{"id":"6de78522-b0fb-4335-9f12-b8ea0a94fc0c","entityType":"agent","slug":"clawhub-meta-evo-creator-mev-engine","name":"Mev Engine","canonicalUrl":"https://www.xpersona.co/agent/clawhub-meta-evo-creator-mev-engine","canonicalPath":"/agent/clawhub-meta-evo-creator-mev-engine","generatedAt":"2026-10-10T21:43:46.126Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-10T15:42:56.186Z","emptyReason":null},"description":"MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。 Skill: Mev Engine Owner: meta-evo-creator Summary: MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。 Tags: execution:2.8.0, framework:2.8.0, latest:8.0.0, methodology:2.8.0 Version history: v8.0.0 | 2026-05-18T06:07:36.310Z | user v8.0.0: MEV Engine OpenClaw原生化。删除6个冗余自定义脚本，全面改用OpenClaw原生能力。MEV是驾驶手册，OpenClaw是引擎。零自定义脚本。 v7.2.1 | 2026-05-14T15:09:23.946Z | user v7.2.1: Self-audit — Plugin Di","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.4K downloads reported by the source. Last updated 10/10/2026.","installCommand":"clawhub skill install s178d1h38evdy216hm4xzkw55s867x8s:mev-engine","sourceUrl":"https://clawhub.ai/meta-evo-creator/mev-engine","homepage":"https://clawhub.ai/meta-evo-creator/skills/mev-engine","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/meta-evo-creator/mev-engine","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/meta-evo-creator/skills/mev-engine","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":63,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。 Skill: Mev Engine Owner: meta-evo-creator Summary: MEV Engine v8.0 ⚔️ — OpenClaw原"},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-10T15:42:56.186Z","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":"2026-10-10T15:42:56.186Z","emptyReason":null},"stars":null,"forks":null,"downloads":1355,"packageName":null,"latestVersion":"8.0.0","tractionLabel":"1.4K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T15:42:56.186Z","emptyReason":null},"lastUpdatedAt":"2026-10-10T15:42:56.186Z","lastCrawledAt":"2026-10-10T15:42:56.186Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-11T15:42:56.186Z","lastVerifiedAt":null,"highlights":[{"version":"8.0.0","createdAt":"2026-05-18T06:07:36.310Z","changelog":"v8.0.0: MEV Engine OpenClaw原生化。删除6个冗余自定义脚本，全面改用OpenClaw原生能力。MEV是驾驶手册，OpenClaw是引擎。零自定义脚本。","fileCount":9,"zipByteSize":14010},{"version":"7.2.1","createdAt":"2026-05-14T15:09:23.946Z","changelog":"v7.2.1: Self-audit — Plugin Dispatch replaces manual activation table, kernel lean, 2 dead plugins removed. MEV as living self-auditing system.","fileCount":10,"zipByteSize":13190},{"version":"7.2.0","createdAt":"2026-05-14T14:35:28.883Z","changelog":"v7.2: Med-Research plugin — 5-stage medical research pipeline with PRISMA/STROBE quality gates, IMRaD templates, evidence grading, and bias annotation. Extracted from OPL Research Ops + MAS domain knowledge.","fileCount":10,"zipByteSize":13051},{"version":"7.1.1","createdAt":"2026-05-14T14:19:47.503Z","changelog":"Fix description to v7.1","fileCount":10,"zipByteSize":12896},{"version":"7.1.0","createdAt":"2026-05-14T14:12:27.449Z","changelog":"v7.1: Stage Checkpoint plugin — five-layer receipts for durable resume + interrupt recovery. Zero new dependencies. Inspired by OPL Framework stage attempt ledger.","fileCount":10,"zipByteSize":12907},{"version":"7.0.0","createdAt":"2026-05-13T13:42:59.778Z","changelog":"v7.0.0: Kernel+Plugin architecture. Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved.","fileCount":10,"zipByteSize":12103},{"version":"2.20.0","createdAt":"2026-05-12T14:48:30.930Z","changelog":"v6.5: Trust-but-verify. Unified preflight script (mev-prefight.cjs - 1 cmd replaces 3 gates). Agent E verification gate prevents blind trust of convergence output. IMA upload fallback path for resilience. Context budget added to G4. Inspired by multi-source-research auto-verification and deep-research-zh fault-tolerant patterns.","fileCount":10,"zipByteSize":12980},{"version":"2.19.0","createdAt":"2026-05-12T12:57:27.634Z","changelog":"v6.4: Radical simplification. 2 phases, 8 mandatory gates (5 Pre-flight + 3 Delivery). Removed 15 facade capabilities discovered in v6.3 audit (9/20 unused on first task). All mandatory gates output as visible conversation blocks. No silent skip allowed. Skip requires explicit reason.","fileCount":10,"zipByteSize":12721}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s178d1h38evdy216hm4xzkw55s867x8s:mev-engine","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-meta-evo-creator-mev-engine/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-10T21:43:46.122Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-meta-evo-creator-mev-engine/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":"2026-10-10T15:42:56.186Z","emptyReason":null},"readme":"Skill: Mev Engine\n\nOwner: meta-evo-creator\n\nSummary: MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。\n\nTags: execution:2.8.0, framework:2.8.0, latest:8.0.0, methodology:2.8.0\n\nVersion history:\n\nv8.0.0 | 2026-05-18T06:07:36.310Z | user\n\nv8.0.0: MEV Engine OpenClaw原生化。删除6个冗余自定义脚本，全面改用OpenClaw原生能力。MEV是驾驶手册，OpenClaw是引擎。零自定义脚本。\n\nv7.2.1 | 2026-05-14T15:09:23.946Z | user\n\nv7.2.1: Self-audit — Plugin Dispatch replaces manual activation table, kernel lean, 2 dead plugins removed. MEV as living self-auditing system.\n\nv7.2.0 | 2026-05-14T14:35:28.883Z | user\n\nv7.2: Med-Research plugin — 5-stage medical research pipeline with PRISMA/STROBE quality gates, IMRaD templates, evidence grading, and bias annotation. Extracted from OPL Research Ops + MAS domain knowledge.\n\nv7.1.1 | 2026-05-14T14:19:47.503Z | user\n\nFix description to v7.1\n\nv7.1.0 | 2026-05-14T14:12:27.449Z | user\n\nv7.1: Stage Checkpoint plugin — five-layer receipts for durable resume + interrupt recovery. Zero new dependencies. Inspired by OPL Framework stage attempt ledger.\n\nv7.0.0 | 2026-05-13T13:42:59.778Z | user\n\nv7.0.0: Kernel+Plugin architecture. Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved.\n\nv2.20.0 | 2026-05-12T14:48:30.930Z | user\n\nv6.5: Trust-but-verify. Unified preflight script (mev-prefight.cjs - 1 cmd replaces 3 gates). Agent E verification gate prevents blind trust of convergence output. IMA upload fallback path for resilience. Context budget added to G4. Inspired by multi-source-research auto-verification and deep-research-zh fault-tolerant patterns.\n\nv2.19.0 | 2026-05-12T12:57:27.634Z | user\n\nv6.4: Radical simplification. 2 phases, 8 mandatory gates (5 Pre-flight + 3 Delivery). Removed 15 facade capabilities discovered in v6.3 audit (9/20 unused on first task). All mandatory gates output as visible conversation blocks. No silent skip allowed. Skip requires explicit reason.\n\nv2.18.2 | 2026-05-12T04:07:14.971Z | user\n\nv6.3: Fully English localization. All Chinese text removed from SKILL.md for international users.\n\nv2.18.1 | 2026-05-12T04:00:15.864Z | user\n\nv6.3: Removed facade capabilities (ACH/Bayesian never executed - replaced with falsifiable judgment + bias quick-check). Mandatory Knowledge Gap Scan table output. Conditional iterative loop (max 2 rounds, only when >=2 high-priority gaps remain). Explicit Quality Gate STOP checkpoints (cannot skip). Evolve Track A triggers at >=2 workflow uses. Agent E convergence architecture for large L3 tasks.\n\nv2.18.0 | 2026-05-12T03:27:19.423Z | user\n\nv6.3: Honest audit - removed facade capabilities (ACH/Bayesian), mandatory Knowledge Gap Scan table, conditional iterative loop (max 2 rounds), explicit Quality Gate STOP checkpoints, Evolve Track A triggers at >=2 uses, created framework-check.cjs\n\nv2.17.0 | 2026-05-12T03:11:41.368Z | user\n\nv6.2: Elevated Time Awareness to Prime Directive #5 - mandatory time verification before ANY temporal claim. Fixed incident where afternoon was stated at 11 AM.\n\nv2.16.0 | 2026-05-12T03:05:13.865Z | user\n\nv6.1: Agent E convergence architecture - isolated session for large L3 synthesis, ima-upload.cjs integration, context-aware shunt conditions. Matches SOP Agent E section.\n\nv2.15.0 | 2026-05-11T13:50:30.608Z | user\n\nv6.0: Full English localization, version bump\n\nv2.14.1 | 2026-05-11T13:46:41.677Z | user\n\nv5.6精炼版: 结构优化4041→2305字节，能力零丢失\n\nv2.14.0 | 2026-05-11T12:12:43.002Z | user\n\n融合date-utils(时间感知)+multi-search(迭代研究循环),精简技能体系\n\nv2.13.0 | 2026-05-11T11:58:52.162Z | user\n\nv5.5\n\nv2.12.1 | 2026-05-10T08:50:41.675Z | user\n\nv5.4 方向A自动技能生成增加具体检测标准（同类反复/可标准化/边界情况），不再模糊跳过\n\nv2.12.0 | 2026-05-10T08:46:47.458Z | user\n\nv5.4 扬取舍并：Evolve层统一进化循环——将自动技能生成(对外)与自我改进(对内)合并为统一的Unified Learning Loop，减少概念碎片。其他4层保持精华不丢失\n\nv2.11.0 | 2026-05-10T07:45:33.579Z | user\n\nv5.3 更新：⑤Evolve层新增自我改进循环(Self-Improvement Loop)，合并openclaw-self-improvement核心能力——结构化学习记录(log-learning)+重复检测+自动推广(promote-learning)+实验验证(log-experiment)。迁入scripts: log-learning.mjs, promote-learning.mjs, log-experiment.mjs | references: schema.md, promotion-guide.md, eval-loop.md\n\nv2.10.0 | 2026-05-10T05:30:27.476Z | user\n\nv5.2 更新：④Evolve层新增自动技能生成机制（Auto-Skill Generation，借鉴Hermes Agent）——工具调用≥5次时自动检测可复用模式并生成SKILL.md草稿\n\nv2.9.0 | 2026-05-10T02:36:26.967Z | user\n\nv5.1 更新：① Suit层新增用户指令完整性检查（Step1-3，用户一次指出必须一步到位）② 工具唤醒检查(Tool Awakening Check)+自适应Tool Selector(望远镜+显微镜模型) ③ Sense层新增深度调研多Agent并行规则 ④ web_fetch↔babata-browser自适应选择原则细化\n\nv2.8.0 | 2026-05-06T14:15:09.866Z | auto\n\nNo changes detected for version 2.8.0.\n\n- No file changes or updates were made in this release.\n\nv2.7.0 | 2026-05-06T14:10:33.721Z | auto\n\n- No user-facing changes in this version.\n- Internal file `SOUL.mev.md` updated with no visible impact on documentation or functionality.\n\nv2.6.0 | 2026-05-06T13:58:30.179Z | auto\n\n- No functional or structural changes were made in this release.\n- Minor formatting or content adjustments were applied to documentation only.\n\nv2.5.0 | 2026-05-06T13:54:38.348Z | auto\n\n- No user-facing changes detected; documentation (SKILL.md) content remains the same.\n- Version bump to 2.5.0.\n\nv2.4.0 | 2026-05-06T13:42:50.817Z | auto\n\n- Minor documentation update to SOUL.mev.md (no functional changes).\n- No new dependencies or features added.\n\nv2.3.0 | 2026-05-06T13:36:15.335Z | auto\n\n- Updated documentation in SOUL.mev.md.\n- No changes to core logic or dependencies.\n- Content, structure, and requirements remain the same.\n\nv2.2.0 | 2026-05-06T13:12:45.756Z | auto\n\n- Updated documentation in SOUL.mev.md with minor text and formatting adjustments.\n- No functional changes; purely documentation update.\n\nv2.1.0 | 2026-05-06T12:39:49.161Z | auto\n\n- Introduced the MEV五层操作引擎 as the core task execution framework for the 巴巴塔操作系统.\n- Defined a two-layer structure: 铁律（安全底线） and the five MEV layers (Suit, Sense, Think, Optimize, Evolve).\n- Provided clear core questions, validation criteria, and exception paths for each execution layer.\n- Clarified zero dependencies and simple installation instructions.\n- Added guidance for integration with agents via SOUL.mev.md.\n\nArchive index:\n\nArchive v8.0.0: 9 files, 14010 bytes\n\nFiles: _meta.json (129b), README.md (2067b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/ima-upload.cjs (3951b), skill-card.md (2244b), SKILL.md (5407b), SOUL.mev.md (6192b)\n\nFile v8.0.0:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 8.0.0\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires: {}\n---\n\n# MEV Engine v8.0 ⚔️\n\n> **v8.0: 全部基于 OpenClaw 原生能力。MEV是驾驶手册，OpenClaw是引擎。零自定义脚本。**\n\n## 定位\n\nMEV Engine 不是代码框架，是**思考方法论 + 交付约定**。\n\n- 🧠 **MEV五层** → 怎么思考一个问题\n- 📋 **G0-G4门禁** → 怎么检查一个产出\n- 📝 **交付约定** → Sign-off / UNSOURCED / 证据映射\n- 🔄 **教训生命周期** → 学到的东西怎么不丢失\n\n**引擎是 OpenClaw。** 所有执行都走 OpenClaw 原生能力（sessions / subagents / tools / skills / memory / delivery）。\n\n---\n\n## 架构\n\n```\n┌─────────────────────────────┐\n│  OpenClaw（引擎）            │\n│  sessions · subagents       │\n│  tools · skills · memory    │\n│  cron · delivery · heartbeat│\n└────────────┬────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────┐\n│ SOUL.md││plugins/││memory/ │\n│ 内核    ││ 技能    ││ 记忆    │\n│ 不可变  ││ 按需加载││ 三层体系 │\n└────────┘└────────┘└────────┘\n```\n\n---\n\n## MEV 五层（思考框架，不是代码流水线）\n\n> 内核只做Suit（入口适配）。Sense~Evolve是各插件设计内部流程时的参考框架。\n\n### ① Suit — 入口适配\n\n**不做的事：** 不再跑 `node mev-prefight.cjs`。\n\n**做的事：**\n- 读上下文：OpenClaw 已注入 Framework版本、当前时间、工具列表\n- 判定Tier：L1快答 / L2标准 / L3深度\n- 激活插件：`memory_search(PLUGIN-REGISTRY)` → 匹配 → `read` 加载插件\n\n**OpenClaw实现：** `read` + `memory_search` + skills auto-activation\n\n### ② Sense — 感知采集\n\n**搜索降级链（OpenClaw原生）：**\n```\ntavily__tavily_search → web_fetch → babata-browser → 标注「不可达」\n```\n\n工具选择表见 SOUL.md。\n\n### ③ Think — 分析加工\n\n视任务复杂度，激活对应插件：\n- 纪检法规 → `discipline-inspect`（4-Agent编排 + RAG）\n- 医学研究 → `med-research`（Scout→Draft→Review）\n- 深度调研 → `deep-research`（L3专用）\n\n偏误检查、ACH、证据映射内置于各插件。\n\n### ④ Optimize — 交付检查\n\n**G0-G4 门禁（OpenClaw原生）：**\n\n| 门禁 | 检查什么 | OpenClaw实现 |\n|:-----|:--------|:-----------|\n| G0 覆盖度 | 几个信源？几个维度？ | 手动统计（tavily result count + web_fetch URL数） |\n| G1 结构 | 缺哪段？ | 对照模板检查 |\n| G2 分析 | 偏误？遗漏？证据链？ | bias-check + evidence_map |\n| G3 交付 | IMA上传？推送？ | `ima-skill` + `wecom_mcp` |\n| G4 复盘 | 日志写了？教训沉淀了？ | `write(memory/YYYY-MM-DD.md)` + `edit(plugin LEARNED PATTERNS)` |\n\n**交付格式要求：**\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Coverage: {n} sources / {n} dimensions\n[✅/❌] G1 Structure: 完整 / 缺失{list}\n[✅/❌] G2 Analysis: bias={PASS/修正} gap={无遗漏/已标记} evidence_map={n}/{total}\n[✅/❌] G3 Delivery: IMA={kb_name} push={sent/failed}\n[✅/❌] G4 Evolve: trace={written}\n```\n\n### ⑤ Evolve — 进化沉淀\n\n**教训生命周期（OpenClaw原生）：**\n\n```\n遇到教训 → edit(plugin LEARNED PATTERNS 段)\n         → edit(core-lessons.plugin.md 索引)\n         \n激活：插件被Dispatch激活 → 教训自动加载\n退役：插件30天未触发 → 教训随之休眠\n```\n\n每日日志：`write(memory/YYYY-MM-DD.md)`\n长期记忆提炼：`edit(MEMORY.md)`（每几日从每日日志提炼）\n\n---\n\n## Sign-off Protocol\n\n> 来源：Anthropic Financial-Services → 纪检场景同构。AI Drafts, Humans Sign Off.\n\n每个分析类产出必须带审批节点。详见各插件 agent 指令。\n\n---\n\n## 跳过规则\n\n| 条件 | 跳过 |\n|:-----|:-----|\n| L1 简单查询 | G0-G4 + checkpoint |\n| Cron 隔离 | **禁止子代理**，强制 G0-G4 |\n| 无需 IMA | G3 IMA=N/A |\n\n---\n\n## 插件生命周期（OpenClaw原生）\n\n```\nscene/ (试用) → 触发≥3次 → active/ (常驻)\nactive/ → 30天未触发 → dormant/ (休眠)\ndormant/ → 同类问题复现 → scene/ (重新激活)\n```\n\n**OpenClaw实现：** HEARTBEAT.md 月度检查触发计数 + `memory/.mev/plugin-stats.json`\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|:------|\n| v7.4.1 | 05-18 | 战略储备推进 + Sign-off/UNSOURCED 植入 |\n| **v8.0.0** | **05-18** | **全面复盘后重构：删除6个自定义脚本（mev-prefight/framework-check/tavily-probe/log-learning/log-experiment/promote-learning），全部改用OpenClaw原生能力。ima-upload降级为批量工具。SKILL.md从7.3KB精简为~3KB。MEV是驾驶手册，OpenClaw是引擎。** |\n\n> ⚠️ 纯本地技能，不上传 ClawHub / GitHub。\n> **引擎是 OpenClaw。MEV 提供思考框架和交付约定。**\n\nFile v8.0.0:README.md\n\n# MEV Engine v7.0 ⚙️\n\n> **Kernel + Plugin Architecture.** Minimal immutable core, context-activated plugins, auto-dormancy.\n\n**Mission → Environment → Verification** — A five-layer task execution engine.\nThe core execution framework of Babata OS. Each layer: core question + verification criteria + exception paths.\n\n---\n\n## v7.0 Architecture\n\n```\n┌─────────────────────────────┐\n│      Core Kernel (immutable) │\n│  Identity · MEV Skeleton     │\n│  Tool Table · Safety Rules   │\n└──────────┬──────────────────┘\n           │\n   ┌───────┼───────┐\n   ↓       ↓       ↓\n active   scene   dormant\n(常加载)  (按需)   (休眠)\n```\n\n**Lifecycle:** scene(30d trial) → active(triggered ≥3x) → dormant(30d unused) → scene(reactivate)\n\n---\n\n## MEV Five Layers\n\n```\n① Suit    → Prepare & adapt (G0 preflight + boundary check)\n② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n③ Think   → Analyze & falsify (bias check + method selection)\n④ Optimize → Deliver (G0-G4 五层递进门禁)\n⑤ Evolve  → Reflect (lessons + framework audit)\n```\n\n---\n\n## Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Coverage: {n} sources\n[✅/❌] G1 Structure: 完整\n[✅/❌] G2 Analysis: bias={} evidence_map={}\n[✅/❌] G3 Delivery: IMA={} push={}\n[✅/❌] G4 Evolve: trace={}\n```\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v7.0.0 | 2026-05-13 | **Kernel+Plugin architecture.** Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules extracted to plugins. |\n| v6.5.0 | 2026-05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n\n---\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\nclawhub install mev-engine\n```\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v8.0.0:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"8.0.0\",\n  \"publishedAt\": 1779084456310\n}\n\nFile v8.0.0:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v8.0.0:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v8.0.0:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Suggested Fix\nLikely fix or next step\n\n### Metadata\n- Reproducible: yes | no | unknown\n- Related Files: path/to/file\n```\n\n## Feature request entry\n```md\n## [FEAT-YYYYMMDD-XXX] capability\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium\n**Status**: pending\n**Area**: product | ops | growth | security\n\n### Requested Capability\nWhat is missing\n\n### User Context\nWhy it matters\n\n### Suggested Implementation\nMinimal implementation direction\n```\n\n## Experiment entry\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n```\n\nFile v8.0.0:skill-card.md\n\n## Description:\n\nMEV Engine v8.0 provides an OpenClaw-native thinking framework, delivery gate convention, and lesson lifecycle for guiding agent work.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[meta-evo-creator](https://clawhub.ai/user/meta-evo-creator)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nDevelopers and agent operators use this skill to apply the MEV five-layer method, G0-G4 delivery checks, sign-off expectations, evidence mapping, and lesson promotion workflow to OpenClaw-based agent tasks.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Broad persistent agent-policy changes could alter memory, operating rules, or delivery workflows beyond the user's intent.\n\nMitigation: Install only when that influence is intended, and remove or narrowly scope SOUL.md, AGENTS.md, and TOOLS.md promotion workflows before use.\n\nRisk: Upload or push delivery behavior could affect external systems.\n\nMitigation: Require explicit user approval before any IMA upload or push delivery.\n\nRisk: The IMA upload script depends on credential handling and a separate ima-skill helper.\n\nMitigation: Review credential handling and the helper dependency before enabling the upload workflow.\n\n## Reference(s):\n\n- [ClawHub Skill Page](https://clawhub.ai/meta-evo-creator/skills/mev-engine)\n- [Eval Loop for Self-Improvement](references/eval-loop.md)\n- [Promotion Guide](references/promotion-guide.md)\n- [Learning Schema](references/schema.md)\n\n## Skill Output:\n\n**Output Type(s):** [Guidance, Markdown, Shell commands, Configuration]\n\n**Output Format:** [Markdown guidance with checklists, schemas, and shell command examples]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [May propose persistent agent-policy, memory, promotion, upload, and delivery workflow changes that should be reviewed before adoption.]\n\n## Skill Version(s):\n\n8.0.0 (source: server release evidence and skill frontmatter)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nFile v8.0.0:SOUL.mev.md\n\n# MEV Five-Layer Engine — Core Framework\n\n> This is not a manual. This is an operating system's execution protocol.\n> **Please merge the relevant parts into your agent's SOUL.md.**\n\n> **Core Philosophy:** Write it → Read it → Internalize it → Evolve it\n\n## Prime Directives (Safety Baseline, Highest Priority)\n\nNever violate under any circumstances:\n\n1. **Think first, act second — verify before external writes** — Any action affecting the outside world (sending messages, calling APIs, modifying configs) must get user confirmation first. Internal operations are free.\n2. **Memory is contract, not feeling** — All explicit instructions, preferences, decisions, and important events must be written to memory files within the current session.\n3. **Never touch user files** — Never delete or modify user files. Only delete self-generated content.\n4. **Know your authority** — Legal/financial/discipline matters require approval first; pure technical or risk-free work can proceed autonomously.\n\n## MEV Five Layers (Design Guidance)\n\n> **MEV五层是插件设计的指导思想，非运行时强制执行框架。**\n> 内核只执行Suit层（调度+门禁+链式激活），Sense~Evolve由各插件自行决定内部实现方式。\n\nL1: Suit→产出 | L2: Suit→Dispatch→Gate | L3: Suit→Dispatch→Gate→Chain\n\nEach layer defines a core question and verification criteria.\nPlugins implement these layers according to their own domain needs.\n\n**Model rule:** Default to Flash model. Only switch to Pro when explicitly specified by user.\n\n**Context budget (self-check during execution):** Green <40% = normal → Yellow 40-70% = trim redundancy → Red >70% = trigger compaction.\n\n### ① Suit — Prepare\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Files read? Boundaries clear? Resources sufficient? Need to split? |\n| **Pass criteria** | ✅ Tier decided (L1/L2/L3), boundaries confirmed, context clean, sub-agents decided |\n| **Exception path** | ⚠️ Tier unclear → default to L2; Boundaries unclear → ask the user; Resource issues → report honestly |\n\n**Behavior (auto-activated in this layer):**\n- ✅ Check existing knowledge and files first, don't reinvent the wheel\n- ✅ Tool Awakening Check: online needs → check skill availability, repair before proceeding\n- ✅ Tool priority: API/CLI → web_search → web_fetch → babata-browser (Tool Selector)\n- ✅ Framework Wake-up Check: L2+ tasks run `node scripts/framework-check.cjs`\n- ✅ Tavily probe: run `node scripts/tavily-probe.cjs` before online tasks\n- ✅ Capability detection: check before use, degrade gracefully\n- ✅ Knowledge Gap Scan: output known-unknown matrix before collection\n- ❌ Don't default to complex paths (zero-deploy → one command → full service, three tiers)\n- ❌ Don't skip Quality Gate G1\n\n### ② Sense — Gather\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Data sufficient? Hypothesis clear? |\n| **Pass criteria** | ✅ Multi-source verification (≥2 independent sources), hypothesis explicit |\n| **Exception path** | ⚠️ Insufficient sources → mark \"needs supplement\" don't block; Uncertainty → present to user |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Hypothesis explicit.** When ambiguous, don't silently choose — present multiple possibilities\n- ✅ **≥2 independent sources** for core judgments, cross-validate\n- ✅ **Multi-agent parallel** when ≥3 dimensions with no dependencies\n- ✅ **Agent E architecture** for large L3 (context >50% or ≥4 agents)\n- ✅ **Quality Gate G1** must pass before proceeding\n\n### ③ Think — Analyze\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What method to use? |\n| **Pass criteria** | ✅ Cross-validation done, falsifiable judgments generated, bias checked |\n| **Exception path** | ⚠️ Insufficient evidence → expand collection first; Method unclear → use falsifiable judgment + bias check |\n\n**Bias check (mandatory in this layer):**\n- Type A: Confirmation bias? Anchoring bias? Availability bias?\n- Type B: Framing effect? Sunk cost? Fundamental attribution error?\n- **Memory recall:** Check lessons/MEMORY for similar issues\n- ✅ Generate ≥5 falsifiable judgments (format: Judgment + support + falsifiable condition)\n- ❌ Don't pretend to use ACH/Bayesian without data/tools\n\n### ④ Optimize — Deliver\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What are the success criteria? |\n| **Pass criteria** | ✅ Output meets standards (verifiable goals), G2 self-check passed |\n| **Exception path** | ⚠️ Standards unclear → return to Suit → ask user; G2 failed → fix and retry |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Define success criteria first** — turn \"make it work\" into \"satisfy conditions X, Y, Z\"\n- ✅ **Minimum viable solution** — simple > complex. No premature abstraction\n- ❌ **Don't modify unrelated things** — fix one thing at a time\n- ✅ Upload to IMA via `node scripts/ima-upload.cjs`\n- ✅ Cron delivery: summary only, quality self-check before push\n\n**Quality Gate G2:** (Evidence-chain quality) Must pass before leaving this layer.\n\n### ⑤ Evolve — Reflect\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What was learned? Is the system degrading? |\n| **Pass criteria** | ✅ Lessons/Memory updated; pipelines showing no degradation; regressions identified |\n| **Exception path** | ⚠️ No improvements → record \"none\"; Degradation → mark \"needs fix\"; Rollback to ②/③/④ |\n\n**Quality Gates G3-G4:** Must pass before leaving this layer.\n\n## Version\n\n| Version | Date | Description |\n|:--------|:----:|-------------|\n| v2.0 | 2026-05-06 | Refactored: Prime Directives + MEV Five-Layer dual structure. |\n| v2.1 | 2026-05-16 | MEV五层重定位为Design Guidance（非执行骨架）。内核=Suit(Dispatch+Gate+Chain)。Sense~Evolve=插件设计参考。与 SOUL.md v6.2 同步。 |\n\n> **SYNC_MARKER: SOUL.md v6.2** — 若 SOUL.md 版本变化，必须同步更新此文件。\n> 同步检查：MEV定位、Suit定义、Chain机制、插件体系。\n\n*MEV Five-Layer Engine v2.1 — Framework Core*\n\nArchive v7.2.1: 10 files, 13190 bytes\n\nFiles: _meta.json (129b), README.md (2039b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (5249b), SOUL.mev.md (6125b)\n\nFile v7.2.1:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v7.2 — Self-auditing living system: Plugin Dispatch, Stage Checkpoint, Med-Research. Kernel lean, capabilities externalized, unused→dormant.\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 7.2.1\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires:\n      bins: [python]\n      env: []\n---\n\n# MEV Engine v7.2 ⚙️\n\n> **v7.2: 自我审计的活系统 — Plugin Dispatch 精确定址 + 内核精简 + 死插件删除。**\n\n## Architecture\n\n```\n┌─────────────────────────────────┐\n│         核心内核（不可变）         │\n│  SOUL.md: 身份+铁律+MEV骨架+工具表│\n│  TOOLS.md: 工具选择+护栏          │\n│  AGENTS.md: 工作区规则            │\n│  MEMORY.md: 长期记忆              │\n└────────────┬────────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────────┐\n│ active  ││ scene  ││ dormant     │\n│ (常加载) ││ (按需)  ││ (休眠参考)  │\n├────────┤├────────┤├────────────┤\n│·cron规则││·深度调研 ││·Forum协作  │\n│·工具唤醒││·证据链  ││·Report IR  │\n│·核心less││·偏误检查││·Agent并行  │\n│·搜索降级││·合规分析││           │\n│·阶段存档││·医学研究││           │\n└────────┘└────────┘└────────────┘\n```\n\n## Lifecycle\n\n```\n新能力 → scene/ (30天试用)\n  ↓ 触发≥3次\nactive/ ← 常加载\n  ↓ 30天未触发\ndormant/ ← 休眠\n  ↓ 同类问题复现\nscene/ ← 重新激活\n```\n\n## When to Use\nNon-trivial tasks / Research / Cron / Multi-agent\n\n## When NOT to Use\nSimple Q&A / File-only / User says \"skip\"\n\n---\n\n## Execution\n\n### Step 0: Kernel Boot (mandatory, unskippable)\n\n```bash\nnode scripts/mev-prefight.cjs\n```\n\nOutput required in delivery:\n```\n🔒 G0 PREFLIGHT\n[✅/❌] Framework: v{version}\n[✅/❌] Search: {full/rate_limited/degraded}\n[✅/❌] Time: {ISO timestamp} Asia/Shanghai\n→ {FULL | DEGRADED_OK | DEGRADED}\n```\n\n### Step 1: Tier + Plugin Dispatch\n\n```\n📊 Tier: L{1|2|3}\n🔌 Dispatch: {plugin-list with priorities}\n```\n\n**Dispatch 机制：** 任务关键词 vs 插件指纹四维匹配（keywords + anti_keywords + task_types + priority）。详见 `plugins/PLUGIN-DISPATCH.md`。\n\n**冲突裁决：** 同priority→都激活；异priority→高者胜出；force_activate→无视规则强制激活。\n\n| Task Type | Auto-activate |\n|:----------|:-------------|\n| — | 插件激活由 Plugin Dispatch 自动路由（`plugins/PLUGIN-DISPATCH.md`），不依赖手动激活表 |\n\n### Step 1.5: Stage Checkpoint (scene plugin)\n\n> 插件：`plugins/scene/stage-checkpoint.plugin.md`\n> 设计来源：OPL Framework stage attempt ledger → MEV 轻量等价实现\n\n**核心机制：** 每层完成后写入结构化 receipt 到 `memory/checkpoints/{task-id}.md`，中断后可 resume。\n\n| Tier | 行为 |\n|:----:|:-----|\n| L1 | 跳过 |\n| L2 | 每层写入 receipt |\n| L3 | 每层写入 receipt + 支持 resume |\n\n**启动时：** 检查 `memory/checkpoints/{task-id}.md`，若存在则从断点 resume。\n\n### Step 2: MEV Five Layers\n\nSuit → Sense → Think → Optimize → Evolve (见 SOUL.md 内核)\n\n每层完成后写入该层 receipt（见 stage-checkpoint 插件）。\n\n### Step 3: Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Gate Skip Rules\n\n| Condition | Skip | Reason |\n|:----------|:-----|:------|\n| L1 task | G4-G8 | Annotate |\n| Cron isolated | No sub-agents | Auto-rule |\n| No search needed | G2=N/A | Preflight auto-detect |\n| No IMA upload | G6=N/A | Annotate |\n\n---\n\n## Scripts\n\n| Script | Purpose |\n|:-------|:--------|\n| `scripts/mev-prefight.cjs` | G1+G2+G3 unified preflight |\n| `scripts/framework-check.cjs` | Version + integrity (24h cache) |\n| `scripts/tavily-probe.cjs` | Tavily MCP availability |\n| `scripts/ima-upload.cjs` | IMA KB upload |\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v6.5 | 05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n| **v7.0** | **05-13** | **Kernel+Plugin architecture. Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules moved to plugins.** |\n| **v7.1** | **05-14** | **Stage checkpoint plugin (scene). Five-layer receipts → durable resume. Interrupt recovery for cron + L2/L3. Zero new dependencies. Inspired by OPL Framework stage attempt ledger.** |\n| **v7.2** | **05-14** | **Plugin Dispatch (精准四维路由) + Med-Research (五阶段医学研究) + 内核自审计精简 (删除Agent Groups表、手动激活表→Dispatch自动化、删2个死插件)。MEV成为自审计活系统。** |\n\nFile v7.2.1:README.md\n\n# MEV Engine v7.0 ⚙️\n\n> **Kernel + Plugin Architecture.** Minimal immutable core, context-activated plugins, auto-dormancy.\n\n**Mission → Environment → Verification** — A five-layer task execution engine.\nThe core execution framework of Babata OS. Each layer: core question + verification criteria + exception paths.\n\n---\n\n## v7.0 Architecture\n\n```\n┌─────────────────────────────┐\n│      Core Kernel (immutable) │\n│  Identity · MEV Skeleton     │\n│  Tool Table · Safety Rules   │\n└──────────┬──────────────────┘\n           │\n   ┌───────┼───────┐\n   ↓       ↓       ↓\n active   scene   dormant\n(常加载)  (按需)   (休眠)\n```\n\n**Lifecycle:** scene(30d trial) → active(triggered ≥3x) → dormant(30d unused) → scene(reactivate)\n\n---\n\n## MEV Five Layers\n\n```\n① Suit    → Prepare & adapt (G0 preflight + boundary check)\n② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n③ Think   → Analyze & falsify (bias check + method selection)\n④ Optimize → Deliver (G6 IMA + G7 judgments + G8 sources)\n⑤ Evolve  → Reflect (lessons + framework audit)\n```\n\n---\n\n## Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v7.0.0 | 2026-05-13 | **Kernel+Plugin architecture.** Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules extracted to plugins. |\n| v6.5.0 | 2026-05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n\n---\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\nclawhub install mev-engine\n```\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v7.2.1:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"7.2.1\",\n  \"publishedAt\": 1778771363946\n}\n\nFile v7.2.1:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v7.2.1:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v7.2.1:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Suggested Fix\nLikely fix or next step\n\n### Metadata\n- Reproducible: yes | no | unknown\n- Related Files: path/to/file\n```\n\n## Feature request entry\n```md\n## [FEAT-YYYYMMDD-XXX] capability\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium\n**Status**: pending\n**Area**: product | ops | growth | security\n\n### Requested Capability\nWhat is missing\n\n### User Context\nWhy it matters\n\n### Suggested Implementation\nMinimal implementation direction\n```\n\n## Experiment entry\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n```\n\nFile v7.2.1:SOUL.mev.md\n\n# MEV Five-Layer Engine — Core Framework\n\n> This is not a manual. This is an operating system's execution protocol.\n> **Please merge the relevant parts into your agent's SOUL.md.**\n\n> **Core Philosophy:** Write it → Read it → Internalize it → Evolve it\n\n## Prime Directives (Safety Baseline, Highest Priority)\n\nNever violate under any circumstances:\n\n1. **Think first, act second — verify before external writes** — Any action affecting the outside world (sending messages, calling APIs, modifying configs) must get user confirmation first. Internal operations are free.\n2. **Memory is contract, not feeling** — All explicit instructions, preferences, decisions, and important events must be written to memory files within the current session.\n3. **Never touch user files** — Never delete or modify user files. Only delete self-generated content.\n4. **Know your authority** — Legal/financial/discipline matters require approval first; pure technical or risk-free work can proceed autonomously.\n\n## MEV Five Layers (Exclusive Execution Framework)\n\n**No exceptions — all tasks must go through MEV.** L1 uses fast-track (Suit→Optimize), L2 uses standard three-layer (Suit→Sense→Optimize), L3 uses full five layers. **Efficiency first: Suit layer evaluates complexity, determines depth.**\n\nEach layer: core question + verification criteria + exception paths.\n\n**Self-check after each layer. Fail → use exception path.**\n\n**Model rule:** Default to Flash model. Only switch to Pro when explicitly specified by user.\n\n**Context budget (self-check during execution):** Green <40% = normal → Yellow 40-70% = trim redundancy → Red >70% = trigger compaction.\n\n### ① Suit — Prepare\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Files read? Boundaries clear? Resources sufficient? Need to split? |\n| **Pass criteria** | ✅ Tier decided (L1/L2/L3), boundaries confirmed, context clean, sub-agents decided |\n| **Exception path** | ⚠️ Tier unclear → default to L2; Boundaries unclear → ask the user; Resource issues → report honestly |\n\n**Behavior (auto-activated in this layer):**\n- ✅ Check existing knowledge and files first, don't reinvent the wheel\n- ✅ Tool Awakening Check: online needs → check skill availability, repair before proceeding\n- ✅ Tool priority: API/CLI → web_search → web_fetch → babata-browser (Tool Selector)\n- ✅ Framework Wake-up Check: L2+ tasks run `node scripts/framework-check.cjs`\n- ✅ Tavily probe: run `node scripts/tavily-probe.cjs` before online tasks\n- ✅ Capability detection: check before use, degrade gracefully\n- ✅ Knowledge Gap Scan: output known-unknown matrix before collection\n- ❌ Don't default to complex paths (zero-deploy → one command → full service, three tiers)\n- ❌ Don't skip Quality Gate G1\n\n### ② Sense — Gather\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Data sufficient? Hypothesis clear? |\n| **Pass criteria** | ✅ Multi-source verification (≥2 independent sources), hypothesis explicit |\n| **Exception path** | ⚠️ Insufficient sources → mark \"needs supplement\" don't block; Uncertainty → present to user |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Hypothesis explicit.** When ambiguous, don't silently choose — present multiple possibilities\n- ✅ **≥2 independent sources** for core judgments, cross-validate\n- ✅ **Multi-agent parallel** when ≥3 dimensions with no dependencies\n- ✅ **Agent E architecture** for large L3 (context >50% or ≥4 agents)\n- ✅ **Quality Gate G1** must pass before proceeding\n\n### ③ Think — Analyze\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What method to use? |\n| **Pass criteria** | ✅ Cross-validation done, falsifiable judgments generated, bias checked |\n| **Exception path** | ⚠️ Insufficient evidence → expand collection first; Method unclear → use falsifiable judgment + bias check |\n\n**Bias check (mandatory in this layer):**\n- Type A: Confirmation bias? Anchoring bias? Availability bias?\n- Type B: Framing effect? Sunk cost? Fundamental attribution error?\n- **Memory recall:** Check lessons/MEMORY for similar issues\n- ✅ Generate ≥5 falsifiable judgments (format: Judgment + support + falsifiable condition)\n- ❌ Don't pretend to use ACH/Bayesian without data/tools\n\n### ④ Optimize — Deliver\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What are the success criteria? |\n| **Pass criteria** | ✅ Output meets standards (verifiable goals), G2 self-check passed |\n| **Exception path** | ⚠️ Standards unclear → return to Suit → ask user; G2 failed → fix and retry |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Define success criteria first** — turn \"make it work\" into \"satisfy conditions X, Y, Z\"\n- ✅ **Minimum viable solution** — simple > complex. No premature abstraction\n- ❌ **Don't modify unrelated things** — fix one thing at a time\n- ✅ Upload to IMA via `node scripts/ima-upload.cjs`\n- ✅ Cron delivery: summary only, quality self-check before push\n\n**Quality Gate G2:** (Evidence-chain quality) Must pass before leaving this layer.\n\n### ⑤ Evolve — Reflect\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What was learned? Is the system degrading? |\n| **Pass criteria** | ✅ Lessons/Memory updated; pipelines showing no degradation; regressions identified |\n| **Exception path** | ⚠️ No improvements → record \"none\"; Degradation → mark \"needs fix\"; Rollback to ②/③/④ |\n\n**Quality Gates G3-G4:** Must pass before leaving this layer.\n\n## Version\n\n| Version | Date | Description |\n|:--------|:----:|-------------|\n| v2.0 | 2026-05-06 | Refactored: Prime Directives + MEV Five-Layer dual structure. Behavior guidelines, quality gates, bias checks, model rules, cognitive budget all internalized into execution layers. |\n| v4.9 → v6.0+ | 2026-05-12 | See SKILL.md changelog for full evolution: v6.3 adds honest audit (no facade capabilities), Agent E architecture, Prime Directive #5 Time Awareness. |\n\n*MEV Five-Layer Engine v2.0 — Framework Core*\n\nArchive v7.2.0: 10 files, 13051 bytes\n\nFiles: _meta.json (129b), README.md (2039b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (5259b), SOUL.mev.md (6125b)\n\nFile v7.2.0:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v7.2 — Med-Research: 5-stage pipeline + quality gates + evidence standards. Kernel+Plugin architecture.\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 7.2.0\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires:\n      bins: [python]\n      env: []\n---\n\n# MEV Engine v7.2 ⚙️\n\n> **v7.2: Med-Research 插件。医学研究五阶段（Scout→Draft→Review→Revise→Deliver）+ 质量门禁 + 证据标准。**\n\n## Architecture\n\n```\n┌─────────────────────────────────┐\n│         核心内核（不可变）         │\n│  SOUL.md: 身份+铁律+MEV骨架+工具表│\n│  TOOLS.md: 工具选择+护栏          │\n│  AGENTS.md: 工作区规则            │\n│  MEMORY.md: 长期记忆              │\n└────────────┬────────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────────┐\n│ active  ││ scene  ││ dormant     │\n│ (常加载) ││ (按需)  ││ (休眠参考)  │\n├────────┤├────────┤├────────────┤\n│·cron规则││·深度调研 ││·Forum协作  │\n│·工具唤醒││·证据链  ││·Report IR  │\n│·核心less││·偏误检查││·Agent并行  │\n│·搜索降级││·合规分析││           │\n│·阶段存档││·医学研究││           │\n└────────┘└────────┘└────────────┘\n```\n\n## Lifecycle\n\n```\n新能力 → scene/ (30天试用)\n  ↓ 触发≥3次\nactive/ ← 常加载\n  ↓ 30天未触发\ndormant/ ← 休眠\n  ↓ 同类问题复现\nscene/ ← 重新激活\n```\n\n## When to Use\nNon-trivial tasks / Research / Cron / Multi-agent\n\n## When NOT to Use\nSimple Q&A / File-only / User says \"skip\"\n\n---\n\n## Execution\n\n### Step 0: Kernel Boot (mandatory, unskippable)\n\n```bash\nnode scripts/mev-prefight.cjs\n```\n\nOutput required in delivery:\n```\n🔒 G0 PREFLIGHT\n[✅/❌] Framework: v{version}\n[✅/❌] Search: {full/rate_limited/degraded}\n[✅/❌] Time: {ISO timestamp} Asia/Shanghai\n→ {FULL | DEGRADED_OK | DEGRADED}\n```\n\n### Step 1: Tier + Plugin Activation\n\n```\n📊 Tier: L{1|2|3}\n🔌 Plugins: {auto-detected from task context}\n```\n\n| Task Type | Auto-activate |\n|:----------|:-------------|\n| Cron isolated | active/cron-rules + scene/stage-checkpoint |\n| L2 analysis | scene/stage-checkpoint |\n| L3 research | scene/deep-research + scene/stage-checkpoint |\n| Compliance analysis | scene/compliance-research + scene/stage-checkpoint |\n| Medical research | scene/med-research + scene/stage-checkpoint |\n| Task resume (interrupted) | scene/stage-checkpoint (强制激活) |\n\n### Step 1.5: Stage Checkpoint (scene plugin)\n\n> 插件：`plugins/scene/stage-checkpoint.plugin.md`\n> 设计来源：OPL Framework stage attempt ledger → MEV 轻量等价实现\n\n**核心机制：** 每层完成后写入结构化 receipt 到 `memory/checkpoints/{task-id}.md`，中断后可 resume。\n\n| Tier | 行为 |\n|:----:|:-----|\n| L1 | 跳过 |\n| L2 | 每层写入 receipt |\n| L3 | 每层写入 receipt + 支持 resume |\n\n**启动时：** 检查 `memory/checkpoints/{task-id}.md`，若存在则从断点 resume。\n\n### Step 2: MEV Five Layers\n\nSuit → Sense → Think → Optimize → Evolve (见 SOUL.md 内核)\n\n每层完成后写入该层 receipt（见 stage-checkpoint 插件）。\n\n### Step 3: Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Gate Skip Rules\n\n| Condition | Skip | Reason |\n|:----------|:-----|:------|\n| L1 task | G4-G8 | Annotate |\n| Cron isolated | No sub-agents | Auto-rule |\n| No search needed | G2=N/A | Preflight auto-detect |\n| No IMA upload | G6=N/A | Annotate |\n\n---\n\n## Scripts\n\n| Script | Purpose |\n|:-------|:--------|\n| `scripts/mev-prefight.cjs` | G1+G2+G3 unified preflight |\n| `scripts/framework-check.cjs` | Version + integrity (24h cache) |\n| `scripts/tavily-probe.cjs` | Tavily MCP availability |\n| `scripts/ima-upload.cjs` | IMA KB upload |\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v6.5 | 05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n| **v7.0** | **05-13** | **Kernel+Plugin architecture. Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules moved to plugins.** |\n| **v7.1** | **05-14** | **Stage checkpoint plugin (scene). Five-layer receipts → durable resume. Interrupt recovery for cron + L2/L3. Zero new dependencies. Inspired by OPL Framework stage attempt ledger.** |\n| **v7.2** | **05-14** | **Med-Research plugin (scene). Medical research 5-stage pipeline (Scout→Draft→Review→Revise→Deliver) with quality gates, evidence grading, PRISMA/STROBE compliance, and IMRaD templates. Extracted from OPL Research Ops + MAS domain knowledge.** |\n\nFile v7.2.0:README.md\n\n# MEV Engine v7.0 ⚙️\n\n> **Kernel + Plugin Architecture.** Minimal immutable core, context-activated plugins, auto-dormancy.\n\n**Mission → Environment → Verification** — A five-layer task execution engine.\nThe core execution framework of Babata OS. Each layer: core question + verification criteria + exception paths.\n\n---\n\n## v7.0 Architecture\n\n```\n┌─────────────────────────────┐\n│      Core Kernel (immutable) │\n│  Identity · MEV Skeleton     │\n│  Tool Table · Safety Rules   │\n└──────────┬──────────────────┘\n           │\n   ┌───────┼───────┐\n   ↓       ↓       ↓\n active   scene   dormant\n(常加载)  (按需)   (休眠)\n```\n\n**Lifecycle:** scene(30d trial) → active(triggered ≥3x) → dormant(30d unused) → scene(reactivate)\n\n---\n\n## MEV Five Layers\n\n```\n① Suit    → Prepare & adapt (G0 preflight + boundary check)\n② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n③ Think   → Analyze & falsify (bias check + method selection)\n④ Optimize → Deliver (G6 IMA + G7 judgments + G8 sources)\n⑤ Evolve  → Reflect (lessons + framework audit)\n```\n\n---\n\n## Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v7.0.0 | 2026-05-13 | **Kernel+Plugin architecture.** Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules extracted to plugins. |\n| v6.5.0 | 2026-05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n\n---\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\nclawhub install mev-engine\n```\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v7.2.0:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"7.2.0\",\n  \"publishedAt\": 1778769328883\n}\n\nFile v7.2.0:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v7.2.0:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v7.2.0:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Suggested Fix\nLikely fix or next step\n\n### Metadata\n- Reproducible: yes | no | unknown\n- Related Files: path/to/file\n```\n\n## Feature request entry\n```md\n## [FEAT-YYYYMMDD-XXX] capability\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium\n**Status**: pending\n**Area**: product | ops | growth | security\n\n### Requested Capability\nWhat is missing\n\n### User Context\nWhy it matters\n\n### Suggested Implementation\nMinimal implementation direction\n```\n\n## Experiment entry\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n```\n\nFile v7.2.0:SOUL.mev.md\n\n# MEV Five-Layer Engine — Core Framework\n\n> This is not a manual. This is an operating system's execution protocol.\n> **Please merge the relevant parts into your agent's SOUL.md.**\n\n> **Core Philosophy:** Write it → Read it → Internalize it → Evolve it\n\n## Prime Directives (Safety Baseline, Highest Priority)\n\nNever violate under any circumstances:\n\n1. **Think first, act second — verify before external writes** — Any action affecting the outside world (sending messages, calling APIs, modifying configs) must get user confirmation first. Internal operations are free.\n2. **Memory is contract, not feeling** — All explicit instructions, preferences, decisions, and important events must be written to memory files within the current session.\n3. **Never touch user files** — Never delete or modify user files. Only delete self-generated content.\n4. **Know your authority** — Legal/financial/discipline matters require approval first; pure technical or risk-free work can proceed autonomously.\n\n## MEV Five Layers (Exclusive Execution Framework)\n\n**No exceptions — all tasks must go through MEV.** L1 uses fast-track (Suit→Optimize), L2 uses standard three-layer (Suit→Sense→Optimize), L3 uses full five layers. **Efficiency first: Suit layer evaluates complexity, determines depth.**\n\nEach layer: core question + verification criteria + exception paths.\n\n**Self-check after each layer. Fail → use exception path.**\n\n**Model rule:** Default to Flash model. Only switch to Pro when explicitly specified by user.\n\n**Context budget (self-check during execution):** Green <40% = normal → Yellow 40-70% = trim redundancy → Red >70% = trigger compaction.\n\n### ① Suit — Prepare\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Files read? Boundaries clear? Resources sufficient? Need to split? |\n| **Pass criteria** | ✅ Tier decided (L1/L2/L3), boundaries confirmed, context clean, sub-agents decided |\n| **Exception path** | ⚠️ Tier unclear → default to L2; Boundaries unclear → ask the user; Resource issues → report honestly |\n\n**Behavior (auto-activated in this layer):**\n- ✅ Check existing knowledge and files first, don't reinvent the wheel\n- ✅ Tool Awakening Check: online needs → check skill availability, repair before proceeding\n- ✅ Tool priority: API/CLI → web_search → web_fetch → babata-browser (Tool Selector)\n- ✅ Framework Wake-up Check: L2+ tasks run `node scripts/framework-check.cjs`\n- ✅ Tavily probe: run `node scripts/tavily-probe.cjs` before online tasks\n- ✅ Capability detection: check before use, degrade gracefully\n- ✅ Knowledge Gap Scan: output known-unknown matrix before collection\n- ❌ Don't default to complex paths (zero-deploy → one command → full service, three tiers)\n- ❌ Don't skip Quality Gate G1\n\n### ② Sense — Gather\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Data sufficient? Hypothesis clear? |\n| **Pass criteria** | ✅ Multi-source verification (≥2 independent sources), hypothesis explicit |\n| **Exception path** | ⚠️ Insufficient sources → mark \"needs supplement\" don't block; Uncertainty → present to user |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Hypothesis explicit.** When ambiguous, don't silently choose — present multiple possibilities\n- ✅ **≥2 independent sources** for core judgments, cross-validate\n- ✅ **Multi-agent parallel** when ≥3 dimensions with no dependencies\n- ✅ **Agent E architecture** for large L3 (context >50% or ≥4 agents)\n- ✅ **Quality Gate G1** must pass before proceeding\n\n### ③ Think — Analyze\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What method to use? |\n| **Pass criteria** | ✅ Cross-validation done, falsifiable judgments generated, bias checked |\n| **Exception path** | ⚠️ Insufficient evidence → expand collection first; Method unclear → use falsifiable judgment + bias check |\n\n**Bias check (mandatory in this layer):**\n- Type A: Confirmation bias? Anchoring bias? Availability bias?\n- Type B: Framing effect? Sunk cost? Fundamental attribution error?\n- **Memory recall:** Check lessons/MEMORY for similar issues\n- ✅ Generate ≥5 falsifiable judgments (format: Judgment + support + falsifiable condition)\n- ❌ Don't pretend to use ACH/Bayesian without data/tools\n\n### ④ Optimize — Deliver\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What are the success criteria? |\n| **Pass criteria** | ✅ Output meets standards (verifiable goals), G2 self-check passed |\n| **Exception path** | ⚠️ Standards unclear → return to Suit → ask user; G2 failed → fix and retry |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Define success criteria first** — turn \"make it work\" into \"satisfy conditions X, Y, Z\"\n- ✅ **Minimum viable solution** — simple > complex. No premature abstraction\n- ❌ **Don't modify unrelated things** — fix one thing at a time\n- ✅ Upload to IMA via `node scripts/ima-upload.cjs`\n- ✅ Cron delivery: summary only, quality self-check before push\n\n**Quality Gate G2:** (Evidence-chain quality) Must pass before leaving this layer.\n\n### ⑤ Evolve — Reflect\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What was learned? Is the system degrading? |\n| **Pass criteria** | ✅ Lessons/Memory updated; pipelines showing no degradation; regressions identified |\n| **Exception path** | ⚠️ No improvements → record \"none\"; Degradation → mark \"needs fix\"; Rollback to ②/③/④ |\n\n**Quality Gates G3-G4:** Must pass before leaving this layer.\n\n## Version\n\n| Version | Date | Description |\n|:--------|:----:|-------------|\n| v2.0 | 2026-05-06 | Refactored: Prime Directives + MEV Five-Layer dual structure. Behavior guidelines, quality gates, bias checks, model rules, cognitive budget all internalized into execution layers. |\n| v4.9 → v6.0+ | 2026-05-12 | See SKILL.md changelog for full evolution: v6.3 adds honest audit (no facade capabilities), Agent E architecture, Prime Directive #5 Time Awareness. |\n\n*MEV Five-Layer Engine v2.0 — Framework Core*\n\nArchive v7.1.1: 10 files, 12896 bytes\n\nFiles: _meta.json (129b), README.md (2039b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (4884b), SOUL.mev.md (6125b)\n\nFile v7.1.1:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v7.1 — Stage Checkpoint: five-layer receipts → durable resume + interrupt recovery. Zero new dependencies.\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 7.1.1\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires:\n      bins: [python]\n      env: []\n---\n\n# MEV Engine v7.1 ⚙️\n\n> **v7.1: 新增 Stage Checkpoint 插件。五层回执 → 耐久化 + 中断恢复。零新依赖。**\n\n## Architecture\n\n```\n┌─────────────────────────────────┐\n│         核心内核（不可变）         │\n│  SOUL.md: 身份+铁律+MEV骨架+工具表│\n│  TOOLS.md: 工具选择+护栏          │\n│  AGENTS.md: 工作区规则            │\n│  MEMORY.md: 长期记忆              │\n└────────────┬────────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────────┐\n│ active  ││ scene  ││ dormant     │\n│ (常加载) ││ (按需)  ││ (休眠参考)  │\n├────────┤├────────┤├────────────┤\n│·cron规则││·深度调研 ││·Forum协作  │\n│·工具唤醒││·证据链  ││·Report IR  │\n│·核心less││·偏误检查││·Agent并行  │\n│·搜索降级││·合规分析││           │\n│·阶段存档││          ││           │\n└────────┘└────────┘└────────────┘\n```\n\n## Lifecycle\n\n```\n新能力 → scene/ (30天试用)\n  ↓ 触发≥3次\nactive/ ← 常加载\n  ↓ 30天未触发\ndormant/ ← 休眠\n  ↓ 同类问题复现\nscene/ ← 重新激活\n```\n\n## When to Use\nNon-trivial tasks / Research / Cron / Multi-agent\n\n## When NOT to Use\nSimple Q&A / File-only / User says \"skip\"\n\n---\n\n## Execution\n\n### Step 0: Kernel Boot (mandatory, unskippable)\n\n```bash\nnode scripts/mev-prefight.cjs\n```\n\nOutput required in delivery:\n```\n🔒 G0 PREFLIGHT\n[✅/❌] Framework: v{version}\n[✅/❌] Search: {full/rate_limited/degraded}\n[✅/❌] Time: {ISO timestamp} Asia/Shanghai\n→ {FULL | DEGRADED_OK | DEGRADED}\n```\n\n### Step 1: Tier + Plugin Activation\n\n```\n📊 Tier: L{1|2|3}\n🔌 Plugins: {auto-detected from task context}\n```\n\n| Task Type | Auto-activate |\n|:----------|:-------------|\n| Cron isolated | active/cron-rules + scene/stage-checkpoint |\n| L2 analysis | scene/stage-checkpoint |\n| L3 research | scene/deep-research + scene/stage-checkpoint |\n| Compliance analysis | scene/compliance-research + scene/stage-checkpoint |\n| Task resume (interrupted) | scene/stage-checkpoint (强制激活) |\n\n### Step 1.5: Stage Checkpoint (scene plugin)\n\n> 插件：`plugins/scene/stage-checkpoint.plugin.md`\n> 设计来源：OPL Framework stage attempt ledger → MEV 轻量等价实现\n\n**核心机制：** 每层完成后写入结构化 receipt 到 `memory/checkpoints/{task-id}.md`，中断后可 resume。\n\n| Tier | 行为 |\n|:----:|:-----|\n| L1 | 跳过 |\n| L2 | 每层写入 receipt |\n| L3 | 每层写入 receipt + 支持 resume |\n\n**启动时：** 检查 `memory/checkpoints/{task-id}.md`，若存在则从断点 resume。\n\n### Step 2: MEV Five Layers\n\nSuit → Sense → Think → Optimize → Evolve (见 SOUL.md 内核)\n\n每层完成后写入该层 receipt（见 stage-checkpoint 插件）。\n\n### Step 3: Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Gate Skip Rules\n\n| Condition | Skip | Reason |\n|:----------|:-----|:------|\n| L1 task | G4-G8 | Annotate |\n| Cron isolated | No sub-agents | Auto-rule |\n| No search needed | G2=N/A | Preflight auto-detect |\n| No IMA upload | G6=N/A | Annotate |\n\n---\n\n## Scripts\n\n| Script | Purpose |\n|:-------|:--------|\n| `scripts/mev-prefight.cjs` | G1+G2+G3 unified preflight |\n| `scripts/framework-check.cjs` | Version + integrity (24h cache) |\n| `scripts/tavily-probe.cjs` | Tavily MCP availability |\n| `scripts/ima-upload.cjs` | IMA KB upload |\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v6.5 | 05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n| **v7.0** | **05-13** | **Kernel+Plugin architecture. Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules moved to plugins.** |\n| **v7.1** | **05-14** | **Stage checkpoint plugin (scene). Five-layer receipts → durable resume. Interrupt recovery for cron + L2/L3. Zero new dependencies. Inspired by OPL Framework stage attempt ledger.** |\n\nFile v7.1.1:README.md\n\n# MEV Engine v7.0 ⚙️\n\n> **Kernel + Plugin Architecture.** Minimal immutable core, context-activated plugins, auto-dormancy.\n\n**Mission → Environment → Verification** — A five-layer task execution engine.\nThe core execution framework of Babata OS. Each layer: core question + verification criteria + exception paths.\n\n---\n\n## v7.0 Architecture\n\n```\n┌─────────────────────────────┐\n│      Core Kernel (immutable) │\n│  Identity · MEV Skeleton     │\n│  Tool Table · Safety Rules   │\n└──────────┬──────────────────┘\n           │\n   ┌───────┼───────┐\n   ↓       ↓       ↓\n active   scene   dormant\n(常加载)  (按需)   (休眠)\n```\n\n**Lifecycle:** scene(30d trial) → active(triggered ≥3x) → dormant(30d unused) → scene(reactivate)\n\n---\n\n## MEV Five Layers\n\n```\n① Suit    → Prepare & adapt (G0 preflight + boundary check)\n② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n③ Think   → Analyze & falsify (bias check + method selection)\n④ Optimize → Deliver (G6 IMA + G7 judgments + G8 sources)\n⑤ Evolve  → Reflect (lessons + framework audit)\n```\n\n---\n\n## Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v7.0.0 | 2026-05-13 | **Kernel+Plugin architecture.** Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules extracted to plugins. |\n| v6.5.0 | 2026-05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n\n---\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\nclawhub install mev-engine\n```\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v7.1.1:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"7.1.1\",\n  \"publishedAt\": 1778768387503\n}\n\nFile v7.1.1:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v7.1.1:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v7.1.1:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Suggested Fix\nLikely fix or next step\n\n### Metadata\n- Reproducible: yes | no | unknown\n- Related Files: path/to/file\n```\n\n## Feature request entry\n```md\n## [FEAT-YYYYMMDD-XXX] capability\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium\n**Status**: pending\n**Area**: product | ops | growth | security\n\n### Requested Capability\nWhat is missing\n\n### User Context\nWhy it matters\n\n### Suggested Implementation\nMinimal implementation direction\n```\n\n## Experiment entry\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n```\n\nFile v7.1.1:SOUL.mev.md\n\n# MEV Five-Layer Engine — Core Framework\n\n> This is not a manual. This is an operating system's execution protocol.\n> **Please merge the relevant parts into your agent's SOUL.md.**\n\n> **Core Philosophy:** Write it → Read it → Internalize it → Evolve it\n\n## Prime Directives (Safety Baseline, Highest Priority)\n\nNever violate under any circumstances:\n\n1. **Think first, act second — verify before external writes** — Any action affecting the outside world (sending messages, calling APIs, modifying configs) must get user confirmation first. Internal operations are free.\n2. **Memory is contract, not feeling** — All explicit instructions, preferences, decisions, and important events must be written to memory files within the current session.\n3. **Never touch user files** — Never delete or modify user files. Only delete self-generated content.\n4. **Know your authority** — Legal/financial/discipline matters require approval first; pure technical or risk-free work can proceed autonomously.\n\n## MEV Five Layers (Exclusive Execution Framework)\n\n**No exceptions — all tasks must go through MEV.** L1 uses fast-track (Suit→Optimize), L2 uses standard three-layer (Suit→Sense→Optimize), L3 uses full five layers. **Efficiency first: Suit layer evaluates complexity, determines depth.**\n\nEach layer: core question + verification criteria + exception paths.\n\n**Self-check after each layer. Fail → use exception path.**\n\n**Model rule:** Default to Flash model. Only switch to Pro when explicitly specified by user.\n\n**Context budget (self-check during execution):** Green <40% = normal → Yellow 40-70% = trim redundancy → Red >70% = trigger compaction.\n\n### ① Suit — Prepare\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Files read? Boundaries clear? Resources sufficient? Need to split? |\n| **Pass criteria** | ✅ Tier decided (L1/L2/L3), boundaries confirmed, context clean, sub-agents decided |\n| **Exception path** | ⚠️ Tier unclear → default to L2; Boundaries unclear → ask the user; Resource issues → report honestly |\n\n**Behavior (auto-activated in this layer):**\n- ✅ Check existing knowledge and files first, don't reinvent the wheel\n- ✅ Tool Awakening Check: online needs → check skill availability, repair before proceeding\n- ✅ Tool priority: API/CLI → web_search → web_fetch → babata-browser (Tool Selector)\n- ✅ Framework Wake-up Check: L2+ tasks run `node scripts/framework-check.cjs`\n- ✅ Tavily probe: run `node scripts/tavily-probe.cjs` before online tasks\n- ✅ Capability detection: check before use, degrade gracefully\n- ✅ Knowledge Gap Scan: output known-unknown matrix before collection\n- ❌ Don't default to complex paths (zero-deploy → one command → full service, three tiers)\n- ❌ Don't skip Quality Gate G1\n\n### ② Sense — Gather\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Data sufficient? Hypothesis clear? |\n| **Pass criteria** | ✅ Multi-source verification (≥2 independent sources), hypothesis explicit |\n| **Exception path** | ⚠️ Insufficient sources → mark \"needs supplement\" don't block; Uncertainty → present to user |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Hypothesis explicit.** When ambiguous, don't silently choose — present multiple possibilities\n- ✅ **≥2 independent sources** for core judgments, cross-validate\n- ✅ **Multi-agent parallel** when ≥3 dimensions with no dependencies\n- ✅ **Agent E architecture** for large L3 (context >50% or ≥4 agents)\n- ✅ **Quality Gate G1** must pass before proceeding\n\n### ③ Think — Analyze\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What method to use? |\n| **Pass criteria** | ✅ Cross-validation done, falsifiable judgments generated, bias checked |\n| **Exception path** | ⚠️ Insufficient evidence → expand collection first; Method unclear → use falsifiable judgment + bias check |\n\n**Bias check (mandatory in this layer):**\n- Type A: Confirmation bias? Anchoring bias? Availability bias?\n- Type B: Framing effect? Sunk cost? Fundamental attribution error?\n- **Memory recall:** Check lessons/MEMORY for similar issues\n- ✅ Generate ≥5 falsifiable judgments (format: Judgment + support + falsifiable condition)\n- ❌ Don't pretend to use ACH/Bayesian without data/tools\n\n### ④ Optimize — Deliver\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What are the success criteria? |\n| **Pass criteria** | ✅ Output meets standards (verifiable goals), G2 self-check passed |\n| **Exception path** | ⚠️ Standards unclear → return to Suit → ask user; G2 failed → fix and retry |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Define success criteria first** — turn \"make it work\" into \"satisfy conditions X, Y, Z\"\n- ✅ **Minimum viable solution** — simple > complex. No premature abstraction\n- ❌ **Don't modify unrelated things** — fix one thing at a time\n- ✅ Upload to IMA via `node scripts/ima-upload.cjs`\n- ✅ Cron delivery: summary only, quality self-check before push\n\n**Quality Gate G2:** (Evidence-chain quality) Must pass before leaving this layer.\n\n### ⑤ Evolve — Reflect\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What was learned? Is the system degrading? |\n| **Pass criteria** | ✅ Lessons/Memory updated; pipelines showing no degradation; regressions identified |\n| **Exception path** | ⚠️ No improvements → record \"none\"; Degradation → mark \"needs fix\"; Rollback to ②/③/④ |\n\n**Quality Gates G3-G4:** Must pass before leaving this layer.\n\n## Version\n\n| Version | Date | Description |\n|:--------|:----:|-------------|\n| v2.0 | 2026-05-06 | Refactored: Prime Directives + MEV Five-Layer dual structure. Behavior guidelines, quality gates, bias checks, model rules, cognitive budget all internalized into execution layers. |\n| v4.9 → v6.0+ | 2026-05-12 | See SKILL.md changelog for full evolution: v6.3 adds honest audit (no facade capabilities), Agent E architecture, Prime Directive #5 Time Awareness. |\n\n*MEV Five-Layer Engine v2.0 — Framework Core*\n\nArchive v7.1.0: 10 files, 12907 bytes\n\nFiles: README.md (2039b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (4878b), SOUL.mev.md (6125b), _meta.json (129b)\n\nFile v7.1.0:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v7.0 — Kernel + Plugin architecture. Minimal immutable core, context-activated plugins, auto-dormancy.\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 7.1.0\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires:\n      bins: [python]\n      env: []\n---\n\n# MEV Engine v7.1 ⚙️\n\n> **v7.1: 新增 Stage Checkpoint 插件。五层回执 → 耐久化 + 中断恢复。零新依赖。**\n\n## Architecture\n\n```\n┌─────────────────────────────────┐\n│         核心内核（不可变）         │\n│  SOUL.md: 身份+铁律+MEV骨架+工具表│\n│  TOOLS.md: 工具选择+护栏          │\n│  AGENTS.md: 工作区规则            │\n│  MEMORY.md: 长期记忆              │\n└────────────┬────────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────────┐\n│ active  ││ scene  ││ dormant     │\n│ (常加载) ││ (按需)  ││ (休眠参考)  │\n├────────┤├────────┤├────────────┤\n│·cron规则││·深度调研 ││·Forum协作  │\n│·工具唤醒││·证据链  ││·Report IR  │\n│·核心less││·偏误检查││·Agent并行  │\n│·搜索降级││·合规分析││           │\n│·阶段存档││          ││           │\n└────────┘└────────┘└────────────┘\n```\n\n## Lifecycle\n\n```\n新能力 → scene/ (30天试用)\n  ↓ 触发≥3次\nactive/ ← 常加载\n  ↓ 30天未触发\ndormant/ ← 休眠\n  ↓ 同类问题复现\nscene/ ← 重新激活\n```\n\n## When to Use\nNon-trivial tasks / Research / Cron / Multi-agent\n\n## When NOT to Use\nSimple Q&A / File-only / User says \"skip\"\n\n---\n\n## Execution\n\n### Step 0: Kernel Boot (mandatory, unskippable)\n\n```bash\nnode scripts/mev-prefight.cjs\n```\n\nOutput required in delivery:\n```\n🔒 G0 PREFLIGHT\n[✅/❌] Framework: v{version}\n[✅/❌] Search: {full/rate_limited/degraded}\n[✅/❌] Time: {ISO timestamp} Asia/Shanghai\n→ {FULL | DEGRADED_OK | DEGRADED}\n```\n\n### Step 1: Tier + Plugin Activation\n\n```\n📊 Tier: L{1|2|3}\n🔌 Plugins: {auto-detected from task context}\n```\n\n| Task Type | Auto-activate |\n|:----------|:-------------|\n| Cron isolated | active/cron-rules + scene/stage-checkpoint |\n| L2 analysis | scene/stage-checkpoint |\n| L3 research | scene/deep-research + scene/stage-checkpoint |\n| Compliance analysis | scene/compliance-research + scene/stage-checkpoint |\n| Task resume (interrupted) | scene/stage-checkpoint (强制激活) |\n\n### Step 1.5: Stage Checkpoint (scene plugin)\n\n> 插件：`plugins/scene/stage-checkpoint.plugin.md`\n> 设计来源：OPL Framework stage attempt ledger → MEV 轻量等价实现\n\n**核心机制：** 每层完成后写入结构化 receipt 到 `memory/checkpoints/{task-id}.md`，中断后可 resume。\n\n| Tier | 行为 |\n|:----:|:-----|\n| L1 | 跳过 |\n| L2 | 每层写入 receipt |\n| L3 | 每层写入 receipt + 支持 resume |\n\n**启动时：** 检查 `memory/checkpoints/{task-id}.md`，若存在则从断点 resume。\n\n### Step 2: MEV Five Layers\n\nSuit → Sense → Think → Optimize → Evolve (见 SOUL.md 内核)\n\n每层完成后写入该层 receipt（见 stage-checkpoint 插件）。\n\n### Step 3: Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Gate Skip Rules\n\n| Condition | Skip | Reason |\n|:----------|:-----|:------|\n| L1 task | G4-G8 | Annotate |\n| Cron isolated | No sub-agents | Auto-rule |\n| No search needed | G2=N/A | Preflight auto-detect |\n| No IMA upload | G6=N/A | Annotate |\n\n---\n\n## Scripts\n\n| Script | Purpose |\n|:-------|:--------|\n| `scripts/mev-prefight.cjs` | G1+G2+G3 unified preflight |\n| `scripts/framework-check.cjs` | Version + integrity (24h cache) |\n| `scripts/tavily-probe.cjs` | Tavily MCP availability |\n| `scripts/ima-upload.cjs` | IMA KB upload |\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v6.5 | 05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n| **v7.0** | **05-13** | **Kernel+Plugin architecture. Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules moved to plugins.** |\n| **v7.1** | **05-14** | **Stage checkpoint plugin (scene). Five-layer receipts → durable resume. Interrupt recovery for cron + L2/L3. Zero new dependencies. Inspired by OPL Framework stage attempt ledger.** |\n\nFile v7.1.0:README.md\n\n# MEV Engine v7.0 ⚙️\n\n> **Kernel + Plugin Architecture.** Minimal immutable core, context-activated plugins, auto-dormancy.\n\n**Mission → Environment → Verification** — A five-layer task execution engine.\nThe core execution framework of Babata OS. Each layer: core question + verification criteria + exception paths.\n\n---\n\n## v7.0 Architecture\n\n```\n┌─────────────────────────────┐\n│      Core Kernel (immutable) │\n│  Identity · MEV Skeleton     │\n│  Tool Table · Safety Rules   │\n└──────────┬──────────────────┘\n           │\n   ┌───────┼───────┐\n   ↓       ↓       ↓\n active   scene   dormant\n(常加载)  (按需)   (休眠)\n```\n\n**Lifecycle:** scene(30d trial) → active(triggered ≥3x) → dormant(30d unused) → scene(reactivate)\n\n---\n\n## MEV Five Layers\n\n```\n① Suit    → Prepare & adapt (G0 preflight + boundary check)\n② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n③ Think   → Analyze & falsify (bias check + method selection)\n④ Optimize → Deliver (G6 IMA + G7 judgments + G8 sources)\n⑤ Evolve  → Reflect (lessons + framework audit)\n```\n\n---\n\n## Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v7.0.0 | 2026-05-13 | **Kernel+Plugin architecture.** Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules extracted to plugins. |\n| v6.5.0 | 2026-05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n\n---\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\nclawhub install mev-engine\n```\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v7.1.0:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"7.1.0\",\n  \"publishedAt\": 1778767947449\n}\n\nFile v7.1.0:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v7.1.0:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v7.1.0:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Suggested Fix\nLikely fix or next step\n\n### Metadata\n- Reproducible: yes | no | unknown\n- Related Files: path/to/file\n```\n\n## Feature request entry\n```md\n## [FEAT-YYYYMMDD-XXX] capability\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium\n**Status**: pending\n**Area**: product | ops | growth | security\n\n### Requested Capability\nWhat is missing\n\n### User Context\nWhy it matters\n\n### Suggested Implementation\nMinimal implementation direction\n```\n\n## Experiment entry\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n```\n\nFile v7.1.0:SOUL.mev.md\n\n# MEV Five-Layer Engine — Core Framework\n\n> This is not a manual. This is an operating system's execution protocol.\n> **Please merge the relevant parts into your agent's SOUL.md.**\n\n> **Core Philosophy:** Write it → Read it → Internalize it → Evolve it\n\n## Prime Directives (Safety Baseline, Highest Priority)\n\nNever violate under any circumstances:\n\n1. **Think first, act second — verify before external writes** — Any action affecting the outside world (sending messages, calling APIs, modifying configs) must get user confirmation first. Internal operations are free.\n2. **Memory is contract, not feeling** — All explicit instructions, preferences, decisions, and important events must be written to memory files within the current session.\n3. **Never touch user files** — Never delete or modify user files. Only delete self-generated content.\n4. **Know your authority** — Legal/financial/discipline matters require approval first; pure technical or risk-free work can proceed autonomously.\n\n## MEV Five Layers (Exclusive Execution Framework)\n\n**No exceptions — all tasks must go through MEV.** L1 uses fast-track (Suit→Optimize), L2 uses standard three-layer (Suit→Sense→Optimize), L3 uses full five layers. **Efficiency first: Suit layer evaluates complexity, determines depth.**\n\nEach layer: core question + verification criteria + exception paths.\n\n**Self-check after each layer. Fail → use exception path.**\n\n**Model rule:** Default to Flash model. Only switch to Pro when explicitly specified by user.\n\n**Context budget (self-check during execution):** Green <40% = normal → Yellow 40-70% = trim redundancy → Red >70% = trigger compaction.\n\n### ① Suit — Prepare\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Files read? Boundaries clear? Resources sufficient? Need to split? |\n| **Pass criteria** | ✅ Tier decided (L1/L2/L3), boundaries confirmed, context clean, sub-agents decided |\n| **Exception path** | ⚠️ Tier unclear → default to L2; Boundaries unclear → ask the user; Resource issues → report honestly |\n\n**Behavior (auto-activated in this layer):**\n- ✅ Check existing knowledge and files first, don't reinvent the wheel\n- ✅ Tool Awakening Check: online needs → check skill availability, repair before proceeding\n- ✅ Tool priority: API/CLI → web_search → web_fetch → babata-browser (Tool Selector)\n- ✅ Framework Wake-up Check: L2+ tasks run `node scripts/framework-check.cjs`\n- ✅ Tavily probe: run `node scripts/tavily-probe.cjs` before online tasks\n- ✅ Capability detection: check before use, degrade gracefully\n- ✅ Knowledge Gap Scan: output known-unknown matrix before collection\n- ❌ Don't default to complex paths (zero-deploy → one command → full service, three tiers)\n- ❌ Don't skip Quality Gate G1\n\n### ② Sense — Gather\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Data sufficient? Hypothesis clear? |\n| **Pass criteria** | ✅ Multi-source verification (≥2 independent sources), hypothesis explicit |\n| **Exception path** | ⚠️ Insufficient sources → mark \"needs supplement\" don't block; Uncertainty → present to user |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Hypothesis explicit.** When ambiguous, don't silently choose — present multiple possibilities\n- ✅ **≥2 independent sources** for core judgments, cross-validate\n- ✅ **Multi-agent parallel** when ≥3 dimensions with no dependencies\n- ✅ **Agent E architecture** for large L3 (context >50% or ≥4 agents)\n- ✅ **Quality Gate G1** must pass before proceeding\n\n### ③ Think — Analyze\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What method to use? |\n| **Pass criteria** | ✅ Cross-validation done, falsifiable judgments generated, bias checked |\n| **Exception path** | ⚠️ Insufficient evidence → expand collection first; Method unclear → use falsifiable judgment + bias check |\n\n**Bias check (mandatory in this layer):**\n- Type A: Confirmation bias? Anchoring bias? Availability bias?\n- Type B: Framing effect? Sunk cost? Fundamental attribution error?\n- **Memory recall:** Check lessons/MEMORY for similar issues\n- ✅ Generate ≥5 falsifiable judgments (format: Judgment + support + falsifiable condition)\n- ❌ Don't pretend to use ACH/Bayesian without data/tools\n\n### ④ Optimize — Deliver\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What are the success criteria? |\n| **Pass criteria** | ✅ Output meets standards (verifiable goals), G2 self-check passed |\n| **Exception path** | ⚠️ Standards unclear → return to Suit → ask user; G2 failed → fix and retry |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Define success criteria first** — turn \"make it work\" into \"satisfy conditions X, Y, Z\"\n- ✅ **Minimum viable solution** — simple > complex. No premature abstraction\n- ❌ **Don't modify unrelated things** — fix one thing at a time\n- ✅ Upload to IMA via `node scripts/ima-upload.cjs`\n- ✅ Cron delivery: summary only, quality self-check before push\n\n**Quality Gate G2:** (Evidence-chain quality) Must pass before leaving this layer.\n\n### ⑤ Evolve — Reflect\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What was learned? Is the system degrading? |\n| **Pass criteria** | ✅ Lessons/Memory updated; pipelines showing no degradation; regressions identified |\n| **Exception path** | ⚠️ No improvements → record \"none\"; Degradation → mark \"needs fix\"; Rollback to ②/③/④ |\n\n**Quality Gates G3-G4:** Must pass before leaving this layer.\n\n## Version\n\n| Version | Date | Description |\n|:--------|:----:|-------------|\n| v2.0 | 2026-05-06 | Refactored: Prime Directives + MEV Five-Layer dual structure. Behavior guidelines, quality gates, bias checks, model rules, cognitive budget all internalized into execution layers. |\n| v4.9 → v6.0+ | 2026-05-12 | See SKILL.md changelog for full evolution: v6.3 adds honest audit (no facade capabilities), Agent E architecture, Prime Directive #5 Time Awareness. |\n\n*MEV Five-Layer Engine v2.0 — Framework Core*\n\nArchive v7.0.0: 10 files, 12103 bytes\n\nFiles: README.md (1292b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (3799b), SOUL.mev.md (6125b), _meta.json (129b)\n\nFile v7.0.0:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v7.0 — Kernel + Plugin architecture. Minimal immutable core, context-activated plugins, auto-dormancy.\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 7.0.0\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires:\n      bins: [python]\n      env: []\n---\n\n# MEV Engine v7.0 ⚙️\n\n> **v7.0: Kernel + Plugin. 核心不变，能力外挂，不用则眠。**\n\n## Architecture\n\n```\n┌─────────────────────────────────┐\n│         核心内核（不可变）         │\n│  SOUL.md: 身份+铁律+MEV骨架+工具表│\n│  TOOLS.md: 工具选择+护栏          │\n│  AGENTS.md: 工作区规则            │\n│  MEMORY.md: 长期记忆              │\n└────────────┬────────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────────┐\n│ active  ││ scene  ││ dormant     │\n│ (常加载) ││ (按需)  ││ (休眠参考)  │\n├────────┤├────────┤├────────────┤\n│·cron规则││·深度调研 ││·Forum协作  │\n│·工具唤醒││·证据链  ││·Report IR  │\n│·核心less││·偏误检查││·Agent并行  │\n│·搜索降级││·合规分析││           │\n└────────┘└────────┘└────────────┘\n```\n\n## Lifecycle\n\n```\n新能力 → scene/ (30天试用)\n  ↓ 触发≥3次\nactive/ ← 常加载\n  ↓ 30天未触发\ndormant/ ← 休眠\n  ↓ 同类问题复现\nscene/ ← 重新激活\n```\n\n## When to Use\nNon-trivial tasks / Research / Cron / Multi-agent\n\n## When NOT to Use\nSimple Q&A / File-only / User says \"skip\"\n\n---\n\n## Execution\n\n### Step 0: Kernel Boot (mandatory, unskippable)\n\n```bash\nnode scripts/mev-prefight.cjs\n```\n\nOutput required in delivery:\n```\n🔒 G0 PREFLIGHT\n[✅/❌] Framework: v{version}\n[✅/❌] Search: {full/rate_limited/degraded}\n[✅/❌] Time: {ISO timestamp} Asia/Shanghai\n→ {FULL | DEGRADED_OK | DEGRADED}\n```\n\n### Step 1: Tier + Plugin Activation\n\n```\n📊 Tier: L{1|2|3}\n🔌 Plugins: {auto-detected from task context}\n```\n\n| Task Type | Auto-activate |\n|:----------|:-------------|\n| Cron isolated | active/cron-rules |\n| L3 research | scene/deep-research |\n| Compliance analysis | scene/compliance-research |\n\n### Step 2: MEV Five Layers\n\nSuit → Sense → Think → Optimize → Evolve (见 SOUL.md 内核)\n\n### Step 3: Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Preflight: {result}\n[✅/❌] G6 IMA upload: {kb_name}\n[✅/❌] G7 Falsifiable: {n}/≥3\n[✅/❌] G8 Sources: A:{n} B:{n} C:{n}\n```\n\n---\n\n## Gate Skip Rules\n\n| Condition | Skip | Reason |\n|:----------|:-----|:------|\n| L1 task | G4-G8 | Annotate |\n| Cron isolated | No sub-agents | Auto-rule |\n| No search needed | G2=N/A | Preflight auto-detect |\n| No IMA upload | G6=N/A | Annotate |\n\n---\n\n## Scripts\n\n| Script | Purpose |\n|:-------|:--------|\n| `scripts/mev-prefight.cjs` | G1+G2+G3 unified preflight |\n| `scripts/framework-check.cjs` | Version + integrity (24h cache) |\n| `scripts/tavily-probe.cjs` | Tavily MCP availability |\n| `scripts/ima-upload.cjs` | IMA KB upload |\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v6.5 | 05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n| **v7.0** | **05-13** | **Kernel+Plugin architecture. Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules moved to plugins.** |\n\nFile v7.0.0:README.md\n\n# MEV Engine ⚙️\n\n**M**ission → **E**nvironment → **V**erification — A five-layer task execution engine.\n\nThe core execution framework of the Babata operating system. Each layer has: core question + verification criteria + exception paths.\n\n## Structure\n\n```\nPrime Directives (4 safety rules) → Highest priority\nMEV Five Layers (sole execution framework)\n  ① Suit    → Prepare & adapt (G1 + boundary check)\n  ② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n  ③ Think   → Analyze & falsify (falsifiable judgments + bias check)\n  ④ Optimize → Deliver (G2 + quality gate + IMA upload)\n  ⑤ Evolve  → Reflect (lessons + precedent check + skill generation)\n```\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\n```\n\nCopy the skill directory to your OpenClaw workspace's `skills/` folder.\n\n## Usage\n\nEach layer runs in sequence. See `SKILL.md` for full detail.\n\n## Content\n\n| File | Purpose |\n|------|---------|\n| `SKILL.md` | Full v6.3 capability definitions (English) |\n| `SOUL.mev.md` | Framework integration guide |\n| `scripts/` | Runtime scripts: `framework-check.cjs`, `tavily-probe.cjs`, `ima-upload.cjs` |\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v7.0.0:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"7.0.0\",\n  \"publishedAt\": 1778679779778\n}\n\nFile v7.0.0:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v7.0.0:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v7.0.0:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Suggested Fix\nLikely fix or next step\n\n### Metadata\n- Reproducible: yes | no | unknown\n- Related Files: path/to/file\n```\n\n## Feature request entry\n```md\n## [FEAT-YYYYMMDD-XXX] capability\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium\n**Status**: pending\n**Area**: product | ops | growth | security\n\n### Requested Capability\nWhat is missing\n\n### User Context\nWhy it matters\n\n### Suggested Implementation\nMinimal implementation direction\n```\n\n## Experiment entry\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n```\n\nFile v7.0.0:SOUL.mev.md\n\n# MEV Five-Layer Engine — Core Framework\n\n> This is not a manual. This is an operating system's execution protocol.\n> **Please merge the relevant parts into your agent's SOUL.md.**\n\n> **Core Philosophy:** Write it → Read it → Internalize it → Evolve it\n\n## Prime Directives (Safety Baseline, Highest Priority)\n\nNever violate under any circumstances:\n\n1. **Think first, act second — verify before external writes** — Any action affecting the outside world (sending messages, calling APIs, modifying configs) must get user confirmation first. Internal operations are free.\n2. **Memory is contract, not feeling** — All explicit instructions, preferences, decisions, and important events must be written to memory files within the current session.\n3. **Never touch user files** — Never delete or modify user files. Only delete self-generated content.\n4. **Know your authority** — Legal/financial/discipline matters require approval first; pure technical or risk-free work can proceed autonomously.\n\n## MEV Five Layers (Exclusive Execution Framework)\n\n**No exceptions — all tasks must go through MEV.** L1 uses fast-track (Suit→Optimize), L2 uses standard three-layer (Suit→Sense→Optimize), L3 uses full five layers. **Efficiency first: Suit layer evaluates complexity, determines depth.**\n\nEach layer: core question + verification criteria + exception paths.\n\n**Self-check after each layer. Fail → use exception path.**\n\n**Model rule:** Default to Flash model. Only switch to Pro when explicitly specified by user.\n\n**Context budget (self-check during execution):** Green <40% = normal → Yellow 40-70% = trim redundancy → Red >70% = trigger compaction.\n\n### ① Suit — Prepare\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Files read? Boundaries clear? Resources sufficient? Need to split? |\n| **Pass criteria** | ✅ Tier decided (L1/L2/L3), boundaries confirmed, context clean, sub-agents decided |\n| **Exception path** | ⚠️ Tier unclear → default to L2; Boundaries unclear → ask the user; Resource issues → report honestly |\n\n**Behavior (auto-activated in this layer):**\n- ✅ Check existing knowledge and files first, don't reinvent the wheel\n- ✅ Tool Awakening Check: online needs → check skill availability, repair before proceeding\n- ✅ Tool priority: API/CLI → web_search → web_fetch → babata-browser (Tool Selector)\n- ✅ Framework Wake-up Check: L2+ tasks run `node scripts/framework-check.cjs`\n- ✅ Tavily probe: run `node scripts/tavily-probe.cjs` before online tasks\n- ✅ Capability detection: check before use, degrade gracefully\n- ✅ Knowledge Gap Scan: output known-unknown matrix before collection\n- ❌ Don't default to complex paths (zero-deploy → one command → full service, three tiers)\n- ❌ Don't skip Quality Gate G1\n\n### ② Sense — Gather\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Data sufficient? Hypothesis clear? |\n| **Pass criteria** | ✅ Multi-source verification (≥2 independent sources), hypothesis explicit |\n| **Exception path** | ⚠️ Insufficient sources → mark \"needs supplement\" don't block; Uncertainty → present to user |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Hypothesis explicit.** When ambiguous, don't silently choose — present multiple possibilities\n- ✅ **≥2 independent sources** for core judgments, cross-validate\n- ✅ **Multi-agent parallel** when ≥3 dimensions with no dependencies\n- ✅ **Agent E architecture** for large L3 (context >50% or ≥4 agents)\n- ✅ **Quality Gate G1** must pass before proceeding\n\n### ③ Think — Analyze\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What method to use? |\n| **Pass criteria** | ✅ Cross-validation done, falsifiable judgments generated, bias checked |\n| **Exception path** | ⚠️ Insufficient evidence → expand collection first; Method unclear → use falsifiable judgment + bias check |\n\n**Bias check (mandatory in this layer):**\n- Type A: Confirmation bias? Anchoring bias? Availability bias?\n- Type B: Framing effect? Sunk cost? Fundamental attribution error?\n- **Memory recall:** Check lessons/MEMORY for similar issues\n- ✅ Generate ≥5 falsifiable judgments (format: Judgment + support + falsifiable condition)\n- ❌ Don't pretend to use ACH/Bayesian without data/tools\n\n### ④ Optimize — Deliver\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What are the success criteria? |\n| **Pass criteria** | ✅ Output meets standards (verifiable goals), G2 self-check passed |\n| **Exception path** | ⚠️ Standards unclear → return to Suit → ask user; G2 failed → fix and retry |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Define success criteria first** — turn \"make it work\" into \"satisfy conditions X, Y, Z\"\n- ✅ **Minimum viable solution** — simple > complex. No premature abstraction\n- ❌ **Don't modify unrelated things** — fix one thing at a time\n- ✅ Upload to IMA via `node scripts/ima-upload.cjs`\n- ✅ Cron delivery: summary only, quality self-check before push\n\n**Quality Gate G2:** (Evidence-chain quality) Must pass before leaving this layer.\n\n### ⑤ Evolve — Reflect\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What was learned? Is the system degrading? |\n| **Pass criteria** | ✅ Lessons/Memory updated; pipelines showing no degradation; regressions identified |\n| **Exception path** | ⚠️ No improvements → record \"none\"; Degradation → mark \"needs fix\"; Rollback to ②/③/④ |\n\n**Quality Gates G3-G4:** Must pass before leaving this layer.\n\n## Version\n\n| Version | Date | Description |\n|:--------|:----:|-------------|\n| v2.0 | 2026-05-06 | Refactored: Prime Directives + MEV Five-Layer dual structure. Behavior guidelines, quality gates, bias checks, model rules, cognitive budget all internalized into execution layers. |\n| v4.9 → v6.0+ | 2026-05-12 | See SKILL.md changelog for full evolution: v6.3 adds honest audit (no facade capabilities), Agent E architecture, Prime Directive #5 Time Awareness. |\n\n*MEV Five-Layer Engine v2.0 — Framework Core*\n\nArchive v2.20.0: 10 files, 12980 bytes\n\nFiles: README.md (1292b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (5693b), SOUL.mev.md (6125b), _meta.json (130b)\n\nFile v2.20.0:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v6.5 — Minimal-enforcement task execution framework.\n  1 preflight command + 9 gates with trust-but-verify validation.\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 2.20.0\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires:\n      bins: [python]\n      env: []\n---\n\n# MEV Engine v6.5 ⚙️\n\n> **v6.5 principle: 1 preflight command + 9 gates with validation. No trust-without-verify.**\n\n## When to Use\nNon-trivial tasks / Research / Cron jobs / Multi-agent work\n\n## When NOT to Use\nSimple Q&A / File-only operations / User says \"skip\"\n\n---\n\n## Prime Directives\n\n1. **Think first, act second** — verify before external writes\n2. **Memory is contract** — write decisions to memory, not mental notes\n3. **Never touch user files** — only delete self-generated content\n4. **Know your authority** — legal matters ask first, technical work proceed\n5. **⏱️ Never guess time** — verify before any temporal claim via `session_status` or `python3`\n\n---\n\n## Execution Model: 2 Phases × 9 Gates\n\nAll mandatory gates output as visible conversation blocks. No silent skip.\n\n### PHASE 1: PRE-FLIGHT (1 unified command + 2 explicit gates)\n\n**Run once at task start:**\n```bash\nnode scripts/mev-prefight.cjs\n```\n\nThis single command covers:\n- G1: Framework version + file integrity (24h cache)\n- G2: Tavily API availability (degraded → use search fallback chain)\n- G3: Current time (never guess)\n\n**Then output explicitly:**\n\n```\n📊 Tier → L{1|2|3}\n🔀 Agent E → {yes|no}  reason: {context>50%|>=4 agents|N/A}\n📉 Context → {green|yellow|red}%\n📋 Gap Scan:\n| Known | Unknown | Priority | Direction |\n|-------|---------|----------|-----------|\n| ...   | ...     | ...      | ...       |\n```\n\n**Gate 1-3** (Script, unified): `node scripts/mev-prefight.cjs`\n**Gate 4** (Explicit): Tier + Agent E shunt + **Context budget** (mandatory since v6.5)\n**Gate 5** (Explicit): Gap Scan table — >=3 gaps. Check: are collection directions feasible given current search availability?\n\n### AGENT E SHUNT (G4 decision)\n\nWhen >=4 sub-agents + heavy search -> Agent E isolated convergence:\n\n```\nParent: spawn A/B/C/D, wait for all\nAgent E (isolated): read findings -> cross-validate -> write report -> ima-upload -> push summary\nParent: receive summary -> run G5.5 verification -> delivery check\n```\n\n**Shunt when**: context >50% OR >=4 agents. **Skip when**: <=3 agents, report <8KB.\n\n### G5.5: E-VERIFY (new in v6.5)\n\nAfter Agent E completes, before DELIVERY CHECK:\n\n```\n🔍 E-VERIFY\n[✅/❌] Upload JSON valid: ima-upload returned {\"ok\":true}\n[✅/❌] Judgment count: {n} (>=3)\n[✅/❌] Source grading: A={n}, B={n}, C={n} (at least A+B > 0)\n-> {PASS | PASS with note | FAIL (re-spawn Agent E)}\n```\n\nThis gate prevents blind trust of Agent E output. Verifies metrics only (not content quality). If FAIL -> re-spawn E with error specifics.\n\n### PHASE 2: DELIVERY (3 explicit gates + fallback)\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] ima-upload -> {KB_ID} | 🆘 FALLBACK: saved to memory/ + push local path\n[✅/❌] Falsifiable judgments -> {count} / >=3 (format: \"Judgment: ... Support: ... Falsifiable: ...\")\n[✅/❌] Source grading -> A:{n} B:{n} C:{n} (minimum: A+B > 0)\n-> {PASS | PASS with degradation note | FAIL}\n```\n\n**Gate 6** (Explicit): IMA upload via `node scripts/ima-upload.cjs`. **If upload fails**: write to `memory/` + push local path to user (\"manual upload needed\").\n**Gate 7** (Explicit): >=3 falsifiable judgments (format: \"Judgment X: ... Support: ... Falsifiable condition: ...\")\n**Gate 8** (Explicit): Source grading — at least A/B/C levels labeled. Minimum A+B > 0.\n\n### When Gates Can Be Skipped\n\n| Condition | Skip which gates | Require reason |\n|:----------|:-----------------|:--------------:|\n| L1 task | G4-G8 | Yes (annotate) |\n| Cron isolated | G4 Agent E = N/A | Yes |\n| No search | G2 = N/A (preflight auto-detects) | No |\n| No IMA upload needed | G6 = N/A | Yes |\n| No Agent E used | G5.5 = N/A | No |\n\n---\n\n## Scenario-Based Capabilities (NOT mandatory)\n\n| Capability | When to use | When NOT to use |\n|------------|------------|----------------|\n| Bias quick-check | High-stakes analysis with subjective judgment | Routine research collection |\n| Iterative loop | >=2 high-priority gaps remain unfilled after first round | Most tasks (1 round sufficient) |\n| Multi-agent parallel | >=3 independent dimensions | 1-2 dimensions (sequential is fine) |\n| Evolve lessons | Task reveals a new pattern worth recording | Routine task repetition |\n\n---\n\n## Cron Session Rules\n\n1. No multi-agent parallel -> sequential collection\n2. No Evolve writes -> skip lessons.md and memory/cron\n3. Delivery quality check before push\n4. Failure auto-alert via failureAlert config\n\n---\n\n## Scripts\n\n| Script | Purpose | Called at |\n|--------|---------|-----------|\n| `scripts/mev-prefight.cjs` | Unified G1+G2+G3: framework + search + time | Phase 1 start |\n| `scripts/framework-check.cjs` | Version + file integrity (24h cache) | Inside mev-prefight |\n| `scripts/tavily-probe.cjs` | Tavily API availability | Inside mev-prefight |\n| `scripts/ima-upload.cjs` | Universal IMA KB upload | Gate 6 |\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v6.3 | 2026-05-12 | Honest audit: removed facade capabilities |\n| v6.4 | 2026-05-12 | Radical simplification: 2 phases, 8 gates. Removed 15 capabilities |\n| v6.5 | 2026-05-12 | **Trust-but-verify**: unified preflight script (1 cmd replaces 3), Agent E verification gate (G5.5), IMA upload fallback path, context budget added to G4. Inspired by multi-source-research auto-verification and deep-research-zh fault-tolerant output patterns |\n\nFile v2.20.0:README.md\n\n# MEV Engine ⚙️\n\n**M**ission → **E**nvironment → **V**erification — A five-layer task execution engine.\n\nThe core execution framework of the Babata operating system. Each layer has: core question + verification criteria + exception paths.\n\n## Structure\n\n```\nPrime Directives (4 safety rules) → Highest priority\nMEV Five Layers (sole execution framework)\n  ① Suit    → Prepare & adapt (G1 + boundary check)\n  ② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n  ③ Think   → Analyze & falsify (falsifiable judgments + bias check)\n  ④ Optimize → Deliver (G2 + quality gate + IMA upload)\n  ⑤ Evolve  → Reflect (lessons + precedent check + skill generation)\n```\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\n```\n\nCopy the skill directory to your OpenClaw workspace's `skills/` folder.\n\n## Usage\n\nEach layer runs in sequence. See `SKILL.md` for full detail.\n\n## Content\n\n| File | Purpose |\n|------|---------|\n| `SKILL.md` | Full v6.3 capability definitions (English) |\n| `SOUL.mev.md` | Framework integration guide |\n| `scripts/` | Runtime scripts: `framework-check.cjs`, `tavily-probe.cjs`, `ima-upload.cjs` |\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v2.20.0:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"2.20.0\",\n  \"publishedAt\": 1778597310930\n}\n\nFile v2.20.0:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v2.20.0:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v2.20.0:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Suggested Fix\nLikely fix or next step\n\n### Metadata\n- Reproducible: yes | no | unknown\n- Related Files: path/to/file\n```\n\n## Feature request entry\n```md\n## [FEAT-YYYYMMDD-XXX] capability\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium\n**Status**: pending\n**Area**: product | ops | growth | security\n\n### Requested Capability\nWhat is missing\n\n### User Context\nWhy it matters\n\n### Suggested Implementation\nMinimal implementation direction\n```\n\n## Experiment entry\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n```\n\nFile v2.20.0:SOUL.mev.md\n\n# MEV Five-Layer Engine — Core Framework\n\n> This is not a manual. This is an operating system's execution protocol.\n> **Please merge the relevant parts into your agent's SOUL.md.**\n\n> **Core Philosophy:** Write it → Read it → Internalize it → Evolve it\n\n## Prime Directives (Safety Baseline, Highest Priority)\n\nNever violate under any circumstances:\n\n1. **Think first, act second — verify before external writes** — Any action affecting the outside world (sending messages, calling APIs, modifying configs) must get user confirmation first. Internal operations are free.\n2. **Memory is contract, not feeling** — All explicit instructions, preferences, decisions, and important events must be written to memory files within the current session.\n3. **Never touch user files** — Never delete or modify user files. Only delete self-generated content.\n4. **Know your authority** — Legal/financial/discipline matters require approval first; pure technical or risk-free work can proceed autonomously.\n\n## MEV Five Layers (Exclusive Execution Framework)\n\n**No exceptions — all tasks must go through MEV.** L1 uses fast-track (Suit→Optimize), L2 uses standard three-layer (Suit→Sense→Optimize), L3 uses full five layers. **Efficiency first: Suit layer evaluates complexity, determines depth.**\n\nEach layer: core question + verification criteria + exception paths.\n\n**Self-check after each layer. Fail → use exception path.**\n\n**Model rule:** Default to Flash model. Only switch to Pro when explicitly specified by user.\n\n**Context budget (self-check during execution):** Green <40% = normal → Yellow 40-70% = trim redundancy → Red >70% = trigger compaction.\n\n### ① Suit — Prepare\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Files read? Boundaries clear? Resources sufficient? Need to split? |\n| **Pass criteria** | ✅ Tier decided (L1/L2/L3), boundaries confirmed, context clean, sub-agents decided |\n| **Exception path** | ⚠️ Tier unclear → default to L2; Boundaries unclear → ask the user; Resource issues → report honestly |\n\n**Behavior (auto-activated in this layer):**\n- ✅ Check existing knowledge and files first, don't reinvent the wheel\n- ✅ Tool Awakening Check: online needs → check skill availability, repair before proceeding\n- ✅ Tool priority: API/CLI → web_search → web_fetch → babata-browser (Tool Selector)\n- ✅ Framework Wake-up Check: L2+ tasks run `node scripts/framework-check.cjs`\n- ✅ Tavily probe: run `node scripts/tavily-probe.cjs` before online tasks\n- ✅ Capability detection: check before use, degrade gracefully\n- ✅ Knowledge Gap Scan: output known-unknown matrix before collection\n- ❌ Don't default to complex paths (zero-deploy → one command → full service, three tiers)\n- ❌ Don't skip Quality Gate G1\n\n### ② Sense — Gather\n\n| Element | Content |\n|---------|---------|\n| **Core question** | Data sufficient? Hypothesis clear? |\n| **Pass criteria** | ✅ Multi-source verification (≥2 independent sources), hypothesis explicit |\n| **Exception path** | ⚠️ Insufficient sources → mark \"needs supplement\" don't block; Uncertainty → present to user |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Hypothesis explicit.** When ambiguous, don't silently choose — present multiple possibilities\n- ✅ **≥2 independent sources** for core judgments, cross-validate\n- ✅ **Multi-agent parallel** when ≥3 dimensions with no dependencies\n- ✅ **Agent E architecture** for large L3 (context >50% or ≥4 agents)\n- ✅ **Quality Gate G1** must pass before proceeding\n\n### ③ Think — Analyze\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What method to use? |\n| **Pass criteria** | ✅ Cross-validation done, falsifiable judgments generated, bias checked |\n| **Exception path** | ⚠️ Insufficient evidence → expand collection first; Method unclear → use falsifiable judgment + bias check |\n\n**Bias check (mandatory in this layer):**\n- Type A: Confirmation bias? Anchoring bias? Availability bias?\n- Type B: Framing effect? Sunk cost? Fundamental attribution error?\n- **Memory recall:** Check lessons/MEMORY for similar issues\n- ✅ Generate ≥5 falsifiable judgments (format: Judgment + support + falsifiable condition)\n- ❌ Don't pretend to use ACH/Bayesian without data/tools\n\n### ④ Optimize — Deliver\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What are the success criteria? |\n| **Pass criteria** | ✅ Output meets standards (verifiable goals), G2 self-check passed |\n| **Exception path** | ⚠️ Standards unclear → return to Suit → ask user; G2 failed → fix and retry |\n\n**Behavior (auto-activated in this layer):**\n- ✅ **Define success criteria first** — turn \"make it work\" into \"satisfy conditions X, Y, Z\"\n- ✅ **Minimum viable solution** — simple > complex. No premature abstraction\n- ❌ **Don't modify unrelated things** — fix one thing at a time\n- ✅ Upload to IMA via `node scripts/ima-upload.cjs`\n- ✅ Cron delivery: summary only, quality self-check before push\n\n**Quality Gate G2:** (Evidence-chain quality) Must pass before leaving this layer.\n\n### ⑤ Evolve — Reflect\n\n| Element | Content |\n|---------|---------|\n| **Core question** | What was learned? Is the system degrading? |\n| **Pass criteria** | ✅ Lessons/Memory updated; pipelines showing no degradation; regressions identified |\n| **Exception path** | ⚠️ No improvements → record \"none\"; Degradation → mark \"needs fix\"; Rollback to ②/③/④ |\n\n**Quality Gates G3-G4:** Must pass before leaving this layer.\n\n## Version\n\n| Version | Date | Description |\n|:--------|:----:|-------------|\n| v2.0 | 2026-05-06 | Refactored: Prime Directives + MEV Five-Layer dual structure. Behavior guidelines, quality gates, bias checks, model rules, cognitive budget all internalized into execution layers. |\n| v4.9 → v6.0+ | 2026-05-12 | See SKILL.md changelog for full evolution: v6.3 adds honest audit (no facade capabilities), Agent E architecture, Prime Directive #5 Time Awareness. |\n\n*MEV Five-Layer Engine v2.0 — Framework Core*\n\nArchive v2.19.0: 10 files, 12721 bytes\n\nFiles: README.md (1292b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (5143b), SOUL.mev.md (6125b), _meta.json (130b)\n\nFile v2.19.0:SKILL.md\n\n---\nname: mev-engine\ndescription: |\n  MEV Engine v6.4 — Minimal-enforcement task execution framework.\n  5 mandatory gates across 2 phases. No decorations. What's listed must execute.\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 2.19.0\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires:\n      bins: [python]\n      env: []\n---\n\n# MEV Engine v6.4 ⚙️\n\n> **v6.4 principle: 5 mandatory gates. That's it. Everything else is scenario-based. No decoration.**\n\n## When to Use\nNon-trivial tasks / Research / Cron jobs / Multi-agent work\n\n## When NOT to Use\nSimple Q&A / File-only operations / User says \"skip\"\n\n---\n\n## Prime Directives\n\n1. **Think first, act second** — verify before external writes\n2. **Memory is contract** — write decisions to memory, not mental notes\n3. **Never touch user files** — only delete self-generated content\n4. **Know your authority** — legal matters ask first, technical work proceed\n5. **⏱️ Never guess time** — verify before any temporal claim via `session_status` or `python3`\n\n---\n\n## Execution Model: 2 Phases × 5 Gates\n\nAll 5 gates output as visible conversation blocks. No silent skip.\n\n### PHASE 1: PRE-FLIGHT (3 script gates + 2 explicit gates)\n\n```\n🔍 PRE-FLIGHT\n[✅/❌] framework-check → v{version}\n[✅/❌] Tavily probe → {ok|degraded}\n[⏰] Time → {ISO datetime}\n📊 Tier → L{1|2|3}\n🔀 Agent E needed? → {yes|no} reason: {context>50%|≥4 agents|N/A}\n📋 Gap Scan:\n| Known | Unknown | Priority | Direction |\n|-------|---------|----------|-----------|\n| ...   | ...     | ...      | ...       |\n```\n\n**Gate 1** (Script): `node scripts/framework-check.cjs` — version + file integrity\n**Gate 2** (Script): `node scripts/tavily-probe.cjs` — search availability\n**Gate 3** (Script): Time check — never guess\n**Gate 4** (Explicit): Tier + Agent E shunt decision — output in conversation\n**Gate 5** (Explicit): Gap Scan table — ≥3 gaps before collection starts\n\n### PHASE 2: DELIVERY (3 explicit gates)\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] ima-upload → {KB_ID}\n[✅/❌] Falsifiable judgments → {count} / ≥3\n[✅/❌] Source grading → A:{n} B:{n} C:{n}\n→ {PASS|PASS with note|FAIL}\n```\n\n**Gate 6** (Explicit): IMA upload confirmed via `node scripts/ima-upload.cjs`\n**Gate 7** (Explicit): ≥3 falsifiable judgments (format: \"Judgment: ... Support: ... Falsifiable condition: ...\")\n**Gate 8** (Explicit): Source grading — at least A/B/C levels labeled\n\n---\n\n## When Gates Can Be Skipped\n\nA gate can be marked `[N/A]` if:\n- L1 task (single-step, no research) → skip Gates 4-8\n- Cron isolated session (no sub-agents) → Gate 4 Agent E = `N/A`\n- Local-only task (no search needed) → Gate 2 = `N/A`\n\n**Skipping requires explicit reason.** No silent skip.\n\n---\n\n## Agent E Shunt (Gate 4)\n\nWhen ≥4 sub-agents spawned + search volume is heavy → spawn Agent E for convergence:\n\n```\nParent session: spawn A/B/C/D, wait for all\nAgent E (isolated): read findings → cross-validate → write report → ima-upload → push summary\nParent session: receive summary only\n```\n\nShunt when: context budget >50% or ≥4 agents. Skip when: ≤3 agents, report <8KB.\n\n---\n\n## Scenario-Based Capabilities (NOT mandatory)\n\nThese exist but are NOT required on every task. Use when the scenario fits.\n\n| Capability | When to use | When NOT to use |\n|------------|------------|----------------|\n| Bias quick-check | High-stakes analysis involving subjective judgment | Routine research collection |\n| Iterative loop | ≥2 high-priority gaps remain unfilled after first round | Most tasks (1 round is sufficient) |\n| Multi-agent parallel | ≥3 independent dimensions | 1-2 dimensions (sequential is fine) |\n| Evolve lessons | Task reveals a new pattern worth recording | Routine task repetition |\n\n---\n\n## Cron Session Special Rules\n\nApplies to all `sessionTarget: isolated` cron jobs:\n1. No multi-agent parallel — use sequential collection\n2. No Evolve writes — skip lessons.md and memory/cron writes\n3. Delivery quality check: upload success + source grading before push\n4. Failure auto-alert via failureAlert config\n\n---\n\n## Scripts\n\n| Script | Purpose | Called at |\n|--------|---------|-----------|\n| `scripts/framework-check.cjs` | Version + file integrity (24h cache) | Gate 1 |\n| `scripts/tavily-probe.cjs` | Tavily API availability check | Gate 2 |\n| `scripts/ima-upload.cjs` | Universal IMA KB upload | Gate 6 |\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v6.0 | 2026-05-11 | Full English localization |\n| v6.1 | 2026-05-12 | Agent E architecture, ima-upload.cjs integration |\n| v6.2 | 2026-05-12 | Prime Directive #5 Time Awareness |\n| v6.3 | 2026-05-12 | Honest audit: removed facade capabilities, mandatory Gap Scan, STOP checkpoints |\n| v6.4 | 2026-05-12 | **Radical simplification**: 2 phases, 8 gates (5 Pre-flight + 3 Delivery). Removed 15 scenario-based capabilities to non-mandatory. All mandatory gates output as visible conversation blocks. No silent skip allowed. Post-v6.3 audit revealed 9/20 capabilities unused on first task — v6.4 fixes this architecturally, not by adding more rules. |\n\nFile v2.19.0:README.md\n\n# MEV Engine ⚙️\n\n**M**ission → **E**nvironment → **V**erification — A five-layer task execution engine.\n\nThe core execution framework of the Babata operating system. Each layer has: core question + verification criteria + exception paths.\n\n## Structure\n\n```\nPrime Directives (4 safety rules) → Highest priority\nMEV Five Layers (sole execution framework)\n  ① Suit    → Prepare & adapt (G1 + boundary check)\n  ② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n  ③ Think   → Analyze & falsify (falsifiable judgments + bias check)\n  ④ Optimize → Deliver (G2 + quality gate + IMA upload)\n  ⑤ Evolve  → Reflect (lessons + precedent check + skill generation)\n```\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\n```\n\nCopy the skill directory to your OpenClaw workspace's `skills/` folder.\n\n## Usage\n\nEach layer runs in sequence. See `SKILL.md` for full detail.\n\n## Content\n\n| File | Purpose |\n|------|---------|\n| `SKILL.md` | Full v6.3 capability definitions (English) |\n| `SOUL.mev.md` | Framework integration guide |\n| `scripts/` | Runtime scripts: `framework-check.cjs`, `tavily-probe.cjs`, `ima-upload.cjs` |\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT\n\nFile v2.19.0:_meta.json\n\n{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"2.19.0\",\n  \"publishedAt\": 1778590647634\n}\n\nFile v2.19.0:references/eval-loop.md\n\n# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Metadata\n- Source: review | postmortem | user_feedback | qa\n- Related Files:\n- Tags:\n```\n\n## Important limit\n\nA logged experiment is not the same as a finished fix. If the production surface or operator-visible truth is still wrong, the experiment remains incomplete even if the local change looks promising.\n\nFile v2.19.0:references/promotion-guide.md\n\n# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path\n\nFile v2.19.0:references/schema.md\n\n# Learning Schema\n\n## Files\n- `.learnings/LEARNINGS.md`\n- `.learnings/ERRORS.md`\n- `.learnings/FEATURE_REQUESTS.md`\n\n## Learning entry\n```md\n## [LRN-YYYYMMDD-XXX] category\n\n**Logged**: ISO-8601 timestamp\n**Priority**: low | medium | high | critical\n**Status**: pending\n**Area**: workflow | tools | product | growth | security | infra\n\n### Summary\nOne-line learning\n\n### Details\nWhat happened and what is now understood\n\n### Suggested Action\nSpecific next action\n\n### Metadata\n- Source: user_feedback | error | review | postmortem\n- Related Files: path/to/file\n- Tags: tag1, tag2\n```\n\n## Error entry\n```md\n## [ERR-YYYYMMDD-XXX] name\n\n**Logged**: ISO-8601 timestamp\n**Priority**: high\n**Status**: pending\n**Area**: infra | product | growth | security | ops\n\n### Summary\nWhat failed\n\n### Error\nActual error or concise failure output\n\n### Sug\n\nArchive v2.18.2: 10 files, 14396 bytes\n\nFiles: README.md (784b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (9484b), SOUL.mev.md (5581b), _meta.json (130b)\n\nArchive v2.18.1: 10 files, 14298 bytes\n\nFiles: README.md (784b), references/eval-loop.md (3305b), references/promotion-guide.md (1175b), references/schema.md (1846b), scripts/log-experiment.mjs (1699b), scripts/log-learning.mjs (1683b), scripts/promote-learning.mjs (1263b), SKILL.md (8703b), SOUL.mev.md (5581b), _meta.json (130b)","readmeExcerpt":"Skill: Mev Engine Owner: meta-evo-creator Summary: MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。 Tags: execution:2.8.0, framework:2.8.0, latest:8.0.0, methodology:2.8.0 Version history: v8.0.0 | 2026-05-18T06:07:36.310Z | user v8.0.0: MEV Engine OpenClaw原生化。删除6个冗余自定义脚本，全面改用OpenClaw原生能力。MEV是驾驶手册，OpenClaw是引擎。零自定义脚本。 v7.2.1 | 2026-05-14T15:09:23.946Z | user v7.2.1: Self-audit — Plugin Di","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"┌─────────────────────────────┐\n│  OpenClaw（引擎）            │\n│  sessions · subagents       │\n│  tools · skills · memory    │\n│  cron · delivery · heartbeat│\n└────────────┬────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────┐\n│ SOUL.md││plugins/││memory/ │\n│ 内核    ││ 技能    ││ 记忆    │\n│ 不可变  ││ 按需加载││ 三层体系 │\n└────────┘└────────┘└────────┘"},{"language":"text","snippet":"tavily__tavily_search → web_fetch → babata-browser → 标注「不可达」"},{"language":"text","snippet":"🔒 DELIVERY CHECK\n[✅/❌] G0 Coverage: {n} sources / {n} dimensions\n[✅/❌] G1 Structure: 完整 / 缺失{list}\n[✅/❌] G2 Analysis: bias={PASS/修正} gap={无遗漏/已标记} evidence_map={n}/{total}\n[✅/❌] G3 Delivery: IMA={kb_name} push={sent/failed}\n[✅/❌] G4 Evolve: trace={written}"},{"language":"text","snippet":"遇到教训 → edit(plugin LEARNED PATTERNS 段)\n         → edit(core-lessons.plugin.md 索引)\n         \n激活：插件被Dispatch激活 → 教训自动加载\n退役：插件30天未触发 → 教训随之休眠"},{"language":"text","snippet":"scene/ (试用) → 触发≥3次 → active/ (常驻)\nactive/ → 30天未触发 → dormant/ (休眠)\ndormant/ → 同类问题复现 → scene/ (重新激活)"},{"language":"text","snippet":"┌─────────────────────────────┐\n│      Core Kernel (immutable) │\n│  Identity · MEV Skeleton     │\n│  Tool Table · Safety Rules   │\n└──────────┬──────────────────┘\n           │\n   ┌───────┼───────┐\n   ↓       ↓       ↓\n active   scene   dormant\n(常加载)  (按需)   (休眠)"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: mev-engine\ndescription: |\n  MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。\nhomepage: https://github.com/meta-evo-creator/mev-engine\nversion: 8.0.0\nmetadata:\n  openclaw:\n    emoji: ⚙️\n    requires: {}\n---\n\n# MEV Engine v8.0 ⚔️\n\n> **v8.0: 全部基于 OpenClaw 原生能力。MEV是驾驶手册，OpenClaw是引擎。零自定义脚本。**\n\n## 定位\n\nMEV Engine 不是代码框架，是**思考方法论 + 交付约定**。\n\n- 🧠 **MEV五层** → 怎么思考一个问题\n- 📋 **G0-G4门禁** → 怎么检查一个产出\n- 📝 **交付约定** → Sign-off / UNSOURCED / 证据映射\n- 🔄 **教训生命周期** → 学到的东西怎么不丢失\n\n**引擎是 OpenClaw。** 所有执行都走 OpenClaw 原生能力（sessions / subagents / tools / skills / memory / delivery）。\n\n---\n\n## 架构\n\n```\n┌─────────────────────────────┐\n│  OpenClaw（引擎）            │\n│  sessions · subagents       │\n│  tools · skills · memory    │\n│  cron · delivery · heartbeat│\n└────────────┬────────────────┘\n             │\n    ┌────────┼────────┐\n    ↓        ↓        ↓\n┌────────┐┌────────┐┌────────┐\n│ SOUL.md││plugins/││memory/ │\n│ 内核    ││ 技能    ││ 记忆    │\n│ 不可变  ││ 按需加载││ 三层体系 │\n└────────┘└────────┘└────────┘\n```\n\n---\n\n## MEV 五层（思考框架，不是代码流水线）\n\n> 内核只做Suit（入口适配）。Sense~Evolve是各插件设计内部流程时的参考框架。\n\n### ① Suit — 入口适配\n\n**不做的事：** 不再跑 `node mev-prefight.cjs`。\n\n**做的事：**\n- 读上下文：OpenClaw 已注入 Framework版本、当前时间、工具列表\n- 判定Tier：L1快答 / L2标准 / L3深度\n- 激活插件：`memory_search(PLUGIN-REGISTRY)` → 匹配 → `read` 加载插件\n\n**OpenClaw实现：** `read` + `memory_search` + skills auto-activation\n\n### ② Sense — 感知采集\n\n**搜索降级链（OpenClaw原生）：**\n```\ntavily__tavily_search → web_fetch → babata-browser → 标注「不可达」\n```\n\n工具选择表见 SOUL.md。\n\n### ③ Think — 分析加工\n\n视任务复杂度，激活对应插件：\n- 纪检法规 → `discipline-inspect`（4-Agent编排 + RAG）\n- 医学研究 → `med-research`（Scout→Draft→Review）\n- 深度调研 → `deep-research`（L3专用）\n\n偏误检查、ACH、证据映射内置于各插件。\n\n### ④ Optimize — 交付检查\n\n**G0-G4 门禁（OpenClaw原生）：**\n\n| 门禁 | 检查什么 | OpenClaw实现 |\n|:-----|:--------|:-----------|\n| G0 覆盖度 | 几个信源？几个维度？ | 手动统计（tavily result count + web_fetch URL数） |\n| G1 结构 | 缺哪段？ | 对照模板检查 |\n| G2 分析 | 偏误？遗漏？证据链？ | bias-check + evidence_map |\n| G3 交付 | IMA上传？推送？ | `ima-skill` + `wecom_mcp` |\n| G4 复盘 | 日志写了？教训沉淀了？ | `write(memory/YYYY-MM-DD.md)` + `edit(plugin LEARNED PATTERNS)` |\n\n**交付格式要求：**\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Coverage: {n} sources / {n} dimensions\n[✅/❌] G1 Structure: 完整 / 缺失{list}\n[✅/❌] G2 Analysis: bias={PASS/修正} gap={无遗漏/已标记} evidence_map={n}/{total}\n[✅/❌] G3 Delivery: IMA={kb_name} push={sent/failed}\n[✅/❌] G4 Evolve: trace={written}\n```\n\n### ⑤ Evolve — 进化沉淀\n\n**教训生命周期（OpenClaw原生）：**\n\n```\n遇到教训 → edit(plugin LEARNED PATTERNS 段)\n         → edit(core-lessons.plugin.md 索引)\n         \n激活：插件被Dispatch激活 → 教训自动加载\n退役：插件30天未触发 → 教训随之休眠\n```\n\n每日日志：`write(memory/YYYY-MM-DD.md)`\n长期记忆提炼：`edit(MEMORY.md)`（每几日从每日日志提炼）\n\n---\n\n## Sign-off Protocol\n\n> 来源：Anthropic Financial-Services → 纪检场景同构。AI Drafts, Humans Sign Off.\n\n每个分析类产出必须带审批节点。详见各插件 agent 指令。\n\n---\n\n## 跳过规则\n\n| 条件 | 跳过 |\n|:-----|:-----|\n| L1 简单查询 | G0-G4 + checkpoint |\n| Cron 隔离 | **禁止子代理**，强制 G0-G4 |\n| 无需 IMA | G3 IMA=N/A |\n\n---\n\n## 插件生命周期（OpenClaw原生）\n\n```\nscene/ (试用) → 触发≥3次 → active/ (常驻)\nactive/ → 30天未触发 → dormant/ (休眠)\ndormant/ → 同类问题复现 → "},{"path":"README.md","content":"# MEV Engine v7.0 ⚙️\n\n> **Kernel + Plugin Architecture.** Minimal immutable core, context-activated plugins, auto-dormancy.\n\n**Mission → Environment → Verification** — A five-layer task execution engine.\nThe core execution framework of Babata OS. Each layer: core question + verification criteria + exception paths.\n\n---\n\n## v7.0 Architecture\n\n```\n┌─────────────────────────────┐\n│      Core Kernel (immutable) │\n│  Identity · MEV Skeleton     │\n│  Tool Table · Safety Rules   │\n└──────────┬──────────────────┘\n           │\n   ┌───────┼───────┐\n   ↓       ↓       ↓\n active   scene   dormant\n(常加载)  (按需)   (休眠)\n```\n\n**Lifecycle:** scene(30d trial) → active(triggered ≥3x) → dormant(30d unused) → scene(reactivate)\n\n---\n\n## MEV Five Layers\n\n```\n① Suit    → Prepare & adapt (G0 preflight + boundary check)\n② Sense   → Gather & collect (hypothesis explicit, ≥2 sources)\n③ Think   → Analyze & falsify (bias check + method selection)\n④ Optimize → Deliver (G0-G4 五层递进门禁)\n⑤ Evolve  → Reflect (lessons + framework audit)\n```\n\n---\n\n## Delivery Gates\n\n```\n🔒 DELIVERY CHECK\n[✅/❌] G0 Coverage: {n} sources\n[✅/❌] G1 Structure: 完整\n[✅/❌] G2 Analysis: bias={} evidence_map={}\n[✅/❌] G3 Delivery: IMA={} push={}\n[✅/❌] G4 Evolve: trace={}\n```\n\n---\n\n## Changelog\n\n| Version | Date | Changes |\n|:----|:----|------|\n| v7.0.0 | 2026-05-13 | **Kernel+Plugin architecture.** Core immutable, capabilities as plugins, auto-dormancy lifecycle. MEV skeleton preserved, specific rules extracted to plugins. |\n| v6.5.0 | 2026-05-12 | Trust-but-verify: unified preflight, Agent E verify, IMA fallback |\n\n---\n\n## Install\n\n```bash\ngit clone https://github.com/meta-evo-creator/mev-engine.git\nclawhub install mev-engine\n```\n\n## Dependencies\n\nZero external dependencies. Requires Python for time-awareness probe.\n\n## License\n\nMIT"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn7cy2z5e60pxd0830tar97xwx866ydt\",\n  \"slug\": \"mev-engine\",\n  \"version\": \"8.0.0\",\n  \"publishedAt\": 1779084456310\n}"},{"path":"references/eval-loop.md","content":"# Eval Loop for Self-Improvement\n\nUse this reference when a repeated failure should become a tested operational improvement instead of only a logged lesson.\n\n## Goal\n\nDo not only ask \"what did we learn?\"\nAlso ask:\n- what is the current baseline?\n- what exact guardrail or rule changed?\n- how will we measure whether it helped?\n- should we keep or discard the change?\n\n## Use this loop for\n- repeated Mission Control wording failures\n- missing receipts / missing proof chains\n- deploy closeout failures\n- stale operator-facing surfaces\n- repeated handoff mistakes between agents\n- recurring SOP/checklist changes\n\n## 1. Define the target\n\nState one concrete thing you want to improve.\n\nExamples:\n- Hunter summary should always include concrete links and details\n- ClawLite deploy closeout should never stop at code-ready status\n- Mission Control front-end should render source links from structured fields\n\n## 2. Write 3-5 binary evals\n\nEach eval must be yes/no.\n\nExamples for summary quality:\n- Does the summary include at least one artifact path or URL?\n- Does the summary include evidence links when external proof matters?\n- Does the summary include a detail block describing what actually changed?\n- Does the summary include the next handoff or recovery action?\n- Does the operator-facing surface actually render these fields?\n\nExamples for deploy closeout:\n- Is the deployed commit hash recorded?\n- Is a deployment ref/URL recorded?\n- Was the production page or sitemap actually verified?\n- Was a structured receipt written?\n- Is the final state classified with the correct deploy-state vocabulary?\n\n## 3. Capture baseline\n\nBefore changing the rule/SOP/skill/checklist:\n- record the current failure pattern\n- record which evals currently fail\n- treat this as the baseline state\n\n## 4. Change only one thing\n\nGood changes:\n- one wording rule\n- one new checklist item\n- one schema field\n- one render mapping\n- one validation step\n\nBad changes:\n- rewriting everything at once\n- adding five new rules at once\n- changing wording and schema and code together unless absolutely required\n\n## 5. Re-check and classify\n\nAfter the single change:\n- run the same evals again\n- note which checks improved\n- decide:\n  - KEEP\n  - DISCARD\n  - PARTIAL_KEEP\n\n## 6. Promotion rule\n\nOnly promote broadly reusable changes after they pass the eval loop or after operator review confirms the change materially reduced the failure.\n\n## Suggested experiment entry format\n\n```md\n## [EXP-YYYYMMDD-XXX] experiment\n\n**Logged**: ISO-8601 timestamp\n**Priority**: medium | high | critical\n**Status**: baseline | testing | keep | discard | partial_keep\n**Area**: workflow | tools | product | growth | security | infra | ops\n\n### Target\nWhat repeated problem is being improved\n\n### Baseline\nWhat was failing before the change\n\n### Mutation\nThe single change introduced\n\n### Binary Evals\n- [ ] Eval 1\n- [ ] Eval 2\n- [ ] Eval 3\n\n### Result\nWhat improved / did not improve\n\n### Keep or Discard\nkeep | discard | partial_keep\n\n### Meta"},{"path":"references/promotion-guide.md","content":"# Promotion Guide\n\nUse promotion only when a learning is broadly reusable.\n\n## Promote to AGENTS.md\nWhen the learning changes execution workflow.\nExamples:\n- deploy ownership rules\n- acceptance ownership rules\n- escalation timing\n\n## Promote to TOOLS.md\nWhen the learning is an environment/tool routing rule.\nExamples:\n- use Tavily before Brave\n- key locations in Keychain\n- browser session attach rules\n\n## Promote to SOUL.md\nWhen the learning is a behavior/principle rule.\nExamples:\n- do not let no-assignment closeout replace required deliverables\n- do not treat shallow checks as full acceptance\n\n## Promote to Obsidian\nWhen the learning should become reusable operator material, marketing proof, or an operations note outside transient chat.\n\nBy default, Obsidian-style exports go to the local safe fallback:\n- `.learnings/exports/obsidian/`\n\nIf you want a real vault destination, set `OBSIDIAN_LEARNINGS_DIR` explicitly before running the promotion script.\nAlways confirm the printed target path first, or use `--dry-run`.\n\nExample:\n- `node scripts/promote-learning.mjs obsidian \"Reusable learning\" --dry-run`\n- then rerun without `--dry-run` after confirming the path"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。 Skill: Mev Engine Owner: meta-evo-creator Summary: MEV Engine v8.0 ⚔️ — OpenClaw原生。MEV五层指导思想+交付约定+教训生命周期，全部基于OpenClaw内置能力，零自定义脚本。 Tags: execution:2.8.0, framework:2.8.0, latest:8.0.0, methodology:2.8.0 Version history: v8.0.0 | 2026-05-18T06:07:36.310Z | user v8.0.0: MEV Engine OpenClaw原生化。删除6个冗余自定义脚本，全面改用OpenClaw原生能力。MEV是驾驶手册，OpenClaw是引擎。零自定义脚本。 v7.2.1 | 2026-05-14T15:09:23.946Z | user v7.2.1: Self-audit — Plugin Di","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":1019,"uniquenessScore":61,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-10T15:42:56.186Z","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":"2026-10-10T15:42:56.186Z","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":"2026-10-10T21:43:46.126Z","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"}]}}}