Agent Docs System Map

Agent docs are portable pointers into the canonical brain, workflows, and skills. They are not another behavior or knowledge layer.

Surface Model

Surface Path Ownership
Repository boot pointers AGENTS.md, nearest SKILL.md Stable local rules and links to deeper owners.
Portable quick references wiki/agent-docs/*.md Small setup and validation cards that point to canonical sources.
Brain graph brain/*.json, packages/brain-runtime/ Goals, loops, durable objects, source bindings, proof, and queries.
Workflow registry wiki/workflows/*.manifest.yaml Typed, callable procedures and capability requirements.
Skills and commands skills/**/SKILL.md, skills/.runtime/all, Slash Command Index Executable agent procedures and retrieval handles.
Capability routing Capability Routing Map, Skill Resolver, Active Stack (What to Actually Use) Chooses the narrowest skill, tool, API, or wiki owner.
Review ledgers wiki/meta/*-audit.json Stateful source and consolidation decisions.

Page Contract

Every hand-written agent-doc page should state:

  1. The canonical owner.
  2. When an agent should open the quick reference.
  3. The portable action or checklist it provides.
  4. The proof command that keeps it current.

Do not let a quick reference become a private fork of source truth. Learned facts and behavior changes belong in a wiki owner, brain object, workflow manifest, skill, code, or configuration.

Closeout

For quick-reference changes, rebuild the wiki index, update/embed qmd, and append wiki/log.md. If the change affects routing or behavior, also validate the brain runtime, workflow registry, relevant skills, and routing doctor.


Timeline