AI Handoff¶
This page is the operational guide for AI agents working on documentation in this repository.
First steps for any docs task¶
Before editing docs:
- read
AGENTS.md - read
README.md - read
docs/planning/DOCS_ARCHITECTURE_PLAN.md - read
docs/planning/DOCS_PROGRESS.md - inspect the relevant source files before assuming behavior
Do not rely only on older docs text when the code is easy to inspect directly.
Claiming work¶
When starting a docs task:
- move the relevant work package in
DOCS_PROGRESS.mdtoin_progress - claim the specific page rows if the work is page-level
- keep ownership explicit so handoffs stay readable
Do not silently start work without updating the tracker first.
While writing¶
- keep the current page tree and navigation model intact unless the architecture plan is intentionally changed
- write task-first, concise content
- document current behavior, including caveats
- prefer small runnable examples over abstract pseudo-code when examples help
If you discover implementation issues while documenting:
- document the current behavior accurately
- mention the caveat in
DOCS_PROGRESS.mdif it matters for future docs work - do not silently “correct” documentation to match intended behavior if the code does something else
Before finishing¶
Every docs agent should:
- update
DOCS_PROGRESS.md - move statuses forward only for actually completed work
- append a new handoff entry
- list blockers and follow-up recommendations explicitly
- run a verification pass appropriate to the environment
Verification expectations¶
Preferred checks:
uv run mkdocs build --strict- targeted tests or examples when docs depend on runtime behavior
- file-level checks for placeholder text, broken naming, or nav drift
If a full docs build is not possible in the current environment, say so in the handoff notes.
Handoff note quality¶
A good handoff entry should say:
- what was completed
- which files changed
- which work package statuses changed
- any blockers or caveats
- the recommended next task
Keep entries short, factual, and easy for the next agent to act on.