groundwork:recall

Rebuild the picture of where a piece of work stands, from what the repository already records.

This skill writes nothing, deliberately. groundwork carried a hand-maintained state file once and removed it in 0.2.6, because a file that has to be kept current is one more thing to forget, and a stale state file is worse than none - it is trusted. Everything below is already being written for other reasons. Reading it costs nothing and cannot go stale in a way the underlying artifacts have not already gone stale.

What to read, and what each one tells you

• prd.md - what this feature is for, and its acceptance criteria. The destination.
• adr/, feature-scoped and project-wide - what was already decided and must not be re-opened. Check for Superseded entries: a reversal is exactly the thing someone returning after a break is most likely to miss.
• the tracker, or tasks.md - which slices are done, open, blocked, or needs-proof. This is the closest thing to a progress bar. A needs-proof slice is built work waiting on evidence, which is a very different resumption than an unstarted one.
• map.md and tickets.md - present when survey ran at map speed. The map's "Not yet specified" section is the live frontier: it says what is still fog.
• git log on the feature's branches, and any open PRs - what was actually done last, as opposed to what was planned.
• quicklog.md - small things done outside the main flow, which are the easiest to forget and the most likely to surprise you.

Report

Give it in this order, because it is the order someone resuming needs:

1. Where this was going - one or two sentences from the PRD, not a summary of the whole file.
2. What is settled - accepted ADRs, in one line each. Flag any supersession.
3. What is done - closed slices, briefly.
4. What is in flight - open slices, needs-proof slices, open PRs, uncommitted work on a branch. Be specific about which of these it is; they need different next actions.
5. What is still open - blocked slices, the map's remaining fog, unanswered questions.
6. The obvious next step, if there is one, and say plainly when there isn't.

Separate what the artifacts state from what you are inferring. "The last commit touched the parser" is a fact; "you were probably mid-refactor" is a guess, and should be labelled as one.

If the artifacts genuinely do not say where things stood, say that. The honest answer is sometimes that the work was left in a state nobody recorded, and the fix for that is finishing this session's work through the normal flow, not reconstructing a story.