verify-batch
Skill1 starsChanged 3 months agoMITView on GitHub
Verify a batch of changes against a fixed recipe: build, tests, lint, type checks. Produces a pass/fail verdict with raw evidence. For verifier agents (Willow) checking a worker's batch, or any agent asked to independently verify someone else's work.
In this repository2 other files it gives its agents.
- migration-patternSkill
- review-checklistSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Run the project's verification recipe in full, in this order, every time:
- Distinctive: line 25
If no recipe has been given to you yet, ask the coordinator for one before verifying anything.
- Distinctive: line 34
Every batch gets exactly one of:
- Distinctive: line 42
Never "mostly passed". Never "should be fine".
- Distinctive: line 47
Before blaming the batch, decide which of these you are looking at:
≈ “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.
Files
Verify a batch
You are verifying someone else's work. Your verdict is what makes their batch "done" — nothing is done until it has been seen to pass.
The recipe
Run the project's verification recipe in full, in this order, every time:
- Build — the project's build command, from a clean state if the recipe says so.
- Tests — the full suite the recipe names, not a subset you judge relevant.
- Lint — the project's linter, with the project's configuration.
- Types — the type checker, if the project has one.
If no recipe has been given to you yet, ask the coordinator for one before verifying anything. Do not invent a recipe from the repo's README mid-batch — recipes are fixed at the start so every batch is judged the same way.
A check skipped because "this batch looks fine" is the one that lets a regression through. Small batch, big batch, cosmetic batch: same recipe.
The verdict
Every batch gets exactly one of:
- PASS — every check ran and every check passed. Say so plainly.
- FAIL — one or more checks failed. Name the failing check, paste the actual failing test names and error output (trimmed, never invented), and send it back to the batch's author by @mention.
- UNVERIFIED — you could not run the checks. Say why. This is not a pass.
Never "mostly passed". Never "should be fine". If you cannot produce the verdict, say which check you could not run.
Broken batch vs. broken environment
Before blaming the batch, decide which of these you are looking at:
| Symptom | Likely | Confirm by |
| Import/module errors on files the batch never touched | Environment | Same command on a clean checkout |
| Missing dependency, stale lockfile | Environment | Diff the lockfile against baseline |
| A test that also fails on the untouched baseline | Flake or pre-existing | Run it on the baseline |
| Failures only in files the batch changed | The batch | — |
Report which one it is. An environment failure goes to the coordinator, not back to the worker.
Patterns worth escalating
Escalate to the coordinator (not the worker) when:
- The recipe itself cannot be run at all.
- The same test fails across batches from different workers — that is a pattern problem, not a batch problem.
- Verification time is growing batch over batch — the batches are getting too big.
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
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.

