agentleFS
tikocitikoci/rosetta/.github/skills/verify-task/SKILL.md

verify-task

Skill54 starsChanged 4 months agoMITView on GitHub

Verify a work item is actually done by running the V-* validation rows its issue cites. Maps V-* IDs to the CI step / test command via VALIDATION.md, runs them, reports pass/fail per row.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If a row is currently GAP in VALIDATION.md:

    line 66
  2. Very specific:

    If the work item under review is the one named in Tracked by, verify that the PR adds the missing proof and updates the row away from GAP.

    line 67
  3. Very specific:

    Run release.yml — the user has to dispatch that themselves.

    line 97

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “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

Verify Task

Closes the loop between a work item's claimed acceptance and what's actually proven. Reads the V-* rows cited in the issue's ## Validation section (per the work-item template), looks each up in VALIDATION.md, runs the proof, reports green/red.

When to use

  • A PR is about to claim Closes #N and you want to confirm the cited validation rows are actually green.
  • A user asks "is this issue really finished?"

When NOT to use

  • For an issue that cites no V-* rows (pure-doc work). There's nothing to run.
  • As a substitute for code review.

Procedure

  1. Resolve the work item. An issue number → gh issue view <n> and pull the V-* IDs from its ## Validation section — falling back to scanning acceptance bullets for older issues that predate the template (also check any archived tasks/done/T-*.md file the issue names as its full spec). T-0037 → read tasks/T-0037-npm-prerelease-dist-tag-channel.md frontmatter directly (the one still-in-flight file-based task).
  2. Look up each V-* in VALIDATION.md. Pull the Proven by, Status, and Tracked by columns. This tells you whether the invariant is already enforced, non-blocking, CI-only, or still a declared gap.
  3. Run the proof.

    Run make verify — it covers V-typecheck, V-lint, V-unit, V-tool-registry, and V-retrieval-floor in one command (requires a populated DB). It runs bun test src/mcp-contract.test.ts without ROSETTA_REAL_DB_TESTS=1, so the real-DB blocks behind V-tool-shapes and V-tool-budget are skipped — run those separately with ROSETTA_REAL_DB_TESTS=1 bun test src/mcp-contract.test.ts against a populated DB. For CI-only invariants (V-db-min-content, V-bunx-*) there is no local equivalent; those are only proven by a release run.

    For more targeted checks, map common V-* IDs to commands:

    V-* IDCommand
    V-typecheckbun run typecheck
    V-lintbun run lint
    V-unitbun test
    V-tool-registrybun test src/mcp-contract.test.ts (Block A, always runs)
    V-tool-shapesROSETTA_REAL_DB_TESTS=1 bun test src/mcp-contract.test.ts (Block C, needs populated DB)
    V-tool-budgetROSETTA_REAL_DB_TESTS=1 bun test src/mcp-contract.test.ts (Block B, needs populated DB)
    V-retrieval-floorbun run src/eval/retrieval.ts
    V-retrieval-selfbun run src/eval/self-supervised.ts
    V-canonicalizebun test src/canonicalize.test.ts src/canonicalize.fuzz.test.ts
    V-classifierbun test src/classify.test.ts
    V-schema-roundtripbun test src/schema-roundtrip.test.ts
    V-extract-videosbun test src/extract-videos.test.ts
    V-http-handshakebun test src/mcp-http.test.ts
    V-stdio-handshakebun test src/mcp-stdio-client.test.ts
    V-tui-mcp-paritybun test src/browse-parity.test.ts
    V-cli-flag-uniformitybun test src/cli-help.test.ts
    V-db-wipe-guardbun test src/query.test.ts
    V-release-structurebun test src/release.test.ts

    For V-bunx-* and V-db-min-content rows, the proof lives in release.yml — these can't be cleanly run locally. Note that explicitly: "this V-* is verified by CI on the next release run."

    If a row is currently GAP in VALIDATION.md:

    • If the work item under review is the one named in Tracked by, verify that the PR adds the missing proof and updates the row away from GAP.
    • Otherwise report the row as an unresolved gap and do not recommend closing the issue.
  4. Report. For each V-*:
    • ✓ pass — quote the relevant Proven by description so the user knows what just ran.
    • ✗ fail — surface the test output's failure summary.
    • ⊘ CI-only — explain that this row is verified on workflow_dispatch of release.yml; can't be checked locally.
    • ⚠ GAP — explain that the invariant is not yet proven, quote Tracked by, and only consider it satisfied if the current work item is the one closing that gap.
  5. Conclusion. Recommend merging (and letting Closes #N fire) only if every row is green or legitimately CI-only. If any row is failing or still GAP, don't recommend it — point at the missing proof or failing acceptance criterion. If acceptance is only partially met, remind that follow-up issues must be opened before merge (issue-pr-linking.instructions.md).

Output format

Issue: #19 — extractor safe import pattern
Validation rows: V-db-wipe-guard

V-db-wipe-guard: ✓ pass
  Proven by: bun test src/query.test.ts (wipe-guard block)

Conclusion: rows green. PR may claim Closes #19 once acceptance bullets are all delivered.

What this skill does NOT do

  • Close issues or open/merge PRs.
  • Run release.yml — the user has to dispatch that themselves.

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.