Miki wrap up

miki-wrap-up

What this does

Reviews the conversation and working tree for outstanding work before the session ends. Produces a structured punch list, or confirms nothing's left. Recommends, never acts — if something needs doing, the user (or the next turn) does it.

The sweep

Six categories. For each: a short check, then a one-line conclusion. Skip empty categories silently — they only appear when they have content.

1. Task list

TaskList items not in completed status. Distinguish in_progress (mid-flight) from pending (queued). Skip if empty.

2. Working tree

git status --porcelain and git status -sb. Surface uncommitted changes related to this session, untracked in-flight files (not random OS cruft), and ahead/behind counts vs. remote. Skip if clean and in sync.

3. PRs touched this session

For each PR mentioned in the conversation, gh pr view <n> --json number,title,state,mergeable,mergeStateStatus. Surface anything still actionable:

• OPEN + CONFLICTING → user action needed.
• OPEN + BLOCKED or awaiting approval → waiting on review.
• OPEN + MERGEABLE without --auto → could enable auto-merge.

PRs that merged or closed during the session aren't loose ends.

4. Background agents and processes

Any Agent runs or background Bash invocations still running — surface with description and runtime. The user may want to wait, kill, or follow up.

5. Flagged follow-ups in the conversation

Re-read the conversation for things either party flagged as "out of scope for this PR", "follow-up", "later", "I'll do that next", "separate PR", etc. — that were not subsequently closed. List as bullets with enough context (what + pointer to the exchange). Skip if none.

6. Stale references noticed but not fixed

CLAUDE.md describing outdated workflow, TODO.md entries that should have been closed, comments pointing at moved code, docs lagging code — only if the conversation surfaced them. Skip if none.

Output format

Lead with a headline:

• "Nothing outstanding." — every category empty.
• "N loose end(s):" — N is the total across fired categories.

Then only the fired categories. One line per entry where possible.

## Wrap-up sweep

<headline>

### <Category that fired>
- <entry>
- <entry>

### <Next category>
- ...

If nothing's outstanding, the whole response is one sentence. Don't fill the page with empty-category stubs — that defeats the point of a quick check.

Operating rules

• Never act. No commits, pushes, edits, killing agents, merging PRs. Report only.
• Quick over thorough. A wrap-up check, not an audit. If a category would take more than a few seconds (e.g. ten PRs to fetch), summarize what'd need to be checked rather than checking all of it.
• Conversation is authoritative for "flagged follow-ups". Don't grep TODO.md whole-repo — only conversation-raised items.
• One-shot. One report per invocation. No loops, no follow-up questions.