agentleFS
dativo-iodativo-io/talon/.claude/skills/verify-change/SKILL.md

verify-change

Skill18 starsChanged 24 days agoApache-2.0View on GitHub

Run the right local verification ladder for a Talon change, map every CI job to its local command, and reproduce a CI failure from a pasted log before changing anything.

  • Reads credentials

24 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

6 rules. 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 20 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Verify $ARGUMENTS (a change, a package set, or a pasted CI log).

    line 6
  2. Very specific:

    Run narrow first, broad before handoff, one heavyweight suite at a time (the dev hosts have few cores.

    line 10
  3. Very specific:

    Use -count=1 for anything you will report.

    line 22
  4. Very specific:

    Read the Run <command> header line: it names the job step and therefore the row above.

    line 55
  5. Very specific:

    Run the step's exact command locally, with the same flags (-race, -tags, -count=1), on the whole package, not only the failing test.

    line 57
  6. Very specific:

    No sleeps, retries, or fixture-specific branches to hide a failure (see .claude/rules/testing.md).

    line 65

“Common” is measured: kept by at least 6 in 10 of the 20 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 $ARGUMENTS (a change, a package set, or a pasted CI log).

Ladder

Run narrow first, broad before handoff, one heavyweight suite at a time (the dev hosts have few cores; never run two race/integration/e2e suites concurrently).

  1. Focused: go test ./internal/<pkg> -run '<TestName>' -count=1
  2. Package(s) touched, with the race detector: go test -race -count=1 ./internal/<pkg>/...
  3. Black-box tiers the change crosses:
    • integration: go test -race -count=1 -tags=integration ./tests/integration/...
    • e2e (builds the binary in TestMain): go test -count=1 -tags=e2e -timeout 5m ./tests/e2e/...
    • offline smoke sections: SMOKE_ONLY="37_mcp_governance 38_openshell 39_budget" make test-smoke (no provider key); the full make test-smoke needs TALON_SECRETS_KEY and OPENAI_API_KEY.
  4. Static: gofmt -l ., go vet ./..., golangci-lint run ./..., git ls-files -z '*.sh' | xargs -0 shellcheck, git diff --check.
  5. Repository guards whenever docs, examples or templates changed: go test ./tests (vocabulary and architecture-claim guards) and bash scripts/check-claim-discipline.sh.
  6. Repository-level confidence: make check (lint, vet, SSOT gate, unit, integration, guards) and make test-e2e.

Use -count=1 for anything you will report. Report exactly which commands ran and their result; name what you did not run and why.

CI job to local command

Workflow / job / stepLocal equivalentNeeds
CI / test / Format check, Vetgofmt -l ., go vet ./...—
CI / test / Run unit testsgo test -race -count=1 $(go list ./internal/... ./cmd/... | grep -v internal/testutil | grep -v internal/llm/testutil | grep -v 'cmd/talon')CGO + SQLite
CI / test / PII benchmark regressionmake benchmark-regressionlinux/amd64 for the baseline
CI / test / Run integration testsgo test -race -count=1 -tags=integration ./tests/integration/...—
CI / test / Run repository guard testsgo test ./tests—
CI / test / Run e2e testsgo test -count=1 -tags=e2e -timeout 5m ./tests/e2e/...—
CI / test / MCP governance smokemake test-mcp-governance-smoke—
CI / test / OpenShell composition smokemake test-openshell-smoke— (fixture supervisor, no Docker)
CI / test / OpenShell showcase runner contractmake test-openshell-showcase-contract—
CI / test / Check coverageunit coverage ≥ 65% (coverage.out from the unit run)—
CI / test / Conformance countmake conformance—
CI / test / OPA/Rego policy testsmake opa-testopa binary
CI / proof-gatesmake proof-gates (fuzz gates; slow)—
CI / product-demo-smokemake smoke-product-demogum; writes under /tmp (copy the script and redirect paths on a host where /tmp is read-only)
CI / lint, shellcheckgolangci-lint run ./..., shellcheck over git ls-files '*.sh'—
CI / build, dockermake build, docker build -t talon:ci .Docker for the image
Docs / Markdown link check, claim disciplinecheck every link you added by hand; bash scripts/check-claim-discipline.sh—
Security / govulncheck, gitleaks, SBOMgovulncheck ./... when go.mod or the toolchain changesnetwork
Shortlist Demomake verify-shortlist-demoDocker compose
OpenShell Live / live (v0.1.3 required, v0.1.4 experimental)make test-openshell-live (OPENSHELL_VERSION=v0.1.x)Docker ≥ 28 on Linux, GHCR pull
OpenShell Live / showcasemake openshell-showcase-validateDocker ≥ 28 on Linux
CodeQLno local equivalent—

Anything in the Docker rows cannot be proven on a host without a Docker socket; say so and let the PR's own CI run be the proof for those rows.

Reproducing a CI failure from a log

  1. Read the Run <command> header line: it names the job step and therefore the row above.
  2. Establish which commit CI ran before concluding anything: gh pr view <n> --json headRefOid (or the GitHub API) against git log -1. A failure is often already fixed by a local commit that has not been pushed; say that instead of re-fixing.
  3. Run the step's exact command locally, with the same flags (-race, -tags, -count=1), on the whole package, not only the failing test.
  4. If it passes in isolation but fails in the package run, look for order or environment dependence: shared global state, cwd-relative files, pricing fixtures (black-box fixtures use the real catalog, where ollama is zero-cost), time boundaries, a pre-existing recovered panic printed by another test. Do not call it flaky until you have a mechanism.
  5. A fix needs a test that fails without it; re-run the step's full command afterwards and quote the result.

Fixture rules that CI enforces

  • Black-box fixtures (internal/server, tests/integration, tests/e2e) price with the production estimator over the embedded catalog. A provider or model outside pricing/models.yaml is denied pricing_unavailable in tests exactly as in production; fix the fixture's provider or model, never the estimator.
  • Preventive claims are proven by counting the upstream or action target, not by a status code.
  • No sleeps, retries, or fixture-specific branches to hide a failure (see .claude/rules/testing.md).

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.