菜单
Load the canonical auto-harness governance skill from platform/skills before changing harness governance, validators, trust tiers, companion rules, or repository operating doctrine.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | harness-governance |
| description | Load the canonical auto-harness governance skill from platform/skills before changing harness governance, validators, trust tiers, companion rules, or repository operating doctrine. |
This Claude-native skill entry is an adapter, not the source of truth.
Before acting, read the canonical skill at:
platform/skills/harness-governance/SKILL.md
Then follow that file completely. Keep durable governance instructions in
platform/skills/, HARNESS.md, AGENTS.md, and the active modules declared in
harness.manifest.yaml. Do not maintain an independent copy of harness doctrine
inside .claude/.
Use when onboarding any existing project — codebase or documentation-only — into the development harness. Produces a governance inventory, repository inventory, module composition recommendation, artifact gap analysis, integration/translation guidance, validator runbook, and a copy-paste lite manifest. Activate with any Agent Skills-compatible client pointed at the target repository.
Trust tier enforcement, lifecycle controls, and companion rule requirements for projects using the development harness. Use when checking if an action requires human approval, creating or validating documentation artifacts, verifying lifecycle stage conditions, running validators, or making changes to governance entrypoints (HARNESS.md, AGENTS.md, CLAUDE.md, CI workflows).
Use for tasks involving digital twins, simulation, scenarios, world state, run logs, model/agent registries, data provenance, geospatial/city/infrastructure models, AI-datacenter or operational twins, or scenario manifests. Enforces maturity honesty, world/scenario/run separation, provenance, uncertainty disclosure, and publication boundaries for projects governed by the management/digital-twin overlay.
Use when a harness-governed project ships or designs an MCP (Model Context Protocol) server, or when an agent reasons about exposing harness governance to runtime harnesses via MCP. Covers the three MCP modes (consumer, producer, exposed-governance), per-tool consumer-tier mapping for exposed tools, prompt-injection defense surface, capability and transport posture, and the explicit boundary against the harness-tools skill (which governs consumed tools) and the harness-governance skill (which governs trust tiers and lifecycle).
Testing governance and practice guidance for projects using the development harness testing-standard module. Use when writing tests, analyzing coverage, choosing a testing strategy, configuring test frameworks, or evaluating whether a PR meets the project's declared coverage thresholds.
Use when invoking MCP developer tools (Linear, Slack, Google Calendar, Gmail, Canva, Ahrefs, Similarweb) in a harness-governed project. Covers trust tier requirements per tool, Linear as the artifact backing store for harness governance docs, Slack as the companion-rule notification surface, and analytics tools under read-only tier policy.