{"id":"7d995bff-ba50-4b2f-9a49-f2fd553e3678","entityType":"agent","slug":"clawhub-athola-nm-abstract-escalation-governance","name":"escalation-governance","canonicalUrl":"https://www.xpersona.co/agent/clawhub-athola-nm-abstract-escalation-governance","canonicalPath":"/agent/clawhub-athola-nm-abstract-escalation-governance","generatedAt":"2026-10-10T06:06:18.532Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-10T05:08:29.035Z","emptyReason":null},"description":"Assess whether to escalate models Skill: escalation-governance Owner: athola Summary: Assess whether to escalate models Tags: latest:1.9.19 Version history: v1.9.19 | 2026-08-26T13:02:37.749Z | user Release v1.9.19 v1.9.18 | 2026-08-15T21:27:20.119Z | user Release v1.9.18 v1.9.17 | 2026-07-30T05:27:15.062Z | user Release v1.9.17 v1.9.16 | 2026-07-14T19:44:12.602Z | user Release v1.9.16 v1.9.15 | 2026-07-04T21:18:53.554Z | user Release v1.9.15 v1.9.14","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.7K downloads reported by the source. Last updated 10/10/2026.","installCommand":"clawhub skill install s17emme0e2m3cpf7k2jvp3a84984b8z9:nm-abstract-escalation-governance","sourceUrl":"https://clawhub.ai/athola/nm-abstract-escalation-governance","homepage":"https://clawhub.ai/athola/skills/nm-abstract-escalation-governance","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/athola/nm-abstract-escalation-governance","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/athola/skills/nm-abstract-escalation-governance","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":40,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Assess whether to escalate models Skill: escalation-governance Owner: athola Summary: Assess whether to escalate models Tags: latest:1.9.19 Version history: v1."},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-10T05:08:29.035Z","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":"2026-10-10T05:08:29.035Z","emptyReason":null},"stars":null,"forks":null,"downloads":1665,"packageName":null,"latestVersion":"1.9.19","tractionLabel":"1.7K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T05:08:29.035Z","emptyReason":null},"lastUpdatedAt":"2026-10-10T05:08:29.035Z","lastCrawledAt":"2026-10-10T05:08:29.035Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-11T05:08:29.035Z","lastVerifiedAt":null,"highlights":[{"version":"1.9.19","createdAt":"2026-08-26T13:02:37.749Z","changelog":"Release v1.9.19","fileCount":3,"zipByteSize":6493},{"version":"1.9.18","createdAt":"2026-08-15T21:27:20.119Z","changelog":"Release v1.9.18","fileCount":3,"zipByteSize":6534},{"version":"1.9.17","createdAt":"2026-07-30T05:27:15.062Z","changelog":"Release v1.9.17","fileCount":3,"zipByteSize":6588},{"version":"1.9.16","createdAt":"2026-07-14T19:44:12.602Z","changelog":"Release v1.9.16","fileCount":3,"zipByteSize":6581},{"version":"1.9.15","createdAt":"2026-07-04T21:18:53.554Z","changelog":"Release v1.9.15","fileCount":3,"zipByteSize":6549},{"version":"1.9.14","createdAt":"2026-06-30T17:49:28.692Z","changelog":"Release v1.9.14","fileCount":3,"zipByteSize":6590},{"version":"1.9.13","createdAt":"2026-06-27T16:14:07.001Z","changelog":"Release v1.9.13","fileCount":3,"zipByteSize":6649},{"version":"1.9.12","createdAt":"2026-06-19T03:07:09.582Z","changelog":"Release v1.9.12","fileCount":3,"zipByteSize":6570}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17emme0e2m3cpf7k2jvp3a84984b8z9:nm-abstract-escalation-governance","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-athola-nm-abstract-escalation-governance/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-10T06:06:18.528Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-athola-nm-abstract-escalation-governance/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":"2026-10-10T05:08:29.035Z","emptyReason":null},"readme":"Skill: escalation-governance\n\nOwner: athola\n\nSummary: Assess whether to escalate models\n\nTags: latest:1.9.19\n\nVersion history:\n\nv1.9.19 | 2026-08-26T13:02:37.749Z | user\n\nRelease v1.9.19\n\nv1.9.18 | 2026-08-15T21:27:20.119Z | user\n\nRelease v1.9.18\n\nv1.9.17 | 2026-07-30T05:27:15.062Z | user\n\nRelease v1.9.17\n\nv1.9.16 | 2026-07-14T19:44:12.602Z | user\n\nRelease v1.9.16\n\nv1.9.15 | 2026-07-04T21:18:53.554Z | user\n\nRelease v1.9.15\n\nv1.9.14 | 2026-06-30T17:49:28.692Z | user\n\nRelease v1.9.14\n\nv1.9.13 | 2026-06-27T16:14:07.001Z | user\n\nRelease v1.9.13\n\nv1.9.12 | 2026-06-19T03:07:09.582Z | user\n\nRelease v1.9.12\n\nv1.8.6 | 2026-06-07T21:21:20.256Z | user\n\nRelease v1.9.11\n\nv1.8.5 | 2026-05-09T02:14:42.463Z | user\n\nRelease v1.9.5\n\nv1.8.4 | 2026-05-06T14:14:09.611Z | user\n\nRelease v1.9.4\n\nv1.8.3 | 2026-04-10T05:44:36.856Z | user\n\nRelease v1.8.3\n\nv1.8.2 | 2026-04-06T14:17:10.666Z | user\n\nRelease v1.8.2\n\nv1.0.0 | 2026-04-06T05:55:06.168Z | auto\n\n- Initial release of the escalation-governance skill, adapted from claude-night-market/abstract.\n- Provides a structured framework and protocol for assessing when to escalate model complexity in agent workflows.\n- Outlines legitimate vs. illegitimate triggers for escalation and questions to consider before escalating.\n- Details orchestrator authority, agent schema escalation hints, and integration with agent workflows.\n- Includes quick reference tables and red flags to help avoid unnecessary escalation.\n- Adds verification steps for model capability and protocol compliance.\n\nArchive index:\n\nArchive v1.9.19: 3 files, 6493 bytes\n\nFiles: skill-card.md (1731b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.19:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.19:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.19\",\n  \"publishedAt\": 1787749357749\n}\n\nFile v1.9.19:skill-card.md\n\n## Description:\n\nAssess whether to escalate models.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[athola](https://clawhub.ai/user/athola)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nDevelopers and agent operators use this skill to decide when model escalation is justified, document the reason, scope the escalated work, and return to a more efficient model after the reasoning task is complete.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Broad triggers may cause the skill to appear in general agent or orchestration discussions.\n\nMitigation: Apply the guidance only when a task involves model escalation, model selection, governance, agent orchestration, or reasoning-depth evaluation.\n\nRisk: Claude-version notes in the artifact may become stale.\n\nMitigation: Check current product documentation before relying on version-specific operational guidance.\n\n## Reference(s):\n\n- [ClawHub Skill Page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance)\n- [OpenClaw Homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract)\n\n## Skill Output:\n\n**Output Type(s):** [Guidance, Markdown]\n\n**Output Format:** [Markdown guidance with decision checklists and escalation criteria]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [No code execution; Markdown-only guidance.]\n\n## Skill Version(s):\n\n1.9.19 (source: server release evidence)\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\nArchive v1.9.18: 3 files, 6534 bytes\n\nFiles: skill-card.md (1841b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.18:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.18:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.18\",\n  \"publishedAt\": 1786829240119\n}\n\nFile v1.9.18:skill-card.md\n\n## Description:\n\nAssess whether to escalate models.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[athola](https://clawhub.ai/user/athola)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nDevelopers and agent operators use this skill to decide when model escalation is justified, how to investigate before escalating, and how to document the reasoning behind a model or effort-level change.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Broad activation terms may cause this advisory skill to appear during adjacent agent, orchestration, or governance conversations.\n\nMitigation: Review and narrow the trigger list if tighter activation behavior is required.\n\nRisk: Escalation guidance can influence model-selection decisions even though the skill is documentation-only.\n\nMitigation: Treat the output as advisory and require human or orchestrator review for high-cost or high-stakes escalation decisions.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance)\n- [OpenClaw homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, configuration, guidance]\n\n**Output Format:** [Markdown guidance with checklists, tables, and configuration examples]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Documentation-only advisory output; no tool access or persistent actions are requested.]\n\n## Skill Version(s):\n\n1.9.18 (source: ClawHub release metadata)\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\nArchive v1.9.17: 3 files, 6588 bytes\n\nFiles: skill-card.md (2006b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.17:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.17:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.17\",\n  \"publishedAt\": 1785389235062\n}\n\nFile v1.9.17:skill-card.md\n\n## Description: <br>\nAssess whether to escalate models. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[athola](https://clawhub.ai/user/athola) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers and agent operators use this skill to decide when model escalation is justified, document the reason for escalation, and return to a more efficient model after the deeper reasoning task is complete. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Users may treat the advisory model-selection guidance as a substitute for evaluating the specific task context. <br>\nMitigation: Require documented investigation, escalation scope, and success criteria before changing model capability. <br>\nRisk: The skill references an optional external plugin that is outside the scanned artifact. <br>\nMitigation: Evaluate and scan the external plugin separately before installing or relying on it. <br>\n\n\n## Reference(s): <br>\n- [ClawHub Skill Page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance) <br>\n- [clawdis homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [guidance, markdown, configuration] <br>\n**Output Format:** [Markdown guidance with YAML configuration examples and decision tables] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Advisory model-selection guidance; no code execution, credential handling, or hidden data access is described in the security evidence.] <br>\n\n## Skill Version(s): <br>\n1.9.17 (source: server release metadata; artifact frontmatter lists 1.9.8) <br>\n\n## Ethical Considerations: <br>\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. <br>\n\nArchive v1.9.16: 3 files, 6581 bytes\n\nFiles: skill-card.md (1984b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.16:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.16:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.16\",\n  \"publishedAt\": 1784058252602\n}\n\nFile v1.9.16:skill-card.md\n\n## Description: <br>\nAssess whether to escalate models. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[athola](https://clawhub.ai/user/athola) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers and agent operators use this skill to decide when model escalation is justified, document the reason, and avoid unnecessary cost or latency from premature escalation. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The skill may influence an agent to use more capable and potentially more costly models. <br>\nMitigation: Require documented investigation, a bounded escalation scope, and a cost-benefit justification before changing model capability. <br>\nRisk: Incorrect escalation guidance could cause unnecessary latency or missed escalation for genuinely complex tasks. <br>\nMitigation: Review the decision framework against local model policy and monitor escalations during rollout. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance) <br>\n- [Source homepage from ClawHub metadata](https://github.com/athola/claude-night-market/tree/master/plugins/abstract) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [guidance, markdown] <br>\n**Output Format:** [Markdown guidance with checklists, decision criteria, and protocol steps] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Documentation-only skill; no code execution, persistence, data access, or credential handling was identified by the security evidence.] <br>\n\n## Skill Version(s): <br>\n1.9.16 (source: ClawHub release metadata) <br>\n\n## Ethical Considerations: <br>\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. <br>\n\nArchive v1.9.15: 3 files, 6549 bytes\n\nFiles: skill-card.md (1904b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.15:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.15:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.15\",\n  \"publishedAt\": 1783199933554\n}\n\nFile v1.9.15:skill-card.md\n\n## Description: <br>\nAssess whether to escalate models. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[athola](https://clawhub.ai/user/athola) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers and agent operators use this skill to decide when model escalation is justified, document the trade-off, and avoid using stronger models before investigating failures or ambiguity. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Following model and effort recommendations can increase cost and latency. <br>\nMitigation: Review the escalation justification, scope, and expected benefit before changing model or effort level. <br>\nRisk: Version-specific model and effort details may become stale as tooling changes. <br>\nMitigation: Check current model availability and effort-control behavior in the target agent environment before relying on version-specific guidance. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance) <br>\n- [OpenClaw homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [guidance, markdown, configuration] <br>\n**Output Format:** [Markdown guidance with decision tables and YAML examples] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Non-executable guidance; no API keys, MCP tools, or credential environment variables detected.] <br>\n\n## Skill Version(s): <br>\n1.9.15 (source: server release metadata) <br>\n\n## Ethical Considerations: <br>\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. <br>\n\nArchive v1.9.14: 3 files, 6590 bytes\n\nFiles: skill-card.md (2016b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.14:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.14:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.14\",\n  \"publishedAt\": 1782841768692\n}\n\nFile v1.9.14:skill-card.md\n\n## Description: <br>\nAssesses whether an agent should escalate to a more capable model. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[athola](https://clawhub.ai/user/athola) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers and agent operators use this skill to decide when model escalation is justified, document the reason and scope, and avoid using larger models as a substitute for investigation. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Automatic documentation or governance updates could introduce incorrect or misleading escalation guidance. <br>\nMitigation: Use report-only wording when automatic changes are not desired, and review proposed AGENTS or CONTRIBUTING changes before applying them. <br>\nRisk: Applying escalation guidance mechanically could increase cost and latency or miss a high-stakes need for deeper reasoning. <br>\nMitigation: Require investigation first, document the reason and scope for escalation, define success, and return to the efficient model after the scoped reasoning task. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance) <br>\n- [Metadata homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Guidance, Markdown] <br>\n**Output Format:** [Markdown guidance] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Decision framework and escalation protocol text for agent reasoning; no tools or credentials required.] <br>\n\n## Skill Version(s): <br>\n1.9.14 (source: server release metadata) <br>\n\n## Ethical Considerations: <br>\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. <br>\n\nArchive v1.9.13: 3 files, 6649 bytes\n\nFiles: skill-card.md (2170b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.13:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.13:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.13\",\n  \"publishedAt\": 1782576847001\n}\n\nFile v1.9.13:skill-card.md\n\n## Description: <br>\nAssess whether to escalate models. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[athola](https://clawhub.ai/user/athola) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers and agent orchestrators use this skill to decide when model escalation is justified, document the reason for escalation, and return to a more efficient model after the higher-capability task is complete. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Escalation guidance can lead to unnecessary cost or latency if an agent changes models before investigating the root cause of a task failure. <br>\nMitigation: Require the agent to document the capability gap, scope the escalated subtask, define success, and return to the efficient model promptly. <br>\nRisk: The artifact includes operational decision guidance that may be misapplied to high-stakes or security-sensitive work. <br>\nMitigation: Review escalation decisions before acting on them and scan the skill before deployment, consistent with the clean security verdict and reviewer guidance. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance) <br>\n- [Abstract plugin homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Guidance, Markdown, Configuration] <br>\n**Output Format:** [Markdown guidance with YAML configuration examples and decision tables] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Supports model-escalation decisions; no API keys or credential environment variables were detected in the submitted artifact.] <br>\n\n## Skill Version(s): <br>\n1.9.13 (source: server release metadata; artifact frontmatter says 1.9.8) <br>\n\n## Ethical Considerations: <br>\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. <br>\n\nArchive v1.9.12: 3 files, 6570 bytes\n\nFiles: skill-card.md (2004b), SKILL.md (12206b), _meta.json (153b)\n\nFile v1.9.12:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.9.12:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.12\",\n  \"publishedAt\": 1781838429582\n}\n\nFile v1.9.12:skill-card.md\n\n## Description: <br>\nAssesses whether an agent should escalate to a higher-capability model after investigation. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[athola](https://clawhub.ai/user/athola) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers and agent orchestrators use this skill to decide when model escalation is justified, document the reason and scope, and return to a lower-cost model after the deeper reasoning task is complete. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The skill can influence model-escalation decisions in agent and orchestration workflows. <br>\nMitigation: Keep activation focused on escalation governance use cases and require documented justification before changing model capability. <br>\nRisk: Version-specific model and effort-control guidance can become stale as provider behavior changes. <br>\nMitigation: Verify current model availability and effort-control behavior before relying on those operational notes. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/athola/nm-abstract-escalation-governance) <br>\n- [Homepage metadata](https://github.com/athola/claude-night-market/tree/master/plugins/abstract) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Guidance, Markdown, Configuration] <br>\n**Output Format:** [Markdown guidance with tables and YAML examples] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Advisory decision framework; no executable code or hidden access behavior was reported in the security evidence.] <br>\n\n## Skill Version(s): <br>\n1.9.12 (source: ClawHub release metadata) <br>\n\n## Ethical Considerations: <br>\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. <br>\n\nArchive v1.8.6: 3 files, 6407 bytes\n\nFiles: skill-card.md (1575b), SKILL.md (12206b), _meta.json (152b)\n\nFile v1.8.6:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72 for Opus 4.6, and `high` became the ceiling on that model. Claude Code 2.1.111 reintroduced `max` and added `xhigh` (between `high` and `max`) for Opus 4.7 only; on other models `xhigh` falls back to `high`. Symbols: ○ (low) ◐ (medium) ● (high) ◉ (xhigh) ★ (max). Use `/effort` (interactive slider since 2.1.111) or `/effort auto` to reset. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n| Opus 4.7@high → escalate | Opus 4.7@xhigh or @max | Deep architectural analysis on Opus 4.7 specifically |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\nFile v1.8.6:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.8.6\",\n  \"publishedAt\": 1780867280256\n}\n\nFile v1.8.6:skill-card.md\n\n## Description: <br>\nAssess whether to escalate models. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[athola](https://clawhub.ai/user/athola) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers and agent operators use this skill to decide when model escalation is justified after investigation, especially for complex, high-stakes, or ambiguous tasks. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The skill may appear in broader agent or governance discussions and be treated as an automatic decision rule. <br>\nMitigation: Treat its advice as optional guidance and require an explicit investigation-backed escalation rationale. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/athola/nm-abstract-escalation-governance) <br>\n- [Project homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [guidance, markdown, text] <br>\n**Output Format:** [Markdown guidance] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Documentation-only; no code execution or data access.] <br>\n\n## Skill Version(s): <br>\n1.8.6 (source: server release metadata) <br>\n\n## Ethical Considerations: <br>\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. <br>\n\nArchive v1.8.5: 3 files, 6571 bytes\n\nFiles: skill-card.md (1984b), SKILL.md (12120b), _meta.json (152b)\n\nFile v1.8.5:SKILL.md\n\n---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.5\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capability | Question your assumptions |\n| Time pressure | Urgency doesn't change task complexity | Systematic investigation is faster |\n| Uncertainty without investigation | You haven't tried to understand yet | Gather evidence first |\n| \"Just to be safe\" | False safety - wastes resources | Assess actual complexity |\n\n## Decision Framework\n\nBefore escalating, answer these questions:\n\n### 1. Have I understood the problem?\n\n- [ ] Can I articulate why the current model is insufficient?\n- [ ] Have I identified what specific reasoning capability is missing?\n- [ ] Is this a capability gap or a knowledge gap?\n\n**If knowledge gap:** Gather more information, don't escalate.\n\n### 2. Have I investigated systematically?\n\n- [ ] Did I read error messages/outputs carefully?\n- [ ] Did I check for similar solved problems?\n- [ ] Did I form and test a hypothesis?\n\n**If not investigated:** Complete investigation first.\n\n### 3. Is escalation the right solution?\n\n- [ ] Would a different approach work at current model level?\n- [ ] Is the task inherently complex, or am I making it complex?\n- [ ] Would breaking the task into smaller pieces help?\n\n**If decomposable:** Break down, don't escalate.\n\n### 4. Can I justify the trade-off?\n\n- [ ] What's the cost (latency, tokens, money) of escalation?\n- [ ] What's the benefit (accuracy, safety, completeness)?\n- [ ] Is the benefit proportional to the cost?\n\n**If not proportional:** Don't escalate.\n\n## Escalation Protocol\n\nWhen escalation IS justified:\n\n1. **Document the reason** - State why current model is insufficient\n2. **Specify the scope** - What specific subtask needs higher capability?\n3. **Define success** - How will you know the escalated task succeeded?\n4. **Return promptly** - Drop back to efficient model after reasoning task\n\n## Common Rationalizations\n\n| Excuse | Reality |\n|--------|---------|\n| \"This is complex\" | Complex for whom? Have you tried? |\n| \"Better safe than sorry\" | Safety theater wastes resources |\n| \"I tried and failed\" | How many times? Did you investigate why? |\n| \"The user expects quality\" | Quality comes from process, not model size |\n| \"Just this once\" | Exceptions become habits |\n| \"Time is money\" | Systematic approach is faster than thrashing |\n\n## Agent Schema\n\nAgents can declare escalation hints in frontmatter:\n\n```yaml\nmodel: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n**Key points:**\n- Hints are advisory, not mandatory\n- Orchestrator has final authority\n- Orchestrator can escalate without hints (broader context)\n- Orchestrator can ignore hints (task is actually simple)\n\n## Orchestrator Authority\n\nThe orchestrator (typically Opus) makes final escalation decisions:\n\n**Can follow hints:** When hint matches observed conditions\n**Can override to escalate:** When context demands it (even without hints)\n**Can override to stay:** When task is simpler than hints suggest\n**Can escalate beyond hint:** Go to opus even if hint says sonnet\n\nThe orchestrator's judgment, informed by conversation context, supersedes static hints.\n\n## Red Flags - STOP and Investigate\n\nIf you catch yourself thinking:\n- \"Let me try with a better model\"\n- \"This should be simple but isn't working\"\n- \"I've tried everything\" (but haven't investigated why)\n- \"The smarter model will know what to do\"\n- \"I don't understand why this isn't working\"\n\n**ALL of these mean: STOP. Investigate first.**\n\n## Integration with Agent Workflow\n\n```\n**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first.\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\n## Quick Reference\n\n| Situation | Action |\n|-----------|--------|\n| Task inherently requires nuanced reasoning | Escalate |\n| Agent uncertain but hasn't investigated | Investigate first |\n| Multiple attempts failed | Question approach, not model |\n| Security/high-stakes decision | Escalate |\n| \"Maybe smarter model knows\" | Never escalate on this basis |\n| Hint fires, task is actually simple | Override, stay at current model |\n| No hint fires, task is actually complex | Override, escalate |\n\n## Model Capability Notes\n\n**MCP Tool Search (Claude Code 2.1.7+)**: Haiku models do not support MCP tool search. If a workflow uses many MCP tools (descriptions exceeding 10% of context), those tools load upfront on haiku instead of being deferred. This can consume significant context. Consider escalating to sonnet for MCP-heavy workflows or ensure haiku agents use only native tools (Read, Write, Bash, etc.).\n\n**Claude.ai MCP Connectors (Claude Code 2.1.46+)**: Users with claude.ai connectors configured may have additional MCP tools auto-loaded, increasing the total tool description footprint. This makes it more likely that haiku agents will exceed the 10% tool search threshold. When escalation decisions involve MCP-heavy workflows, factor in claude.ai connector tool count via `/mcp`.\n\n**Effort Controls as Escalation Alternative (Opus 4.6 / Claude Code 2.1.32+)**: Opus 4.6 introduces adaptive thinking with effort levels (`low`, `medium`, `high`). The `max` level was removed in 2.1.72; `high` is now the ceiling. Symbols: ○ (low) ◐ (medium) ● (high). Use `/effort auto` to reset to default. Before escalating between models, consider whether adjusting effort on the current model would suffice:\n\n| Instead of... | Consider... | When |\n|--------------|-------------|------|\n| Haiku → Sonnet | Stay on Haiku | Task is still deterministic, just needs more context |\n| Sonnet → Opus | Opus@medium | Moderate reasoning, not deep architectural analysis |\n| Opus@medium → \"maybe try again\" | Opus@high or \"ultrathink\" | Genuine complexity needing deeper reasoning |\n\n**Default effort change (2.1.68+)**: Opus 4.6 now\ndefaults to **medium effort** for Max and Team\nsubscribers. Use `/model` to change effort level, or\ntype \"ultrathink\" in your prompt to enable high effort\nfor the next turn.\n\n**Opus 4/4.1 removed (2.1.68+)**: Opus 4 and 4.1 are\nno longer available on the first-party API. Users with\nthese models pinned are automatically migrated to\nOpus 4.6. No action needed for agents using `model`\nfrontmatter, as the migration is transparent.\n\n**Sonnet 4.5 → 4.6 migration (2.1.69+)**: Sonnet 4.5\nusers on Pro/Max/Team Premium are automatically migrated\nto Sonnet 4.6. Agent model frontmatter referencing\nSonnet resolves transparently. The `--model` flags for\n`claude-opus-4-0` and `claude-opus-4-1` now correctly\nresolve to Opus 4.6 instead of deprecated versions.\n\n**Effort parameter fix (2.1.70+)**: Fixed API 400 error\n`This model does not support the effort parameter` when\nusing custom Bedrock inference profiles or non-standard\nClaude model identifiers. Effort controls now work\nreliably across all deployment configurations.\n\n**Default Opus 4.6 on providers (2.1.73+)**: Bedrock,\nVertex, and Microsoft Foundry now default to Opus 4.6\n(was Opus 4.1). Subagent `model: opus`/`sonnet`/`haiku`\naliases now resolve to the current version on all\nproviders; previously they were silently downgraded to\nolder versions (e.g., Opus 4.1 instead of 4.6). This\nfix means agent dispatch workflows on third-party\nproviders now match first-party API behavior.\n\n**`modelOverrides` setting (2.1.73+)**: Maps model\npicker entries to provider-specific IDs (Bedrock\ninference profile ARNs, Vertex version names, Foundry\ndeployment names). Use when routing model selections to\nspecific inference profiles. See the model optimization\nguide for configuration details.\n\n**`/output-style` deprecated (2.1.73+)**: Use `/config`\ninstead. Output style is now fixed at session start for\nbetter prompt caching.\n\n**Full model IDs in agent frontmatter (2.1.74+)**: Agent\n`model:` fields now accept full model IDs (e.g.,\n`claude-opus-4-6`) in addition to aliases (`opus`,\n`sonnet`, `haiku`). Previously, full IDs were silently\nignored. Agents now accept the same values as `--model`.\n\nEffort controls do NOT replace the escalation governance\nframework: they provide an additional axis. The Iron Law\nstill applies: investigate before changing either model\nor effort level.\n\n## Troubleshooting\n\n### Common Issues\n\n**Command not found**\nEnsure all dependencies are installed and in PATH\n\n**Permission errors**\nCheck file permissions and run with appropriate privileges\n\n**Unexpected behavior**\nEnable verbose logging with `--verbose` flag\n\nFile v1.8.5:_meta.json\n\n{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.8.5\",\n  \"publishedAt\": 1778292882463\n}\n\nFile v1.8.5:skill-card.md\n\n## Description: <br>\nAssess whether to escalate models. <br>\n\nThis skill is ready for commercial/non-commercial use. <br","readmeExcerpt":"Skill: escalation-governance Owner: athola Summary: Assess whether to escalate models Tags: latest:1.9.19 Version history: v1.9.19 | 2026-08-26T13:02:37.749Z | user Release v1.9.19 v1.9.18 | 2026-08-15T21:27:20.119Z | user Release v1.9.18 v1.9.17 | 2026-07-30T05:27:15.062Z | user Release v1.9.17 v1.9.16 | 2026-07-14T19:44:12.602Z | user Release v1.9.16 v1.9.15 | 2026-07-04T21:18:53.554Z | user Release v1.9.15 v1.9.14","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"NO ESCALATION WITHOUT INVESTIGATION FIRST"},{"language":"yaml","snippet":"model: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly"},{"language":"text","snippet":"**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first."},{"language":"text","snippet":"NO ESCALATION WITHOUT INVESTIGATION FIRST"},{"language":"yaml","snippet":"model: haiku\nescalation:\n  to: sonnet                 # Suggested escalation target\n  hints:                     # Advisory triggers (orchestrator may override)\n    - security_sensitive     # Touches auth, secrets, permissions\n    - ambiguous_input        # Multiple valid interpretations\n    - novel_pattern          # No existing patterns apply\n    - high_stakes            # Error would be costly"},{"language":"text","snippet":"**Verification:** Run the command with `--help` flag to verify availability.\nAgent starts task at assigned model\n├── Task succeeds → Complete\n└── Task struggles →\n    ├── Investigate systematically\n    │   ├── Root cause found → Fix at current model\n    │   └── Genuine capability gap → Escalate with justification\n    └── Don't investigate → WRONG PATH\n        └── \"Maybe escalate?\" → NO. Investigate first."}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: escalation-governance\ndescription: Assess whether to escalate models\nversion: 1.9.8\ntriggers:\n  - escalation\n  - model-selection\n  - governance\n  - agents\n  - orchestration\n  - evaluating reasoning depth\nmetadata: {\"openclaw\": {\"homepage\": \"https://github.com/athola/claude-night-market/tree/master/plugins/abstract\", \"emoji\": \"\\ud83e\\udd9e\"}}\nsource: claude-night-market\nsource_plugin: abstract\n---\n\n> **Night Market Skill** — ported from [claude-night-market/abstract](https://github.com/athola/claude-night-market/tree/master/plugins/abstract). For the full experience with agents, hooks, and commands, install the Claude Code plugin.\n\n\n## Table of Contents\n\n- [Overview](#overview)\n- [The Iron Law](#the-iron-law)\n- [When to Escalate](#when-to-escalate)\n- [When NOT to Escalate](#when-not-to-escalate)\n- [Decision Framework](#decision-framework)\n- [1. Have I understood the problem?](#1-have-i-understood-the-problem)\n- [2. Have I investigated systematically?](#2-have-i-investigated-systematically)\n- [3. Is escalation the right solution?](#3-is-escalation-the-right-solution)\n- [4. Can I justify the trade-off?](#4-can-i-justify-the-trade-off)\n- [Escalation Protocol](#escalation-protocol)\n- [Common Rationalizations](#common-rationalizations)\n- [Agent Schema](#agent-schema)\n- [Orchestrator Authority](#orchestrator-authority)\n- [Red Flags - STOP and Investigate](#red-flags-stop-and-investigate)\n- [Integration with Agent Workflow](#integration-with-agent-workflow)\n- [Quick Reference](#quick-reference)\n\n\n# Escalation Governance\n\n## Overview\n\nModel escalation (haiku→sonnet→opus) trades speed/cost for reasoning capability. This trade-off must be justified.\n\n**Core principle:** Escalation is for tasks that genuinely require deeper reasoning, not for \"maybe a smarter model will figure it out.\"\n\n## The Iron Law\n\n```\nNO ESCALATION WITHOUT INVESTIGATION FIRST\n```\n**Verification:** Run the command with `--help` flag to verify availability.\n\nEscalation is never a shortcut. If you haven't understood why the current model is insufficient, escalation is premature.\n\n## When to Escalate\n\n**Legitimate escalation triggers:**\n\n| Trigger | Description | Example |\n|---------|-------------|---------|\n| Genuine complexity | Task inherently requires nuanced judgment | Security policy trade-offs |\n| Reasoning depth | Multiple inference steps with uncertainty | Architecture decisions |\n| Novel patterns | No existing patterns apply | First-of-kind implementation |\n| High stakes | Error cost justifies capability investment | Production deployment |\n| Ambiguity resolution | Multiple valid interpretations need weighing | Spec clarification |\n\n## When NOT to Escalate\n\n**Illegitimate escalation triggers:**\n\n| Anti-Pattern | Why It's Wrong | What to Do Instead |\n|--------------|----------------|---------------------|\n| \"Maybe smarter model will figure it out\" | This is thrashing | Investigate root cause |\n| Multiple failed attempts | Suggests wrong approach, not insufficient capabil"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn7d107jg9jv602h9ytsegydq184a42s\",\n  \"slug\": \"nm-abstract-escalation-governance\",\n  \"version\": \"1.9.19\",\n  \"publishedAt\": 1787749357749\n}"},{"path":"skill-card.md","content":"## Description:\n\nAssess whether to escalate models.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[athola](https://clawhub.ai/user/athola)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nDevelopers and agent operators use this skill to decide when model escalation is justified, document the reason, scope the escalated work, and return to a more efficient model after the reasoning task is complete.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Broad triggers may cause the skill to appear in general agent or orchestration discussions.\n\nMitigation: Apply the guidance only when a task involves model escalation, model selection, governance, agent orchestration, or reasoning-depth evaluation.\n\nRisk: Claude-version notes in the artifact may become stale.\n\nMitigation: Check current product documentation before relying on version-specific operational guidance.\n\n## Reference(s):\n\n- [ClawHub Skill Page](https://clawhub.ai/athola/skills/nm-abstract-escalation-governance)\n- [OpenClaw Homepage](https://github.com/athola/claude-night-market/tree/master/plugins/abstract)\n\n## Skill Output:\n\n**Output Type(s):** [Guidance, Markdown]\n\n**Output Format:** [Markdown guidance with decision checklists and escalation criteria]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [No code execution; Markdown-only guidance.]\n\n## Skill Version(s):\n\n1.9.19 (source: server release evidence)\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."}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Assess whether to escalate models Skill: escalation-governance Owner: athola Summary: Assess whether to escalate models Tags: latest:1.9.19 Version history: v1.9.19 | 2026-08-26T13:02:37.749Z | user Release v1.9.19 v1.9.18 | 2026-08-15T21:27:20.119Z | user Release v1.9.18 v1.9.17 | 2026-07-30T05:27:15.062Z | user Release v1.9.17 v1.9.16 | 2026-07-14T19:44:12.602Z | user Release v1.9.16 v1.9.15 | 2026-07-04T21:18:53.554Z | user Release v1.9.15 v1.9.14","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":921,"uniquenessScore":57,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-10T05:08:29.035Z","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":"2026-10-10T05:08:29.035Z","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":"2026-10-10T06:06:18.532Z","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"}]}}}