agentleFS
etretr/groundwork/skills/finalize-task/SKILL.md

finalize-task

Skill45 starsChanged 4 days agoMITView on GitHub

Use when a task worktree has passed validation - manually commits, merges, and cleans up, or prepares runner-owned commits and publication

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 10 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    When invoked by the external runner, treat its exact project_root, worktree_path, branch, base_branch, base_head, and validated_head values …

    line 16
  2. Distinctive:

    When invoked manually from the task worktree:

    line 18
  3. Distinctive:

    Never reset, stash, rebase, force-delete, or silently switch either worktree to another branch.

    line 37
  4. Distinctive:

    Update its status-table row when an index exists.

    line 44
  5. Distinctive:

    Leave unrelated tasks unchanged.

    line 45
  6. Distinctive:

    Do not create an empty commit.

    line 54
  7. Distinctive:

    Check whether the current base head is an ancestor of the task branch.

    line 61
  8. Distinctive:

    If it is not:

    line 63
  9. Very specific:

    If resolution needs product, architecture, or ownership judgment, safely abort the integration when possible and return RESULT: NEEDS_INPUT.

    line 71
  10. Very specific:

    When invoked manually, merge with --no-ff, then continue to cleanup below.

    line 81
  11. Distinctive:

    After a successful manual merge:

    line 85
  12. Distinctive:

    Remove exactly the verified task worktree.

    line 89
  13. Distinctive:

    Delete exactly the verified task branch with safe deletion.

    line 90
  14. Very specific:

    If the merge succeeded but cleanup fails, return RESULT: CLEANUP_REQUIRED.

    line 92

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Finalize Task

Finalize one validated task. Use judgment for commit and merge messages. Preserve the worktree and branch whenever state is ambiguous.

Inputs

Accept a task ID and optional --project <name>.

When invoked by the external runner, treat its exact project_root, worktree_path, branch, base_branch, base_head, and validated_head values as authoritative. Missing or inconsistent runner inputs are a failure.

When invoked manually from the task worktree:

  1. Resolve --project directly from the repository's .groundwork.yml; do not depend on or change persisted project selection.
  2. Derive the current registered worktree and checked-out task branch.
  3. Derive the primary worktree and its checked-out base branch.
  4. Treat the current task HEAD as validated_head. If validation did not just pass for this exact tree and working state, stop and ask the user to run validate first. Exception — resume entry: when a resume gate (node ${CLAUDE_PLUGIN_ROOT}/lib/task-resume.js check <task-id>) classified this task as resume_at: finalize, its terminal validated session already covers this exact tree; proceed (see ${CLAUDE_PLUGIN_ROOT}/references/task-resume-protocol.md).

Workflow

1. Verify state

Before changing anything:

  1. Confirm the task path is a registered worktree for exactly the task branch.
  2. Confirm the primary worktree is on the base branch and clean.
  3. Confirm the task branch HEAD equals validated_head.
  4. Record the base branch's current head.
  5. Inspect staged, unstaged, and untracked task-worktree changes without discarding any.

Never reset, stash, rebase, force-delete, or silently switch either worktree to another branch.

2. Complete task bookkeeping

Within the selected project's task files:

  1. Change only this task's status to Complete.
  2. Update its status-table row when an index exists.
  3. Leave unrelated tasks unchanged.

3. Commit remaining validated work

Inspect the actual remaining diff. Include validation fixes and task bookkeeping belonging to this task.

  • In runner mode, validation has already been sealed by the runner. Only task-status bookkeeping may remain; reject any other changed path. Do not stage or commit it. Return it to the runner with action: "commit" and propose <task-id>: Mark task complete plus an explanatory body. Use action: "none" only when no bookkeeping change remains.
  • Outside runner mode, commit bookkeeping changes on the task branch as <task-id>: Mark task complete.
  • In manual mode, describe any validated remaining change in the subject.
  • Do not create an empty commit.
  • Require a clean task worktree afterward only in manual mode.

In runner mode, never ask a question. Return RESULT: FAILURE for missing information or unsafe state.

4. Integrate a moved base

Check whether the current base head is an ancestor of the task branch.

If it is not:

  1. Merge the current base head into the task branch with --no-ff --no-commit.
  2. Resolve conflicts only when the task definition and surrounding code make the resolution unambiguous.
  3. Record whether the merge command reported conflicts, even when every conflict was resolved unambiguously.
  4. In runner mode, do not run git add or create the merge commit. Leave the prepared merge for the runner and propose an expressive task-prefixed commit subject and body. Return RESULT: BASE_INTEGRATED, bind it to the exact integrated base head, and report conflicts_resolved: true only when the merge command reported conflicts. The runner owns the opt-in revalidation policy and invokes finalization again after sealing the integration.
  5. Outside runner mode, commit the resolution and continue to prepare and merge the task into the base branch.

If resolution needs product, architecture, or ownership judgment, safely abort the integration when possible and return RESULT: NEEDS_INPUT.

5. Prepare the merge

In runner mode:

  1. Reconfirm the primary worktree is clean, the task worktree contains only the permitted bookkeeping change, and the base branch is still at the head recorded in step 1.
  2. Choose a concise merge message containing the task ID and actual title or outcome.
  3. Return the versioned JSON RESULT: READY_TO_MERGE receipt with the exact runner token, task identity, bookkeeping commit proposal when needed, and expressive outward merge subject/body. Do not merge outward, remove the worktree, or delete the branch; the runner verifies and commits the prepared tree before performing those operations.

When invoked manually, merge with --no-ff, then continue to cleanup below. If an unexpected conflict occurs, abort it when safe and return RESULT: NEEDS_INPUT. Do not resolve it on the base branch because the resolution would not have been validated.

6. Verify and clean up

After a successful manual merge:

  1. Confirm the task branch is an ancestor of the base branch.
  2. Confirm the primary worktree is clean and record its merge commit.
  3. Remove exactly the verified task worktree.
  4. Delete exactly the verified task branch with safe deletion.

If the merge succeeded but cleanup fails, return RESULT: CLEANUP_REQUIRED; do not undo the merge.

Result contract

Make the final line exactly one of:

RESULT: FINALIZED | task_id=<id> | task_head=<sha> | merge_commit=<sha> | base_branch=<branch>
RESULT: NEEDS_INPUT | reason=<one-line reason>
RESULT: CLEANUP_REQUIRED | merge_commit=<sha> | worktree_path=<path> | branch=<branch> | reason=<one-line reason>
RESULT: FAILURE | reason=<one-line reason> | worktree_path=<path> | branch=<branch>

In runner mode, report only one of these compact JSON receipts or a failure result:

RESULT: READY_TO_MERGE | {"v":1,"token":"<exact-runner-token>","task_id":"TASK-NNN","phase":"finalize","action":"commit","commit":{"subject":"TASK-NNN: Mark task complete","body":"<bookkeeping summary>"},"merge":{"subject":"Merge TASK-NNN: <actual outcome>","body":"<publication summary>"}}
RESULT: BASE_INTEGRATED | {"v":1,"token":"<exact-runner-token>","task_id":"TASK-NNN","phase":"finalize","action":"commit","base_head":"<full-integrated-base-sha>","conflicts_resolved":true,"commit":{"subject":"TASK-NNN: Integrate updated base branch","body":"<resolution summary>"}}
RESULT: REVALIDATE | {"v":1,"token":"<exact-runner-token>","task_id":"TASK-NNN","phase":"finalize","action":"commit","base_head":"<full-integrated-base-sha>","reason":"<why revalidation is required>","commit":{"subject":"TASK-NNN: Integrate updated base branch","body":"<resolution summary>"}}

For READY_TO_MERGE, use action: "none" and omit commit when the task worktree is clean. BASE_INTEGRATED always uses action: "commit"; REVALIDATE remains accepted only for compatibility with older skill exports. Report FINALIZED only after a manual merge is present, both remaining worktrees are clean, and cleanup succeeded.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.