closeout

Use this before saying work is done, especially after edits, test runs, local servers, BEAM residents, or any cleanup request from the operator.

Before the final response, if you created actionable KG findings this session, revisit at most three of your own still-open bugs, questions or improvements. Use the UUID returned by this process's start_session (not a display name, predecessor or harness account), and the existing read:

knowledge(action="search", agent_id_filter="<this process UUID>",
          status="open", limit=10, include_details=false, response_mode="lean")

Open details for candidates you can assess. If addressed, use the existing update_finding with a closure class and evidence; fix_verified needs both deployment and a positive observation, so a merged PR alone is insufficient. Otherwise keep it open with a brief unresolved condition or artifact pointer when that adds information. Permanent insights, decisions and ordinary notes may legitimately stay open. This is a bounded reminder, not a quota or an extra check-in. If you wrote no actionable findings, the graph is unavailable, or the remaining claims cannot be verified now, continue closeout; do not mint an identity just for this reminder or automatically archive anything.

First run a non-mutating check:

• python3 scripts/dev/workspace_closeout.py
• python3 scripts/dev/workspace_closeout.py --branch-hygiene when branch or worktree cleanup is part of the question

This uses .unitares/workspace-closeout-baseline.json when present. The baseline is written by /governance-start through workspace_closeout.py --start-check; it lets the closeout check ignore resident/control-plane processes that already existed before the agent began work and focus on newly started repo-rooted processes.

Report:

• whether git is clean
• the delivery: line, including whether work is local-only, unpushed, pushed-but-not-proven-merged, or synced with default upstream
• any staged, unstaged, or untracked files
• any repo-rooted processes still running
• whether a remaining process is managed by a LaunchAgent label (macOS only; other platforms report no labels)
• the branch hygiene: line when requested, including cleanup candidates, safe deletions performed, held branches, and sweep errors

If work should be delivered instead of left local, stage the intentional files and use the ship helper:

• When the operator asks to ship, finish, deliver, open a PR, or otherwise complete a delivery workflow, treat the default draft-PR route as authorized; do not ask for a second confirmation just to push or open the draft PR.
• ./scripts/dev/ship.sh --plan "commit message" previews the delivery route
• ./scripts/dev/ship.sh "commit message" is the default route: a draft PR for every change, per docs/operations/github-workflow-conventions.md. Once CI is green and the review passes, mark it ready and apply approved-to-merge (the merge queue lands it; a governance-sensitive PR waits for the operator). Runtime/detached work mints a fresh agent-prefixed branch; non-runtime work on a named branch opens the draft PR on that branch
• ./scripts/dev/ship.sh --direct "commit message" opts out for docs/tests-only pushes where you knowingly skip the PR
• ./scripts/dev/ship.sh --auto-merge "commit message" opts into auto-merge-on-green; use only when the operator explicitly asks, not by default

If the operator asked to clean the workspace, or if you are finishing a task whose intended work is already committed/stashed, run:

• python3 scripts/dev/workspace_closeout.py --stash-dirty --stop-repo-processes --bootout-launch-agents --branch-hygiene-live

--bootout-launch-agents applies only on macOS LaunchAgent installs. --branch-hygiene-live reuses the branch-hygiene safety contract in docs/operations/branch-hygiene-runbook.md: patch-equivalent or empty stale branches may be pruned, clean stale worktrees may be removed, and branches with unique commits, dirty worktrees, or a protected checkout are held for review instead of deleted.