agentsclimarketplace

Done gate

Skill tarangdeep-goel-by/claude-harness/claude/skills/done-gate

Dev-focused Claude Code harness — workflow-engine skills, in-repo memory, session continuity (/recall ↔ /vault-push), local telemetry, + a ready-to-fill vault scaffold. clone + ./install.sh and go.

Install
npx -y skills add tarangdeep-goel-by/claude-harness --skill done-gate

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

The completion gate. Use RIGHT BEFORE claiming work is done/fixed/passing/working — run the proof command, read the output, only claim with fresh evidence. Distinct from /verify (which drives the app to confirm runtime behavior); done-gate gates the 'done' CLAIM itself. Fires last in the dev workflow, after the work + /code-review.

SKILL.md

5.2 KB, as published. Nobody here has run it

Done-Gate — Verify Before Claiming Done

The Rule

No completion claims without fresh verification evidence.

Before saying "done", "fixed", "passing", "works", or "complete":

  1. IDENTIFY — What command proves this claim?
  2. RUN — Execute it now. Fresh. Not from memory or cache.
  3. READ — Full output. Check exit code. Don't skim.
  4. CONFIRM — Does the output actually prove what you're about to claim?
  5. REPORT — Include the verification evidence in your response.

What Counts as Verification

ClaimRequired Evidence
"Tests pass"Run test suite, show output with pass count and exit code 0
"Bug is fixed"Run the reproduction case, show it no longer fails
"Feature works"Run/demo the feature, show expected output. If it touches an API endpoint, E2E tests must pass: pytest tests/e2e/ -v
"Build succeeds"Run the build command, show completion without errors
"No regressions"Run full test suite, show all passing

Banned Language (Before Verification)

These phrases are NOT allowed until you have evidence:

  • "should work", "probably fixed", "seems to pass", "likely resolved"
  • "Done!", "All good!", "That should do it!"
  • "I believe this fixes...", "This should resolve..."

Instead say: "Let me verify..." and then run the command.

When You Can't Verify

Sometimes verification isn't possible (no test suite, requires manual UI check, external dependency). In that case, say so explicitly:

"I've made the change but can't verify automatically because [reason]. You'll need to [specific manual check]."

Never pretend you verified when you didn't.

Logging (MANDATORY)

After verification (pass or fail), log it via Bash:

echo '{"ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","skill":"done-gate","project":"<project>","result":"<pass|fail>","task":"<brief task description>"}' >> ~/vault/logs/workflow.jsonl

Task Tracking Check (MANDATORY)

Before reporting completion, verify the task list is up to date:

  1. Call TaskList to see all tracked tasks.
  2. Confirm the current task is marked completed (use TaskUpdate if not).
  3. Confirm no tasks are stuck in in_progress that should be completed or need attention.
  4. If all tasks for the plan/feature are done, state: "All tasks complete — [X/X] done."
  5. If tasks remain, state what's next.

Never claim a feature is "done" while tracked tasks are still pending or in_progress.

After Verification

Report concisely with evidence:

  • "Tests pass (42 passed, 0 failed, exit code 0)"
  • "Bug fixed — reproduction case now returns expected 200 instead of 500"
  • "Build succeeds (docker build completed in 23s)"

Self-Review Checklist (Run Before Claiming Done)

After verification passes, run this quick scan on changed files to catch leftover code smells:

# Get changed files and scan for common issues
git diff --name-only HEAD~1 2>/dev/null | xargs grep -nE \
  'TODO|FIXME|HACK|XXX|NotImplementedError|raise NotImplemented|console\.log\(|print\(.*debug|debugger;|\.only\(|_unused|PLACEHOLDER' \
  2>/dev/null || echo "Clean — no code smells found"

What to check

PatternWhy it matters
TODO / FIXME / HACK / XXXUnfinished work left behind
NotImplementedErrorStub that was never filled in
console.log() / print(...debugDebug logging left in
debugger;JS debugger statement
.only(Focused test that skips other tests
PLACEHOLDERPlaceholder text never replaced

Rules

  • Quick tasks: Skip this checklist (overkill for typo fixes)
  • Small+ tasks: Run the grep scan. Fix any hits or justify why they should stay.
  • If hits are intentional (e.g., a TODO for a future ticket), note it in the completion message.
  • Do NOT add this as a pre-commit hook — it's advisory, not blocking.

Vault Push — After Successful Verification (Small+ Tasks)

After verification passes for Small, Medium, or Large tasks, push a completion note to the vault. Skip for Quick tasks.

What to push

Append an entry to ~/vault/notes/<project>-completed.md (create if it doesn't exist):

### YYYY-MM-DD — <task title>
- **Tier:** Small/Medium/Large
- **What:** One-line summary of what was done
- **Decisions:** Key architectural or design choices made (if any)
- **Files:** Main files touched (3-5 max)
- **Learnings:** Anything worth remembering for next time (if any)

How to push

  1. Write/append the entry to ~/vault/notes/<project>-completed.md
  2. Run qmd update && qmd embed via Bash
  3. Mention briefly: "Pushed completion note to vault."

Rules

  • Keep entries short — 3-5 lines max per task
  • Use the project name from the working directory (e.g., todo, ai-data-analyst-v2, ladder)
  • If project is unclear, use general-completed.md
  • Don't push duplicate entries — check if the task was already logged

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.