agentleFS
Sign inStart free
pahlevikunpahlevikun/levistack/skills/engineering/verify/SKILL.md

verify

Skill0 starsChanged 18 days agoMITView on GitHub

Evidence before claims and before fixes. Use before saying work is done, fixed, passing or ready (tests pass, build succeeds, bug fixed, ready to merge, handing off, before a commit, push or PR), when asked only to verify or run the gates, or when a failure has an unknown cause and must be diagnosed before editing, when someone claims a change is safe or asks what it could break, or before you report a measured performance number. Requires a fresh, claim-matched proof, honest reporting of what was and was not checked, and stopping once the proof is complete.

188 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use instinct-export, instinct-import and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • TheKREATORSSkill

    instinct-export

    Export instincts from project/global scope to a file

    Also in 3 projects that use this

  • TheKREATORSSkill

    instinct-import

    Import instincts from file or URL into project/global scope

    Also in 3 projects that use this

  • AviadPSkill

    risk-assess

    Assess the risk and blast radius of a proposed code change before implementing it. Evaluates CI impact, framework-wide vs localized scope, rollback path, and identifies blindspots. Use when user says "what's the risk", "is this safe", "assess the impact", "risk assess", "blast radius", or before applying framework-wide or CI-affecting changes. Do NOT use for code quality review (use code-quality) or plan validation (use design-validator).

    Also in 3 projects that use this

  • diego-tobalinaSkill

    build-fix

    Fix build and compilation errors systematically. Run when build fails.

    Also in 6 projects that use this

  • AaravChadhaSkill

    secure

    Security review with a confidence gate - a finding exists only with a concrete exploit scenario and a high/medium/low confidence rating; everything else goes to a worth-hardening list, never inflated. Sweeps five surfaces: auth gates, secrets hygiene, injection and unsafe sinks, unsafe deserialization/crypto/transport, and LLM tool-use trust boundaries. Reports only, never fixes. Use when the user asks for a security review or to check vulnerabilities, secrets, or auth.

    Also in 3 projects that use this

  • ArvorCoSkill

    specify

    Create feature specification from natural language description. Use when starting a new feature or creating a spec. Triggers on: create spec, specify feature, new feature spec.

    Also in 10 projects that use this

  • +1 more from TheKREATORS

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 167 edited versions we read keep these.

  • 4

    Distinctive

    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
  1. Distinctive:

    Run this before any claim, a commit, a push, a PR, or moving to the next task.

    line 58
  2. Distinctive:

    Scope: If the request is ambiguous about how much to touch, inspect the repository and state the safe assumption.

    line 60
  3. Distinctive:

    Read: all of it: the exit code, the executed and passed counts, warnings, skipped tests, missing artifacts.

    line 63
  4. Distinctive:

    Every claim in the final message is backed by output from this turn, or is labeled as not verified.

    line 118

“Common” is measured: kept by at least 6 in 10 of the 167 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

A claim is only as good as the evidence behind it, and a fix is only as good as the diagnosis behind it. This skill does five jobs:

JobWhenWhere
Gate a claim (default)You are about to say something is done, fixed, passing or readyThis file
Verify onlyThe task is validation, not change: run the gates, prove acceptance, stopverify-only.md
Investigate firstA failure has an unknown cause, or the problem is intermittent or a regressioninvestigate-first.md
Prove it is safeSomeone says a change is safe, or asks what it could break elsewhereprove-safe.md
Vet a numberYou are about to report a measured number: a speedup, a regression, a benchmarkvet-a-number.md

The rule

No completion claim without fresh evidence that matches the claim.

  • Fresh: produced after the last change, by a command you ran in this turn. An earlier run, a remembered pass, a subagent's report and confidence are not evidence.
  • Matches the claim: a focused test supports the behavior it names, not "all tests pass". A linter does not prove the build. Passing tests do not prove the requirements were met.
  • Claim only what the evidence shows. A narrower claim that is true beats a broad one that is assumed.

Rephrasing does not exempt you. "Looks good", "that should do it", "all set" and a relieved tone are claims too.

The gate

Run this before any claim, a commit, a push, a PR, or moving to the next task.

  1. Scope. If the request is ambiguous about how much to touch, inspect the repository and state the safe assumption; ask only when readings differ materially, and do not edit the disputed part meanwhile. Check git status --porcelain and leave unrelated changes alone.
  2. Identify the command that proves this claim. For a ship-level claim run the whole applicable chain (build, types, lint, tests, security scan, diff review) in the project's own order and stop at the first failure. Use the gates the project declares in CLAUDE.md, AGENTS.md or CONTRIBUTING.md; do not invent gates.
  3. Run it now, in full, against the current state.
  4. Read all of it: the exit code, the executed and passed counts, warnings, skipped tests, missing artifacts. A suite that ran nothing exits 0. Confirm the intended binary, revision and entry point actually ran. read-run.mjs does the counting for common runners.
  5. Match. Does the output cover the requirement and one failure path? Re-read the requirements line by line. A stub, a mock, an unreachable branch or a refusal-only path is partial.
  6. Claim what the evidence establishes, plus what was not checked and the real residual risk. Use evidence-report.md when a report helps; skip empty sections.

What each claim needs

ClaimProofNot proof
Tests passRunner output with a positive executed count and 0 failuresA previous run, "should pass", a green exit with 0 tests
Build succeedsThe build command, exit 0A clean linter, readable logs
Types or lint cleanThat tool's own outputA different tool, a partial path
Bug fixedThe original reproduction now passes"I changed the code"
Regression test is realRed then green: fails without the fix, passes with itOne passing run
Feature completeEach acceptance criterion checkedTests passing
No regressionsThe full suite, same command set as beforeOnly the new tests
Subagent finishedThe diff shows the change and you ran the checkThe agent's summary

Proof by kind of change (frontend, endpoint, queue, CLI, migration, infra, package, docs) is in proof-by-change.md.

Do not make the proof easier

Never weaken a test, assertion, validator or acceptance criterion to get green. Never delete, skip or mark a test expected-to-fail to pass. Never hard-code the exercised value or success path. Regenerate snapshots or expected output only after reviewing and justifying the change. Never bypass a failing pre-commit hook that this work caused (--no-verify). Fixtures, mocks and recorded responses support deterministic tests; they are not live behavior. Details: proof-integrity.md.

When verification fails

Report the failure with its output. Do not rerun unchanged hoping it passes, and do not claim success. Fix the cause and restart the gate. Calling a failure pre-existing, environmental or flaky needs evidence: reproduce it on the base revision, or name the independent cause. Details and handoff rules: failures-and-handoff.md.

A clean result is a valid result. Do not invent findings to look thorough, and do not keep adding checks, polish or unrelated tests after the stated criteria are met. Stop when the proof is complete.

Which reference

SituationRead
The task is only to verify or run the gatesverify-only.md
Unknown cause, intermittent failure, regression, "why", or fixes that keep failinginvestigate-first.md
"Is this safe", "what could this break", a wide change you do not trustprove-safe.md
A speedup, a regression, a throughput or latency number, a benchmarkvet-a-number.md
Choosing the right check for a changeproof-by-change.md
Dirty tree, upgrades, strict validation, fixture vs liveproof-integrity.md
Empty results, wrapper scripts, wrong binary, totals that add uptrustworthy-output.md
"Every file", repo-wide rename, audit all, ledgerssweeps.md
A check failed, delegated work, deliverables, staleness, the final reportfailures-and-handoff.md

Related skills

  • trace: the immediate cause is proven, and the user wants the root cause behind it.
  • second-opinion: a wide change needs reviewers on different models. prove-safe supplies the proof, and second-opinion supplies the review.
  • tdd: the fix needs a failing test first.
  • code-review: verify a review comment before you act on it.
  • atomic-semantic-commit: the work is proven and ready to commit.
  • write-mr-description: the work is proven and ready for an MR.
  • handoff: say what was and was not verified when you stop.

Done when

  • Every claim in the final message is backed by output from this turn, or is labeled as not verified.
  • The report names what was run, what was skipped, and the residual risk, with no empty sections.
  • Nothing was weakened, skipped or bypassed to get the result.

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.

Check your setup, free