trellis-finish-work

name: trellis-finish-work description: "Finalize a completed task: verify gates passed, archive task, update workspace journal, update task index, summarize commits/PR, record follow-ups, and mark task done. Does NOT write code, fix bugs, bypass failed gates, or auto-push. Rejects if preconditions are not met."

Trellis Finish Work

Preconditions

ALL of the following must be satisfied:

1. Implementation is complete.
2. trellis-check has PASS.
3. Selected review gates have PASS.
4. finish.md records the explicit Finish Approval from the user.
5. Spec update decision is recorded in finish.md.
6. Observable outcomes are recorded in finish.md with verification evidence.
7. Delivery Sync Check is recorded in finish.md.
8. Guardrail Overrides review is recorded when runtime/guardrail-overrides.jsonl exists.
9. Merge review has PASS when required.
10. Code is committed, or PR is created, or explicitly no commit needed.
11. Build/Test/Smoke are recorded in validation/test-results.md, Ready for finish-work? is marked yes, and Overall PASS or skipped-with-valid-reason evidence is recorded.

Any missing precondition — finish-work refuses to execute.

Core Rules

What Finish-Work DOES

1. Verify all preconditions are met.
2. Run the team-kit archive wrapper.
3. Re-open the archived task and validate the archived artifacts.
4. Update the workspace journal.
5. Update the task index.
6. Summarize commits and PR.
7. Record follow-ups.
8. Mark task done.

What Finish-Work Does NOT Do

• Write code — implementation is done before finish-work runs.
• Fix bugs — bugs must be fixed before finish-work runs.
• Bypass failed gates — failed gates must pass before finish-work runs.
• Auto-push — push is always manual and requires explicit user action.
• Commit code — commits happen in Phase 3.2, before finish-work.
• Make spec update decisions — those happen in Phase 3.1, before finish-work.

Workflow

1. Verify preconditions — check each of the eleven preconditions listed above.
2. If any precondition fails — reject with a clear message stating what is missing.
3. If all preconditions pass — proceed with the following steps.

Step 1: Verify Preconditions

Implementation complete? [YES/NO]
trellis-check PASS? [YES/NO]
Review gates PASS? [YES/NO]
Finish Approval recorded? [YES/NO]
Spec update decision recorded? [YES/NO]
Observable outcomes recorded? [YES/NO]
Delivery Sync Check recorded? [YES/NO]
Guardrail Overrides reviewed when ledger exists? [YES/NO / N/A]
Merge review PASS when required? [YES/NO / N/A]
validate_review_gates.py PASS? [YES/NO]
validate_agent_results.py PASS? [YES/NO]
Code committed? [YES/NO / N/A with reason]
Build/Test/Smoke recorded with Ready yes and Overall PASS? [YES/NO / N/A with reason]

Any NO — stop and report what is missing.

Step 2: Archive Task

python3 ./.trellis/scripts/finalize_task_archive.py <task-dir>

Step 3: Validate Archived Artifacts

The archive wrapper must leave the archived task in a validator-clean state:

python3 ./.trellis/scripts/validate_task.py <archived-task-dir>
python3 ./.trellis/scripts/validate_review_gates.py <archived-task-dir>
python3 ./.trellis/scripts/validate_guardrail_overrides.py <archived-task-dir>
python3 ./.trellis/scripts/validate_agent_results.py <archived-task-dir>
python3 ./.trellis/scripts/validate_workflow_state.py <archived-task-dir>

Required post-archive checks:

• task.json still contains level
• implement.jsonl / check.jsonl still resolve after archive
• JSONL still contains only spec/research context, not duplicated task artifacts
• agent-results/*.json is present and valid when Execution Mode Decision selects single Trellis subagent, Trellis subagents, Trellis-native parallel + worktree, or OMC ulw/ultrawork + worktree + parent/child; if results are not required, any existing agent-results still validate cleanly
• each selected review gate still has a concrete review/*.md verdict, no unresolved template placeholders, and a matching PASS reviewer agent result
• journal / workspace index entries are updated with real commit information
• no .omc/state/* runtime state files remain in the tracked dirty set

If any validator fails, finish-work is not complete yet.

Step 4: Update Workspace Journal

Add an entry to the workspace journal summarizing:

• What was accomplished
• Key decisions made
• Specs updated (if any)
• Follow-ups recorded (if any)
• Real commits made for this task (do not leave placeholder text)

Step 5: Update Task Index

Ensure the task is marked as done in the task index.

Step 6: Summarize Commits/PR

List the commits made during this task:

[commit hash] [commit message]
[commit hash] [commit message]

If a PR was created, include the PR URL.

Step 7: Record Follow-ups

If any follow-up items were identified (out-of-scope items, future improvements, known limitations), record them.

Step 8: Mark Task Done

The task is now DONE. No further changes should be made to this task.

Output Format

finish.md (required shape)

# Finish: [Task Title]

## Finish Approval

Approval status:
- [x] approved

Approval source:
- user message: [exact user message entering Finish]
- timestamp: [timestamp]
- summary approved: [short summary]

Allowed to proceed with finish?
- [x] yes
- [ ] no

## Task Summary
[1-3 sentences describing what was accomplished]

## Observable Outcomes
- Outcome: [what someone can now see, do, or verify]
- Evidence: [command, test, screenshot, or manual flow]
- Remaining gap / risk: none

## Changed Files

| File | Change |
|------|--------|
| `path/to/file` | modified: [what changed] |

## Acceptance Criteria Coverage

| AC | Status | Evidence |
|----|--------|----------|
| AC1: [description] | PASS / FAIL / PARTIAL | [how verified] |

## Delivery Sync Check

- [x] README / user docs reviewed
- [x] Example commands / scripts reviewed
- [x] Public API paths / contracts reviewed
- [x] Implemented vs planned status reviewed

Files checked:
- [file] — [what was verified]

## Guardrail Overrides
- [ ] override ledger reviewed
- Ledger: runtime/guardrail-overrides.jsonl
- Decision: N/A - no overrides / accepted - [reason] / denied - [follow-up]

## Spec Update Decision
- **Need update?**: yes / no
- **Reason**: [why or why not]
- **Updated files**: [spec files updated, or N/A]

## Follow-ups
- [follow-up item] — [priority]
- (or "none")

## Risks
- [risk] | Status: accepted / mitigated / needs follow-up
- (or "none")

Quality Bar

• All eleven preconditions are verified before proceeding.
• If any precondition fails, finish-work is rejected with a clear message.
• No code changes are made during finish-work.
• No failed gates are bypassed.
• No auto-push occurs.
• Archived artifacts still pass validators after archive.
• The task is archived and marked done.
• Follow-ups are recorded, not lost.