Skip to content

Latest commit

 

History

History
65 lines (47 loc) · 2.09 KB

File metadata and controls

65 lines (47 loc) · 2.09 KB

AGENTS.md

Applies to

This card applies to .agents/ and all descendants unless a nearer AGENTS.md narrows the path.

Role

.agents/ holds agent-facing companion lanes for aoa-techniques.

It can expose practice-canon help to local agents, but it does not author technique meaning, skill workflow meaning, proof authority, route dispatch, or runtime state.

Read before editing

Read root AGENTS.md, DESIGN.AGENTS.md, docs/guardrails/AGENTS_MESH_PROTOCOL.md, and the nearest lane card before changing files here.

For Codex Spark work, also read .agents/spark/README.md, .agents/spark/registry.json, and the chosen scenario README.md plus PROMPT.md.

For technique-facing help, also read techniques/AGENTS.md, TECHNIQUE_INDEX.md, and the selected source TECHNIQUE.md bundle.

Boundaries

  • Do not create .agents/skills/ as a local cache of shared AoA bundles.
  • Do not create an empty top-level skills/ home. A future repository-owned bundle must first pass the admission boundary in root AGENTS.md.
  • Use host or user-profile projections for shared skills; use authored technique routes and source-returning derived readers for technique help.
  • Do not hand-edit generated companion content as the first move.
  • Do not make agent lanes stronger than the source technique, docs, schemas, builders, or owning AoA repository.
  • Do not add private project assumptions, secrets, local host paths, or hidden capability claims.

Validation

Run the smallest covering checks:

python scripts/validate_agents_md_shape.py
python scripts/validate_agents_mesh.py
python .agents/spark/scripts/validate_spark_lane.py
python -m unittest discover -s .agents/spark/tests -p 'test*.py'
python scripts/validate_semantic_agents.py
python scripts/validate_repo.py

For broad agent-lane or generated-surface changes, run:

python scripts/release_check.py

Closeout

Report changed agent lanes, source technique or builder surfaces consulted, generated or exported companions rebuilt or left untouched, validation run, validation skipped, and any remaining migration-card risk.