Files
Magnus HedemarkGitHubfactory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
0223e2bc18 fix: repair dead cross-skill routing references in methodology skills (#251)
Route methodology skills to real catalog skills instead of dead role names:
- platform-engineering: docker-management -> docker-compose
- backend-engineering: technical-architect -> api-design-and-evolution,
  reviewer -> qa-methodology
- frontend-engineering: ux-designer -> product-design-and-ux,
  technical-architect -> api-design-and-evolution, writer -> neutral,
  reviewer -> qa-methodology
- security-audit-methodology: reviewer -> qa-methodology,
  debugger -> systematic-debugging
- data-engineering: UX designer -> product-design-and-ux
- technical-documentation: writer -> neutral, technical architect ->
  software-architecture-analysis, debugger -> systematic-debugging,
  reviewer -> qa-methodology
- operational-design: executive-methodology -> strategy-frameworks
- technical-documentation/references/agent-facing-docs.md: drop dead
  deployment-methodology skill_view example

Also bring modified skills' descriptions up to the quality validator's
imperative-verb + negative-boundary requirement and regenerate catalogs.

agent-council's external hermes-profiles list is retained unchanged.

Closes #234

Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
2026-08-03 13:57:25 -04:00

1.3 KiB

Agent-Facing Documentation

AGENTS.md Structure

Agent-facing docs have different requirements than human-facing docs. Agents read the entire file sequentially, so structure matters differently.

Required Sections

# Project Name — Agent Guidance

## Trigger Patterns

| User Says | What It Means |
|---|---|
| "Deploy this service" | Full deployment: build → configure → deploy → verify |
| "Check service status" | Status check with health endpoint verification |

## Loading Order

Specify which skills to load and in what order. Skills should be loaded explicitly,
not assumed. For example: load `artifact-pyramids` for the output format, then
`platform-engineering` (or `release-engineering`) for the deployment methodology.

## Output Contract

Be explicit about what the agent produces and how to interpret it. If using
artifact pyramids, say so.

Skill Documentation

When documenting a skill for agent consumption:

Element Required Purpose
Trigger conditions Yes When does an agent know to load this skill?
Loading order Yes What to load first, what depends on what
Reference files table Yes What each reference is for and when to load it
Output contract Yes What format does the agent produce?
Pitfalls Recommended What commonly goes wrong?