RegistrationChangedOrCancelled | transfer/change/renewal/cancellation becomes effective | credentials, contracts, consumers, accounting/legal review |\n| AuthorizationAgreementEffective | approved resource/product/service scope becomes usable | operating platform, usage control, disclosure, supervision |\n| UsagePolicyViolated | a participant exceeds purpose, scope, action, quantity, location or term | block/revoke, incident, audit, notification, liability |\n| DatasetVersionReleased | a governed training/evaluation/preference dataset passes release gates | model teams, registry, benchmark, lineage and consumers |\n| DatasetContaminationDetected | held-out or prohibited data is found in training/feedback inputs | release block, affected models, re-split/retrain decision and audit |\n| ModelFeedbackAccepted | evaluated application feedback is approved as a dataset change | sampling backlog, label review, new dataset version; never mutate the old version |\n\n## State Machines\n\nData source:\n\n```text\ndraft -> connected -> syncing -> active -> degraded -> suspended -> retired\nsyncing -> failed -> syncing\n```\n\nPipeline:\n\n```text\ndraft -> test_run -> scheduled -> running -> succeeded\nrunning -> failed -> retrying -> running\nscheduled -> paused -> scheduled\n```\n\nData asset:\n\n```text\nraw -> profiled -> curated -> certified -> deprecated -> retired\ncurated -> blocked_by_quality -> curated\n```\n\nSemantic model / ontology:\n\n```text\ndraft -> reviewed -> published -> versioned -> deprecated\npublished -> rollback_pending -> published\n```\n\nReport/fill task:\n\n```text\ndraft -> created -> collecting -> aggregating -> completed -> archived\ncollecting -> cancelled\naggregating -> failed -> aggregating\n```\n\nData agent:\n\n draft -> evaluated -> pilot -> production -> degraded -> disabled -> retired\n pilot/production -> rollback_pending -> previous_version\n\nData registration:\n\n draft -> submitted -> accepted_for_review -> reviewing -> public_notice -> issued -> changed/renewed/transferred -> cancelled/expired\n submitted/reviewing/public_notice -> returned/rejected\n public_notice -> objection_pending -> reviewing\n\nAuthorization and trusted use:\n\n planned -> approved -> contracted -> active -> suspended -> revoked/expired -> exited\n active -> violation_pending -> active/revoked\n\nAI dataset:\n\n planned -> collecting -> processing -> labeling -> quality_review -> released -> in_use -> superseded -> retired\n quality_review -> rejected -> processing/labeling\n released/in_use -> contamination_hold -> corrected_new_version\n\nEvery state transition must define role/tool permission, data/right/purpose guard, quality/freshness guard, audit event, notification, retry/idempotency, affected consumers, and rollback or correction behavior.\n\n## Metric / Indicator Governance\n\nAI+Data PRDs must separate data correctness from business meaning. For every trusted metric, specify all layers below.\n\n### Indicator Layering\n\n| Layer | Definition | Required Fields |\n|---|---|---|\n| 原子指标 | base measurable fact from a trusted source | source table/API, field, aggregation, unit, owner |\n| 派生指标 | atomic metric plus dimension/filter/period | base metric, dimension, filter, statistical period |\n| 复合指标 | expression over multiple metrics | dependency graph, formula, recalculation rule |\n| 目标/阈值 | planned target, warning line, assessment rule | owner, period, exception rule, effective time |\n\n### Caliber Dual Track\n\n| Field | Required Content |\n|---|---|\n| 指标编码 | stable ID such as `IND-001` |\n| 业务定义 | plain-language meaning, inclusion/exclusion, boundary time |\n| 技术计算逻辑 | SQL, MQL, DAX, formula, semantic expression, or rule |\n| 统计周期 | day/week/month/quarter/custom/cumulative and time-zone/boundary |\n| 数据来源 | source system, ODS/DWD/DWS/ADS/API/table/field |\n| 维度范围 | dimensions, hierarchies, allowed drill paths |\n| 权限范围 | tenant/org/region/enterprise/row/column/action |\n| 质量规则 | null/range/duplicate/freshness/referential/anomaly |\n| 负责人 | business owner, data owner, technical owner |\n| 版本 | effective time, superseded version, migration note |\n\n### Dimension Dictionary\n\n| Dimension ID | Name | Value Source | Drill Path | Related Metrics | Scope Rule |\n|---|---|---|---|---|---|\n| DIM-AREA | 区域 | region master data | province -> city -> district | enterprise_count, risk_count | user.home_region subtree |\n\n### Ontology And Semantic Contract\n\n| Contract | Required Content |\n|---|---|\n| Object type | business entity/event, properties, owner, source mapping |\n| Link type | object relationship, cardinality, join/source field, lifecycle |\n| Action type | user/agent operation, parameters, guard, side effect, audit |\n| Query type | approved read/query pattern, filters, row/column scope |\n| Rule | quality, policy, state, permission, or writeback condition |\n| Synonym/glossary | business terms, aliases, forbidden ambiguous terms |\n\n### Caliber Change Propagation\n\n| Change | Required Impact Handling |\n|---|---|\n| source schema changes | list affected pipelines, metrics, semantic models, dashboards, agents, exports |\n| metric caliber changes | preview affected templates, reports, dashboards, AI answers, and historical data |\n| dimension values change | refresh filters, drill paths, permission scopes, and cached queries |\n| ontology action changes | verify downstream state transitions, audit, side effects, and agent write scope |\n| history data recalculates | keep old/new version markers and user-facing explanation |\n\n## AI Context Sources\n\nData agents and ChatBI must be grounded in governed context rather than raw table guessing.\n\n| AI Scenario | Required Context | Guard |\n|---|---|---|\n| ChatBI / NL2SQL | semantic model, metric/dimension dictionary, synonyms, query examples, permission scope | generate constrained query plan and cite metric/source; no arbitrary cross-source join unless approved |\n| NL2Metrics / NL2Semantics | approved metrics, dimensions, object types, business terms | resolve question to metric/dimension/object first, then query |\n| Data engineering assistant | source schema, lineage, pipeline templates, quality rules | propose pipeline/SQL; human approves before production write |\n| Data governance agent | catalog, classification, policy, lineage, access logs | recommend policy/owner; no silent permission grants |\n| Data quality agent | profiles, anomaly baselines, exception history | raise issue/recommend fix; preserve raw data and audit |\n| Insight / attribution agent | time series, dimension drill path, benchmark, quality/freshness state | label correlation vs cause; require evidence for causal claims |\n| Report-writing agent | certified metrics, report structure, source citations, audience | generated narrative cites metric IDs and periods |\n| Action / decision agent | ontology actions, workflow state, write scope, human gate | draft or route tasks unless explicit production write permission exists |\n\n### AI Dataset Supply Contract\n\nDo not collapse collection, annotation, training, evaluation and feedback into one\ndataset status. For every released dataset version, specify:\n\n| Contract Area | Required Content |\n|---|---|\n| purpose | target industry, task, model family, application scenario, excluded uses and accountable owner |\n| provenance and rights | source, collection method/time, lawful/right basis, license/contract, personal/sensitive data treatment and downstream restrictions |\n| population and coverage | unit of observation, time/region/device/language/domain coverage, rare/edge cases, known gaps and representativeness limits |\n| processing | extraction, cleaning, deduplication, de-identification, synthesis/augmentation, filtering and human/automated steps |\n| labeling | schema/version, instructions, annotator qualifications, agreement/adjudication, acceptance sampling and rejected work |\n| splits | train/validation/test/preference/benchmark purpose, isolation key, leakage and contamination checks |\n| quality and fitness | intrinsic quality plus model/scenario benchmark, threshold, failure slices, bias/safety tests and reviewer |\n| release and lineage | immutable version, parent/derived datasets, sample hash, code/rule version, model consumers and retirement |\n| feedback | application/evaluation signal, sampling criteria, human approval, label correction and target next version |\n| value | public/operational/commercial outcome, baseline, cost, attributable benefit and renewal/exit decision |\n\nSFT, preference optimization, DPO, reward learning and reinforcement-learning\ndatasets require separate schemas and quality evidence. Production conversations\nor user feedback never enter training automatically; collection purpose,\nauthorization, filtering, approval and versioned lineage must be explicit.\n\nMinimum AI data runtime contract:\n\n```yaml\ndata_agent_contract:\n agent_id:\n use_case: chatbi | insight | data_quality | data_engineering | governance | report_writer | action_agent\n allowed_sources: [semantic_model, catalog, lineage, certified_dataset, ontology]\n forbidden_sources: [raw_sensitive_table, unapproved_export]\n tool_scope: [sql_query, semantic_query, vector_search, catalog_lookup, lineage_lookup, create_task]\n write_scope: none | draft_only | workflow_task | ontology_action\n permission_mode: inherit_user_scope | service_role_with_policy\n citation_required: true\n freshness_required: true\n confidence_threshold:\n human_gate: before_write | before_publish | exception_only | none_with_reason\n eval_sets: [golden_questions, permission_denial, stale_data, ambiguous_terms, adversarial_queries]\n```\n\n## Content / Knowledge Assets\n\n| Asset | Required Metadata |\n|---|---|\n| Source catalog | source name, system, credential owner, schema, cadence, SLA, classification |\n| Pipeline registry | source, transform steps, schedule, retry, backfill, failure owner |\n| Data asset catalog | asset ID, layer, owner, certification, lineage, access policy |\n| Quality rule library | rule ID, severity, threshold, exception path, owner |\n| Business glossary | term, synonym, definition, domain, owner, ambiguity guard |\n| Semantic model | dataset, logical tables, relationships, metric/dimension bindings, examples |\n| Ontology model | object/link/action/query/rule types, write guards, side effects |\n| Metric dictionary | ID, name, layer, formula, unit, source, period, version |\n| Dimension dictionary | ID, values, hierarchy, value source, scope rule |\n| Report/template library | purpose, columns, sys/ext split, filters, snapshot/version |\n| Agent knowledge pack | instructions, examples, refusal rules, evaluation cases, citations |\n| Registration dossier | applicant/object, source and right evidence, review/public notice/objection, credential, validity and lifecycle history |\n| Authorization portfolio | approved plan, agreement, resource/product/service list, participant, purpose, usage policy, disclosure, evaluation and exit |\n| Digital-contract registry | parties, product, permitted/prohibited use, count/time/location, enforcement, revocation and evidence |\n| Dataset registry/card | version, intended use, provenance/right basis, population, processing, labels, splits, quality/benchmark, risks and consumers |\n| Labeling specification | task/schema/version, instructions, examples, annotator qualification, agreement, adjudication and acceptance sampling |\n| Model-data feedback ledger | model/application/eval version, failure slice, accepted feedback, dataset change, retraining decision and result |\n| Accounting/value evidence | economic purpose, costs, recognition judgment, disclosures, valuation assumptions and restrictions kept distinct from registration |\n| Runbook | freshness, incident, rollback, cost, access review, retirement |\n\n## Core Workflows\n\n| Workflow | Minimum Steps |\n|---|---|\n| multi-source onboarding | register source -> classify -> approve credential -> profile schema -> test sync -> publish source asset |\n| data pipeline delivery | define transform -> run sample -> validate quality -> schedule -> monitor -> backfill/rollback |\n| governance certification | assign owner -> classify sensitivity -> define policy -> verify lineage -> certify asset |\n| semantic/ontology modeling | define objects/metrics/dimensions/actions -> map source -> review -> publish -> version |\n| BI/dashboard delivery | choose certified data -> bind metric/dimension -> design drill/filter -> verify permission/freshness -> publish |\n| report/fill-in delivery | create template -> mark sys/ext -> set scope/period -> collect fill -> aggregate -> export/audit |\n| ChatBI enablement | prepare semantic model -> add glossary/examples -> evaluate golden questions -> authorize users -> monitor feedback |\n| data agent launch | define tools/write scope -> evaluate -> pilot -> monitor traces -> rollback/iterate |\n| public-data resource registration | inventory object -> prove provider/catalog/update/quality -> submit -> review/correct -> issue -> change/cancel |\n| data-property registration | choose type -> prove source/right boundary -> submit -> accept/review -> public notice/objection -> issue -> transfer/change/renew/cancel |\n| authorized operation | justify scenario/value -> approve plan -> select operator -> sign agreement -> register resource/product/service -> controlled delivery -> disclose/supervise/evaluate -> renew/exit |\n| trusted data use | onboard participant -> verify identity/qualification -> negotiate digital contract -> authorize purpose/action -> execute in controlled environment -> meter/audit -> settle/revoke |\n| high-quality dataset delivery | define model/scenario -> inventory rights/data gaps -> collect/process -> label/adjudicate -> split/isolate -> benchmark -> release card/version -> monitor |\n| model-data-scene iteration | detect model/application failure slice -> approve feedback sample -> relabel/augment -> release new dataset -> retrain/evaluate -> compare outcome |\n| data-resource accounting support | identify economic purpose/cost -> preserve evidence -> authorized accounting judgment -> disclosure/impairment review; never infer recognition from registration |\n| insight-to-action loop | detect signal -> explain/cite -> human confirm -> create task/action -> track outcome |\n| retirement | inventory consumers/contracts/models -> revoke/migrate -> freeze writes/use -> retain/export/delete -> close registration/accounting/audit |\n\n## Role Path Patterns\n\n| Role | Main Job | Critical Permissions |\n|---|---|---|\n| Sponsor / owner | approve business value, risk, budget, launch | release scope, risk acceptance |\n| 数据产品经理 | shape scenarios, metric value, lifecycle, acceptance | PRD, priority, evaluation |\n| 数据架构师 | design source-to-serving architecture | model/layer decisions |\n| 数据工程师 | build ingestion, processing, storage, quality | pipelines, schema, backfill |\n| 数据治理负责人 | own catalog, lineage, policy, quality | certification, access review |\n| 业务负责人 | approve metric definitions and exception rules | caliber/target decisions |\n| 分析师 | build dashboards, analysis paths, reports | semantic assets, dashboard publish |\n| 运营/监管人员 | create report/fill tasks, monitor progress, act on insights | report scope, reminders, returns |\n| 企业/一线填报人 | submit ext fields and correct returned data | own tasks and own history |\n| AI/算法工程师 | define agent runtime, eval, fallback, observability | prompts, tools, eval, traces |\n| 数据提供方/持有方 | prove source, quality, update and permitted use | register, authorize, correct and revoke owned/controlled scope |\n| 数据登记申请人/代理人 | submit truthful dossier and manage lifecycle | initial/transfer/change/renewal/cancellation within authority |\n| 登记机构审查人员 | accept, review, publicize, handle objections and issue credentials | independent review, reasoned return/reject, audit |\n| 实施机构/主管部门 | approve public-data operation plan and supervise | scope, operator selection, disclosure, evaluation and exit |\n| 运营机构 | develop and operate approved products/services without exceeding scope | controlled processing, product/service delivery, cost/revenue and audit |\n| 数据空间运营方 | onboard participants and enforce shared rules/digital contracts | identity, contract, usage control, metering, incident and exit |\n| 标注项目经理/领域标注员/质检员 | turn domain judgment into versioned labels | task allocation, qualification, annotation, agreement, adjudication and acceptance |\n| 模型/应用负责人 | define data fitness and feed evaluated failures back | benchmark, failure slices, retraining decision and outcome evidence |\n| 财务/法务/安全合规 | make separate accounting, contract, rights and security judgments | no cross-discipline automatic conclusion |\n| 管理层/查看者 | ask questions, inspect dashboards, decide actions | authorized metrics, data-product outcome and drill paths |\n\n## UI / Mobile Patterns\n\n| Surface | Pattern | Required States |\n|---|---|---|\n| source connection | connector list, credential drawer, schema preview, sync history | draft/connected/syncing/failed/degraded |\n| pipeline builder | DAG/step list, sample output, quality panel, run log | draft/test_run/scheduled/running/failed |\n| catalog/governance | asset card, owner, classification, lineage graph, access panel | raw/curated/certified/deprecated |\n| semantic/ontology manager | object/metric/dimension/action model, examples, publish diff | draft/reviewed/published/versioned |\n| BI/dashboard | metric cards, filters, drill path, freshness marker, export | loading/empty/no_permission/stale/error |\n| analysis workspace | drag/drop dimensions, charts, pivot, attribution, annotations | editing/saved/shared/locked |\n| ChatBI | question box, semantic match, generated query, answer, citations, feedback | thinking/answered/ambiguous/denied/stale |\n| Data Agent console | task/instruction, tool trace, confidence, human review, rollback | draft/evaluated/pilot/production/degraded |\n| registration workbench | application list, object/evidence form, review timeline, correction, notice/objection and credential | draft/submitted/returned/reviewing/notice/issued/rejected/changed/cancelled |\n| authorization operation | plan/agreement, resource-product-service relation, participant, disclosure, evaluation and exit | planned/approved/contracted/active/suspended/expired/exited |\n| trusted data-space console | participants, products, digital contracts, usage meter, violations, settlement and revocation | onboarding/active/denied/violation/suspended/exited |\n| dataset factory | demand, collection, processing, label schema, task/worker, sampling, split, benchmark, card and release diff | collecting/labeling/review/rejected/released/contamination_hold |\n| model-data feedback | model/application version, failure slices, sample review, dataset change and before/after evaluation | proposed/accepted/rejected/in_next_version/verified |\n| value cockpit | product/service usage, outcome, cost/revenue, settlement and renewal evidence | current/stale/disputed/under_review/closed |\n| fill-in mobile | task card, sys locked values, ext editable cells, save/submit | filling/submitted/returned/overdue/offline |\n\n## Policy / Privacy Constraints\n\n- Apply row-level, column-level, action-level, and agent-tool-level permissions before query generation, aggregation, export, or writeback.\n- Do not let agents bypass semantic model, source freshness, lineage, sensitivity labels, or user data scope.\n- Record who changed source schemas, pipelines, metrics, dimensions, ontology actions, permissions, prompts, eval sets, and report scopes.\n- Keep test/shadow records out of KPI, dashboard, BI, ChatBI, agent eval, and exported reports by default.\n- For sensitive or regulated data, define masking, desensitization, retention, deletion/export approval, access review, and audit retention.\n- For ontology/action agents, define human accountability for consequential state changes.\n- For public-sector delivery, specify private deployment, network boundaries, domestic database/OS constraints, and external model/data-exit restrictions when applicable.\n- Keep public-data resource registration, data-property registration, accounting recognition, appraisal/valuation, authorization, transaction and financing as linked but separately accountable decisions.\n- Do not directly or indirectly release uncontrolled non-public raw public data into the market; deliver approved products/services through the confirmed environment and scope.\n- Bind every participant and use to purpose, product, action, quantity, time, location, onward-transfer, retention, revocation and audit rules; a download permission alone is not a usage contract.\n- Training, validation, testing, preference and application-feedback data require provenance, rights, purpose, isolation and version lineage. Production feedback is opt-in/authorized and human-gated before dataset inclusion.\n- Protect held-out evaluation from training contamination, including retrieval corpora, synthetic derivatives, cached prompts, agent memory and human feedback.\n- Copyright/license, personal information, trade secrets, state/public interest, export/cross-border and sector rules are project-specific P0 questions; a dataset card does not cure an unlawful source.\n- Model performance gain does not by itself prove data-product value. Tie renewal, price or public benefit claims to an agreed baseline, measurement window and attribution boundary.\n\n## Domain Test Scenarios\n\n| Scenario | Must Verify |\n|---|---|\n| source schema change | impacted pipelines, semantic models, metrics, dashboards, agents, and exports are listed |\n| ingestion retry/backfill | duplicate records are not created and failed batches can replay |\n| cleaning/dedup | raw value is preserved; corrected value is versioned and auditable |\n| quality failure | downstream dashboards/agents show stale/quality warning or block strict use |\n| lineage trace | metric -> semantic model -> pipeline -> source field can be traced |\n| permission denial | unauthorized user/agent cannot query, export, or infer restricted rows/columns |\n| ChatBI ambiguous question | asks clarification or uses approved synonym; does not invent metric |\n| ChatBI stale data | answer cites freshness or refuses when strict freshness is required |\n| data agent writeback | low-confidence or disallowed ontology action is blocked and audited |\n| report/fill sys/ext split | sys fields are locked/refreshed; ext fields validate, submit, return, audit |\n| dashboard drill | filter/drill respects dimension hierarchy and row-scope permissions |\n| retirement | consumers are notified, migration/retention is completed, old asset is blocked |\n| metric definition version change | owner approval, old/new caliber, lineage impact, historical recalculation decision, consumer notification and rollback are linked |\n| data-agent permission denial | denied query/write cannot be bypassed by prompt, tool delegation or cached context; denial is audited |\n| registration lifecycle | initial/transfer/change/renewal/cancellation keep source/right evidence, review, notice/objection, credential validity and history linked |\n| registration-accounting separation | issued credential does not automatically mark the resource as accounting-recognized, valued, financeable or dispute-free |\n| authorized-operation scope | operator cannot use an unapproved resource/product/purpose or bypass disclosure, supervision and exit |\n| digital-contract enforcement | expired, revoked, over-count, wrong-purpose, wrong-region or onward-transfer use is blocked and evidenced |\n| dataset provenance/right withdrawal | affected samples, derived versions, models and applications are identified; freeze/retrain/delete/retain decisions are accountable |\n| labeling disagreement | qualification, agreement threshold, adjudication and rejected work remain visible by schema/version |\n| split leakage/contamination | entity/time/source/near-duplicate leakage and test-to-train feedback are detected before release |\n| preference-data quality | ranking/critique schema, annotator policy, safety taxonomy, agreement and model/version bias are testable |\n| model feedback flywheel | accepted failure slices create a new dataset version; old data and held-out benchmark are not silently mutated |\n| value verification | product usage, public/operational outcome, cost/revenue and attribution limits support renew/stop decisions |\n\n## Cross-Domain Requirement Patterns\n\n- `PAT-METRIC-CALIBER-001`: dashboards, reports, semantic metrics and value verification must share one reproducible caliber and version.\n- `PAT-LONG-RUNNING-JOB-001`: ingestion, backfill, index rebuild, export and dataset build need observable, idempotent and resumable task semantics.\n- `PAT-VERSION-COMPATIBILITY-001`: schema, metric, semantic model, label policy and dataset releases must preserve historical/in-flight interpretation.\n- `PAT-FEDERATED-RECONCILIATION-001`: use when registered, operated and source platforms retain overlapping authoritative attributes.\n\n## Evaluation Profile\n\nDomain knowledge is not execution evidence. Register coverage and maturity in\n`references/domain-coverage.yaml`; keep behavioral scenarios and run evidence\noutside this knowledge file.\n\nBefore raising maturity, independently evaluate:\n\n- one primary happy path;\n- one validation or exception path;\n- one permission/privacy path;\n- one lifecycle transition;\n- one coding-agent no-guess handoff path;\n- applicable migration, integration-failure, AI, and high-risk human-gate paths.\n\nRecord executor, input, environment, timestamp, result, and evidence location.\nMocked matrices and simulated reviewers cannot satisfy expert review or audit.\n\n## Acceptance Checklist\n\n- [ ] Every data source has owner, credential, schema, sync mode, cadence, SLA, backfill, retry, and classification.\n- [ ] Every pipeline has transform rules, quality gates, idempotency, sample evidence, lineage, and failure owner.\n- [ ] Every curated asset has catalog metadata, certification state, sensitivity label, access policy, and consumer list.\n- [ ] Every storage/retrieval path declares freshness, latency, retention, cost, permission, and index/materialization strategy.\n- [ ] Every semantic/ontology model defines objects/metrics/dimensions/actions/relations, examples, versioning, and rollback.\n- [ ] Every metric has ID, owner, layer, business definition, technical expression, source, period, unit, and quality rule.\n- [ ] Every dashboard/report shows source freshness, caliber version, permission state, empty/error state, and export rule.\n- [ ] Every report/fill workflow declares sys/ext fields, validation, submit/review/return state, deadline, and audit.\n- [ ] Every Data Agent/ChatBI feature has allowed sources/tools, permission inheritance, citations, freshness handling, refusal rules, eval sets, and human gate.\n- [ ] Every insight-to-action flow identifies who is accountable for final business decision and what the agent may not write.\n- [ ] Every registration flow declares object/type, applicant/agent, source/right evidence, review/return/reject, notice/objection, credential validity, change/renewal/transfer/cancellation and audit.\n- [ ] Every authorization operation declares approving/implementing/operating roles, resource-product-service scope, agreement/term, controlled environment, price/settlement, disclosure/supervision, evaluation and exit.\n- [ ] Every trusted use declares participant identity, product/purpose/action, digital contract, usage controls, metering/evidence, revocation and incident/liability path.\n- [ ] Every AI dataset version has intended use, provenance/right basis, population/coverage, processing, labeling, split isolation, quality/benchmark, risks/gaps, lineage, consumers and retirement.\n- [ ] SFT, validation, test, preference/DPO/RLHF and application-feedback data remain separately identifiable and cannot contaminate the held-out evaluation baseline.\n- [ ] Registration credential, accounting recognition, appraisal/valuation, transaction and financing decisions have separate owners, evidence and states.\n- [ ] Data-product value defines beneficiary, baseline, measurement window, cost/revenue or public outcome, attribution limit and renew/stop decision.\n- [ ] Acceptance tests cover registration, authorization, digital-contract denial, AI-data provenance/label/split/benchmark/feedback, accounting separation, ingestion, governance, BI/agent, permission, freshness, export and retirement.\n\nFile v5.5.2:references/domains/domain-education-it.md\n\n# Domain: Higher-Education Informationization / 高校教育信息化\n\nSource authority and freshness metadata: `references/domains/domain-sources.yaml`.\nCoverage and maturity: `references/domain-coverage.yaml`.\n\nUse this replaceable domain module for higher-education digital campus, vocational college digital transformation, academic affairs, student affairs, research administration, teaching quality, smart classroom, professional/program construction, one-stop service, data governance, data middle platform, AI assistant, and education data/BI scenarios.\n\nThis module is distilled from multi-year higher-education informationization materials. Keep customer/project details out of the public protocol; preserve reusable domain rules only.\n\n## Contents\n\n- Domain Purpose\n- First-Principles Domain Lens\n- Vocabulary\n- Aggregates and Entities\n- Domain Events\n- State Machines\n- Metric / Indicator Governance\n- AI Context Sources\n- Content / Knowledge Assets\n- Core Workflows\n- Role Path Patterns\n- UI / Mobile Patterns\n- Policy / Privacy Constraints\n- Domain Test Scenarios\n- Evaluation Profile\n- Acceptance Checklist\n\n## Domain Purpose\n\n- Business outcome: make university teaching, learning, student growth, research activity, campus service, quality diagnosis, professional construction, and data governance visible, serviceable, measurable, and continuously improvable.\n- Primary users: school leaders, academic affairs office, student affairs office, research office, teaching quality office, departments/colleges, program directors, professional-group leaders, teachers, counselors, students, data center/network office, IT administrators, external evaluators.\n- Sensitive areas: student identity, grades, awards/punishments, financial aid, mental-health risk, employment data, attendance/location, classroom video/audio, teacher evaluation, research projects, research funding, contracts/IP, personnel data, and cross-system master data.\n- AI may optimize: policy Q&A, one-stop service guidance, learning/career suggestions, counselor work assistance, data Q&A, classroom quality analysis, teaching report drafting, form/process filling, and risk hints.\n- AI must not decide automatically: student disciplinary action, scholarship/aid final result, psychological crisis classification, final grades, graduation eligibility, teacher performance conclusion, official evaluation result, or any legally/accountably binding school decision.\n- Representative system families to abstract from: academic affairs suites such as training-plan, course, scheduling, selecti...","readmeExcerpt":"Skill: AI Delivery Spec｜需求判断与交付 Owner: franklinxkk Summary: Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变 Tags: latest:5.5.2 Version history: v5.5.2 | 2026-09-29T23:","codeSnippets":[],"executableExamples":[{"language":"powershell","snippet":"python scripts/ai_delivery_spec_cli.py triage --input examples/minimal-v5/intake.yaml --format json\npython scripts/ai_delivery_spec_cli.py gate --profile prd --prd examples/minimal-v5/requirement-card.md --stage specify"},{"language":"bash","snippet":"npx skills add franklinxkk/ai-delivery-spec"},{"language":"text","snippet":"使用 ai-delivery-spec：给现有列表增加“仅看已启用”筛选，\n沿用系统已有的启用状态，默认显示全部，保留现有权限。\n把这次改动的规则和验收说明白。"},{"language":"text","snippet":"Use ai-delivery-spec: add an \"Enabled only\" filter to the existing list.\nReuse the existing enabled status, show all records by default, and preserve permissions.\nSpecify the rules and acceptance criteria for this change."},{"language":"bash","snippet":"python -m pip install -r scripts/requirements.txt\npython scripts/ai_delivery_spec_cli.py version\npython scripts/ai_delivery_spec_cli.py check\npython scripts/ai_delivery_spec_cli.py triage --input examples/minimal-v5/intake.yaml --format json\npython scripts/ai_delivery_spec_cli.py gate --profile prd --prd examples/minimal-v5/requirement-card.md --stage specify"},{"language":"bash","snippet":"python scripts/ai_delivery_spec_cli.py gate --profile prd --prd analysis.md --stage explore\npython scripts/ai_delivery_spec_cli.py query-domain --search confidence --limit 8\npython scripts/ai_delivery_spec_cli.py query-domain --search 完成率 --limit 8\npython scripts/ai_delivery_spec_cli.py query-domain --domain ai-native --section \"Metric / Indicator Governance\"\npython scripts/ai_delivery_spec_cli.py query-domain --domain medical-hospital-it --section \"Policy / Privacy Constraints\" --source-detail full --language en-US\npython scripts/ai_delivery_spec_cli.py gate --profile prototype --prototype app.html --prototype-baseline old-app.html"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: ai-delivery-spec\nlicense: Apache-2.0\ndescription: Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变更与验收；一句话想法、表单/表格规则和微小改动也适用。明确的纯翻译、排版、抄录或既定步骤执行由对应工具直接处理。\n---\n\n# AI Delivery Spec 5.5.2 — 需求判断与交付\n\n帮助用户管清需求，让产研和 AI 少猜、少漏、少返工。跟随用户语言；代码、字段和已有稳定 ID 保留原名。\n\n面向业务/产品、设计、前后端、QA 与 Coding Agent；从想法、存量材料或变更进入，各角色共用业务约定，不重复建文档。\n\n## 从当前目标进入\n\n识别用户要解决问题、比较方案、明确规则、获得原型，还是审查已有改变。已有决定直接继承；不因模板要求重复批准。`/ads`、`/dig`、`/prd`、`/proto` 只表达意图，宿主是否支持裸命令由实际能力决定。\n\n明确局部小改：读取相关基线，完成差异、继承边界和正反验收；没有关键未知就不提问、不建生命周期文件。不默认走全流程。\n\n按依赖推进：澄清当前决定 → 写清业务切片或走通原型 → 按需补追溯 → 验证。已有答案直接进入所需阶段。每阶段先解决业务分歧；编号、模板和门禁不能挤占业务内容。标题用业务名称，ID 留在引用位置。\n\n办公任务的目标取舍、口径、权限或流程改变同样适用；澄清后交对应工具执行。纯格式整理或既定步骤执行无需制造需求问题。\n\n| 当前需要 | 按需读取 |\n|---|---|\n| 判断问题、比较方案、澄清 | [discover.md](references/discover.md) |\n| 轻量规格、正式 PRD 与模块交接 | [specify.md](references/specify.md) |\n| 存量盘点、生成或修改可操作原型 | [prototype.md](references/prototype.md) |\n| 用户需要双态评审 | [review-workspace.md](references/review-workspace.md) |\n| 准入、处置、责任与基线 | [lifecycle.md](references/lifecycle.md) |\n| 变更、交接反馈或验收 | [change-acceptance.md](references/change-acceptance.md) |\n| 多文件、大上下文或跨会话 | [context.md](references/context.md) |\n| 机器路由、模板或检查命令 | [stages.md](references/stages.md) |\n\n不预加载全部模板或领域包。涉及行业规则，先按 [领域检索指引](references/stages.md#领域检索) 取当前问题切片和来源基线，再核实有权原文及版本；中英文同样执行。语言不决定法域，来源须匹配辖区与适用对象，不能照抄成项目真相。\n\n## 保持业务含义与决定权\n\n- 分清已核实事实、授权决定、观察、建议和未知。来源按主题、版本和授权范围判断；原型行为不自动成为产品规则。\n- 最新有效决定覆盖同主题旧内容。文档待同步、旧评审待复验不使该决定重新变成待批准。真正超出授权或存在冲突时只处理相应范围。\n- 建议暂缓、不做、缩范围写回现有产物；未获处置权不得改变需求状态。记录理由、依据和复议条件；已获授权不重复询问。\n- 未定规则保持未定。退路可以限制执行，不能借“保守默认”选定补考、口径、权限或晚到数据政策。未知只阻断依赖它的交付。\n\n## 最小充分规格\n\n实施者仍可能作出互不兼容的关键业务选择时，补足该处语义；技术实现保留合理空间。说明行为前提、允许者、业务结果、失败恢复及可判验收。按实际风险补状态、权限、指标、外部数据或历史对象约定，不按角色数或旧等级加长文档。\n\n事实只在一处人工定义，模块内就近引用或展开，使接收者连续读懂任务。业务审批和需求评审是不同对象；审核通过仅为发布前提时，不擅自合并成自动发布。\n\n关键规则用能区分错误实现的反例验证；标签缺失不能掩盖风险，模型一致不能代替来源。小改不附加全角色报告。\n\n正文 GAP 是待核实分歧，未命中也不证明通过。回读规则与来源，检查条件、反向路径和下游；不能为消除 GAP 编造政策、来源或 pass。\n\n## 变更与完成\n\n存量原型先盘点受影响页面、角色、入口、动作/处理器、状态、实体、数据源和 Mock 边界；保护未经取消的基线功能与视觉约定。产品态默认可操作；双态评审按用户需要启用。\n\n关键变更找到写入者、读取者、指标、入口、旧对象和受影响证据。候选依赖与核实结果分开，核实依赖不等于批准修改。多方修改前核对当前基线，不能静默覆盖漂移。\n\n达到用户目标就停止。检查只在需要的里程碑执行，不默认 full/handoff。静态、语义评阅、浏览器、真实系统与业务签署分别说明范围、版本及结果；没运行写未运行。小范围通过不代表全项目完成，建议被采纳不代表实现已验收。\n\n本 Skill 管需求及其产物；工程方案、排期、编码、部署和运营由相应工作流负责，只接收必要反馈与证据。私人材料与凭据不进入公共示例；外部写入遵守用户实际授权。"},{"path":"examples/minimal-v5/README.md","content":"# 最小需求示例\n\n这个虚构示例说明如何为制度列表增加“仅看当前有效”筛选。现有权限、时间边界、正常/空/失败结果与验收仍需说清，但不需要完整生命周期。\n\n```powershell\npython scripts/ai_delivery_spec_cli.py triage --input examples/minimal-v5/intake.yaml --format json\npython scripts/ai_delivery_spec_cli.py gate --profile prd --prd examples/minimal-v5/requirement-card.md --stage specify\n```\n\n预期为 card 路由建议及静态 PASS。建议不改变需求状态；PASS 不能证明自然语言风险发现完整、浏览器交互、真实系统或客户验收。\n\n示例保留一份可读取的旧式卡片，演示 5.4.x 内容兼容。新任务可用更短的模板或直接答复；状态、集成、指标等复杂点只补相关语义，不自动升级长 PRD。"},{"path":"README.md","content":"# AI Delivery Spec 5.5.2\n\n**帮你管清需求，让产研和 AI 少猜、少漏、少返工。**
\n**Manage requirements with less guesswork, fewer omissions and less rework—for your team and AI.**\n\n面向**产研团队与 AI Agent**的需求管理内核，以 Skill 形式使用。从一句话、现有材料或变更进入，帮你定清该做什么、交代清楚业务规则、找出改动影响。按当前任务生成或维护需求卡、PRD、可操作原型与验收条件，让接手者知道依据什么做、哪些还没定。\n\nA requirements management core for **product teams and AI agents**, delivered as a skill. Start with an idea, existing material or a change. Decide what needs doing, make business rules clear and identify change impacts. Create or update only the requirement cards, PRDs, interactive prototypes and acceptance criteria the task needs, so whoever takes over knows what to work from and what remains undecided.\n\n[![ClawHub downloads: 2.6k](https://img.shields.io/badge/ClawHub-2.6k_downloads-2563eb)](https://clawhub.ai/franklinxkk/skills/ai-delivery-spec)\n[![SkillHub AI score: 4.7/5](https://img.shields.io/badge/SkillHub_AI-4.7%2F5-f59e0b)](https://skillhub.cn/skills/user_12c92261/ai-delivery-spec)\n[![License: Apache 2.0](https://img.shields.io/badge/License-Apache_2.0-64748b)](LICENSE)\n\n2026-09-13 社区快照：ClawHub 约 2.6k 次下载；SkillHub 4.7/5 为 v5.4.8 历史 AI 评分，非当前版本新评测。 / Community snapshot: approximately 2.6k ClawHub downloads; SkillHub's 4.7/5 is a historical AI rating of v5.4.8, not a new evaluation of this release.\n\n**[角色价值 / Role value](#roles) · [中文上手](#zh) · [English guide](#en) · [四个快捷入口 / Shortcuts](#shortcuts) · [安装 / Install](#install) · [示例 / Examples](#examples) · [中英社区 / Community](#community)**\n\n\n\n## 各产研角色能得到什么\n\n| 核心用户 | 经常遇到的问题 | 这次能拿走什么 |\n|---|---|---|\n| **初级产品经理** | 收到一句需求，不知道该问什么、写到多细 | 关键问题、范围与边界、能开始评审的需求卡或 PRD |\n| **中高级产品 / 产品负责人** | 需求都合理，但优先做什么、跨模块如何一致还没定 | 问题证据、方案取舍、最小验证、当前决定与变更影响 |\n| **业务 / 售前 / 实施 / 设计** | 客户说法、业务规则和页面体验之间有断层 | 可确认的业务行为、可操作的产品原型、待决定事项 |\n| **前端研发** | 页面有了，入口、状态、权限和失败反馈仍不明确 | 与规格一致的交互路径、状态结果与验收条件 |\n| **后端 / 架构** | 同一句话会推导出不同口径、状态或写入方式 | 数据权威、允许的状态变化、副作用、恢复与集成边界 |\n| **QA / 验收方** | “显示正确”无法变成可重复的验收 | 正反例、权限与边界场景、变更回归范围和证据缺口 |\n| **Coding Agent** | 换个会话就丢背景，或自行补出业务政策 | 当前有效规则、来源与稳定引用、未知及可接续的任务范围 |\n\n适用于 ToC 产品、ToB/ToG 业务系统及 AI Native 场景。这些角色共用同一份业务约定，各自按需要读取；已有 PRD、需求系统和批准基线可以继续作为权威位置。\n\n## 从决定到交付，重点做好三件事\n\n**1. 更快找到当前要决定什么。** 从目标、受影响的人和事实出发，比较方案与最小验证。有依据的“先验证、暂缓、缩范围或不做”也可以完成分析；改变需求状态仍取决于实际授权。\n\n**2. 让规格可以体验，让评审有具体落点。** PRD 说明业务规则，原型呈现操作与结果，验收条件判断是否符合约定。“审批通过后可发布”要在规则、按钮行为与验收中区分发布资格和发布动作。需要双态评审时，可以边操作产品，边在当前页面旁查看规则、边界与验收依据。只补影响关键业务选择的内容，保留合理的工程实现空间。\n\n**3. 变更之后，相关产物仍然说同一件事。** 沿写入者、读取者、入口、指标和旧对象找具体依赖；区分候选影响与已核实影响，让 PRD、原型和交接引用同一规则。\n\n清晰小改直接完成；复杂需求按问题深入。工作量跟随当前目标，不要求先选 L0–L4、跑完整生命周期或填完全部模板。产品态原型默认可操作；需要面向产研的双态评审时，再开启评审标记与工作区。\n\n**先把业务讲清、操作走通，再补必要的追溯。** PRD 标题用业务名称，编号放在引用位置；快速原型不先搭全套评审设施。已定规则、待决影响和研发可自行选择的实现分别说清，避免把“有待决记录”误当成“已经可以开发”。
\n**Explain the business and make the task work before adding traceability.** Use business titles, keep IDs in references, and distinguish settled behavior, unresolved dependencies and engineering choices. A recorded question is not implementation readiness.\n\n\n\n## 安装到"},{"path":"_meta.json","content":"{\n \"ownerId\": \"kn7789n9h3n4chvd2jehwxy2jh88w52h\",\n \"slug\": \"ai-delivery-spec\",\n \"version\": \"5.5.2\",\n \"publishedAt\": 1790726245062\n}"},{"path":"references/change-acceptance.md","content":"# 变更、交接反馈与验收\n\n## 从一个有效决定开始\n\n记录来源、授权、基线、前后变化与影响种子；继承已有决定。写入前核对版本，按主题合并并发变化，不以最后写入者胜出。\n\n一条规则只改一个权威定义。页面、流程、原型说明、AC 与交接引用同步。显示序号变化不等于业务身份变化；需要替换身份时保存旧 ID 与替代关系，不改写历史。\n\n## 找消费者并保留依据\n\n从规则找到写入者、读取者、指标、操作入口、事件/通知、旧数据、进行中流程、验收与接收者。每项说明依赖链；区分直接、传递、仅回归和有理由的无影响。\n\n推测消费者记为候选，核实后更新同项及依据，排除保留理由。核实依赖不代表批准修改；高影响候选未核实就限制相应完成声明。\n\n`impact` 仅沿已登记关系找候选，暴露缺种子、未解析关系和深度截断；命令见 [工具](stages.md)，输入见 [变更模板](templates/change-request-template.yaml)。复杂审计才用 change-package Schema；小改写在当前产物。\n\n已有对象如何处理须依授权决定；不能默认全部迁移或沿用。记录哪些证据仍适用、哪些失效以及谁接收新版本。工程反馈回链为规格缺陷、实现缺陷、来源失效或新增范围，不把它们混成同一种返工。\n\n## 证据按主张、范围、版本成立\n\n| 证据 | 能证明 | 不能替代 |\n|---|---|---|\n| 静态检查 | 指定声明、引用、结构及可确定约束 | 自然语言业务真值和实际行为 |\n| 语义评阅 | 指定对象的来源核对、反例及处置 | 运行验证；两模型一致也不是真值 |\n| 浏览器原型 | 指定环境的操作、可见/Mock结果 | 真实接口、持久、并发与生产权限 |\n| 真实系统 | 指定版本环境中的实际对象与结果 | 客户批准规则或验收签署 |\n| 业务/客户确认 | 有权人在对应范围作出的决定或接受 | 实现已完成、范围外能力或上线 |\n\n关键主张绑定范围、版本、证据类型、环境/责任人、结果与位置。验证后修改重验受影响范围；未执行、缺来源、不适用分开。\n\n断言数、唯一 AC 数和已验证行为分开；编号注释不算覆盖。按同一输入核对来源→PRD→原型实际操作→实现结果，检查对象/字段含义、状态、权限、口径、拒绝和副作用，允许有明确映射的不同命名。AC 前提须能独立造数或重置；删除对象使旧 AC 失效。声明可配置的规则，用改值前后样本验证各消费者响应及生效时点，不把所有阈值升级成配置平台。\n\n复用 ARUN/EVD；真实验收保留 acceptance_ref、requirement_refs、result、actual_result、evidence_refs，见 [验收模板](templates/acceptance-run-template.yaml)。本地证据可解析且不越出记录目录；远程链接或字符串不证明执行。\n\n## 关闭\n\n实际验证覆盖指定主路径及高价值拒绝/恢复；验收接受需相应有权签署。残留问题明确范围、责任及复议条件；有条件批准到期重新处理，不用条件措辞掩盖未完成的关键规则。\n\n需求通过验收不等于已经发布。外部上线、运营和结果反馈只记录引用，必要时作为新来源或 CHG 返回；不由需求状态自动驱动生产操作。"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变 Skill: AI Delivery Spec｜需求判断与交付 Owner: franklinxkk Summary: Clarify, create, review or change product, service and office workflow requirements, PRDs and interactive prototypes. Use for vague goals, process improvements, spreadsheet/form rules, small UI edits and changes to existing systems, even without the word requirement. 中文：产品、服务及办公流程的需求判断、深挖澄清、PRD、原型、变 Tags: latest:5.5.2 Version history: v5.5.2 | 2026-09-29T23:","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":1290,"uniquenessScore":48,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-09T10:56:40.890Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-09T10:56:40.890Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-10T02:10:06.419Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}