| name | feam-agent-context-maintainer |
| description | Use when AGENTS.md is missing, stale, contradictory, or needs self-healing; when repo structure, Cargo workspace layout, CI, commands, tests, or source-of-truth docs change; or when asked to update Feather Mesh AI-SDLC, agent, onboarding, or project skill context. |
Feather Mesh Agent Context Maintainer
Use this skill to keep project agent context small, accurate, and durable.
Sources To Read
Read only the high-signal files needed for the drift being investigated:
AGENTS.md
README.md
feather-mesh/README.md
feather-mesh/Cargo.toml
.github/workflows/rust.yml
- Changed files relevant to the current task
Do not load long project artifacts, PDFs, generated outputs, or the Python MVP unless the task specifically targets them.
Self-Healing Rules
- Patch
AGENTS.md when it is missing, stale, or contradicted by durable repo facts.
- Keep
AGENTS.md as a routing and decision guide, not a duplicate README.
- Prefer durable facts: source-of-truth directories, crate boundaries, validation commands, tested contracts, and ownership rules.
- Remove stale statements instead of adding caveats around them.
- Keep the file concise; target under 120 lines.
- Do not include temporary branch status, implementation plans, sprint notes, or speculative roadmap.
- Keep Python MVP context explicitly secondary unless it becomes the active implementation source of truth.
Drift Check
When useful, run:
.codex/skills/feam-agent-context-maintainer/scripts/check_agents_context.sh
Treat the script as a smoke check. Passing does not prove the context is complete; it only verifies the most important anchors exist.
Validation
After updating context, verify that a future agent could answer:
- Where should a CLI behavior change go?
- What tests should run after service logic changes?
- Is
python_mvp/ the source of truth?
- Where are stable CLI exit codes documented and tested?