onboard

skills/groundwork/commands/onboard.md skills/groundwork/templates/briefing.template.md

Every other command produces context; this one consumes it and emits a task-specific subset. A fresh agent or teammate gets exactly the slice that applies to the task, and nothing more.

When to use it

  • Starting a new chat on a familiar repo.
  • Handing a task to a teammate.
  • Opening a project after weeks away.
  • Compacting a long, noisy session into a clean restart.

What it produces

One markdown file at .context/briefings/<YYYY-MM-DD>-<slug>.md, optionally pasted inline in chat. Sections:

  • What you are working on. Task, why, outcome.
  • What you need to know about this repo. Three to five sentences.
  • Conventions that apply to this area. Only rules that touch the files you will edit.
  • Decisions worth knowing. 3 to 5 ADRs with one-line summaries.
  • Recent activity. Last 5 to 10 commits in the area.
  • How to verify. Scoped fast and full commands.
  • Out of scope. Explicit. The leash on scope creep.
  • Tools available. Relevant MCP servers with HITL flags.

How it works

  1. Understand the task. One-sentence summary. If you cannot summarise, the command pushes back.
  2. Map to files. 3 to 10 paths before continuing.
  3. Score ADRs for relevance. Pick top 3 to 5. Never pad.
  4. Pull recent plans that touched the area. Skip the section if none apply.
  5. Subset the conventions. Prefer already-scoped files (.cursor/rules/<area>.mdc).
  6. Recent activity. git log --oneline --since="14 days ago" against the area.
  7. Verification. Pull from AGENTS.md, scoped form preferred.
  8. Out of scope. Two or three bullets. If you cannot name what is out, the command asks.
  9. Write. Aim 300 to 600 words. Over 800 means the briefing has failed at its job.