| name | eagle-governance |
| description | Install hook-based engineering governance for agent-led projects. Use when the user says: 'govern this project', 'install governance hooks', 'agent governance', 'context bloat guard', 'hook-based engineering governance', 'stop agents ignoring instructions', or asks to make AGENTS.md / CLAUDE.md enforceable through hooks.
|
Eagle Governance
Install a project-local governance runtime that keeps instruction files small and lets hooks enforce the important rules.
What this does
- Adds a small governance policy file:
.eagle-governance.json
- Installs hook scripts for Claude Code and Codex when those targets are selected
- Adds small managed routing blocks to
AGENTS.md and/or CLAUDE.md
- Configures Grok Build through its Claude Code compatibility layer and Antigravity through
.agents/hooks.json
- Verifies hook scripts, JSON wiring, and instruction shims
- Partners with Eagle Mem when available for durable handoff mirroring and feature-verification visibility
- Ships an optional Eagle Eval governance scenario pack for regression checks
Commands
eagle-skills govern status --target all
eagle-skills govern apply --target all --mode warn
eagle-skills govern verify --target all
Use --dry-run before applying when you want to preview writes:
eagle-skills govern apply --target all --mode warn --dry-run
Default policy
- Default mode is
warn
- Hard blocks remain active for destructive commands, secret-file writes, outside-root writes, and high-risk context handoff
- Context bloat uses Claude's real context percentage when available, then Eagle Mem turn budget, then optional transcript thresholds. High risk produces
.eagle-governance/handoff.md and blocks further implementation
- Transcript size and changed-file count are warnings by default; they do not force handoff unless configured
- Eagle Mem bridge defaults to
auto: Governance works without Eagle Mem, but mirrors handoffs and surfaces pending feature verification when Eagle Mem is installed
- Eagle Eval is optional and never runs inside hooks; use the governance pack when you want regression tests for hook behavior
AGENTS.md and CLAUDE.md remain shims; do not duplicate the full policy there
Workflow
- Run
eagle-skills govern status --target all
- Run
eagle-skills govern apply --target all --mode warn --dry-run
- Run
eagle-skills govern apply --target all --mode warn
- Run
eagle-skills govern verify --target all
- For Codex, open
/hooks and trust the new project hooks when prompted
- For Claude Code, restart or resume the project session so new hooks and the context statusline load
- If Eagle Mem is installed, confirm
govern status reports the bridge as available
Output
Return a short PM-readable report:
Eagle Governance:
Policy: installed
Claude Code: hooks installed and wired
Codex: hooks installed and wired; trust review may be required
Eagle Mem: bridge available; handoff mirror enabled
Eagle Eval: governance pack available; CLI optional
Grok Build: Claude-compatible hooks installed and wired
Antigravity: hooks installed and wired
Principles
- Hooks are control; instruction files are routing context
- Keep injected context tiny
- Treat each provider's hook output contract separately
- Let Eagle Mem own durable recall, orchestration lanes, and release-boundary feature verification
- Let Eagle Eval test governance behavior outside the live hook path
- Prefer warn-first adoption, then tighten gates once the project trusts the workflow
- Never overwrite user-written
AGENTS.md, CLAUDE.md, or settings content
