Skip to content

One record per outcome

Use .ai/work/<slug>.md for larger local work without a board-owned assignment. For customer delivery on a GitHub Project, the assigned issue owns the live plan; follow the operating model and the project's working agreement. The dashboard below manages files; Project updates use your host's authorized tools.

Keep acceptance criteria, plan, tasks, feedback, decisions and evidence together. Small fixes need no record. Avoid separate proposal, spreadsheet and decision logs unless the project already requires them.

python3 tools/ai_toolkit.py work --workspace /path/to/project new customer-portal \
  --title "Ship the first customer portal flow" --kind software
python3 tools/ai_toolkit.py work --workspace /path/to/project feedback customer-portal \
  --text "Keep the existing sign-in flow."
python3 tools/ai_toolkit.py work --workspace /path/to/project task customer-portal T1 \
  --status done --evidence "Acceptance scenario verified with the relevant test."
python3 tools/ai_toolkit.py work --workspace /path/to/project csv customer-portal
python3 tools/ai_toolkit.py work --workspace /path/to/project archive customer-portal

Add task rows in your editor with stable T1, T2… IDs and statuses open, in_progress, blocked, done. Preserve the four-column table. The API updates only the selected row or appends feedback with a stable F1 ID. Completion needs evidence. Archiving requires all parsed tasks complete; check acceptance criteria and feedback too.

serve --workspace /path/to/project opens the local dashboard. Markdown is displayed as text. HTTP writes require the current ETag; in-process writes are serialized and the file is rechecked before replacement. This handles normal stale updates but is not a distributed transaction system for uncoordinated editors or OneDrive.

Decide per repository whether to commit work records. This toolkit ignores them by default; explicitly track sanitized records if useful. Private notes never belong in public repositories. Extract only reusable conclusions into existing documentation or the private PARA vault. Archive completed records and load history on demand.

Canonical source: toolkit/WORK-RECORDS.md.