| name | verify-local |
| description | Detect the project toolchain and run fast, deterministic checks locally before pushing. Runs lint and package checks only — lets CI handle full tests. Never push without this skill passing.
|
| argument-hint | [--with-tests] |
| user-invocable | true |
| type | skill |
| mandatory | false |
| triggers | ["verify locally","run local checks","check before push","run CI locally"] |
| metadata | {"author":"Ansible DevTools Team","version":"1.1.0"} |
Verify Local
Fast, deterministic gate before pushing. Runs lint + package checks
only — the same fast-feedback checks CI runs first. Full tests are
left to CI because they are slow, flaky, and environment-dependent.
If this skill reports a failure, do NOT push.
This skill runs commands that modify local state (install deps, build
artifacts). It does not push, commit, or modify remote state.
Why not run tests locally?
- Tests are flaky. A false failure blocks a perfectly good lockfile fix.
- Tests are slow.
tox -e py takes 2-10 min. task wdio takes 30 min and needs Xvfb.
- Local != CI. CI runs across linux/macos/wsl in containers. Local pass doesn't guarantee CI pass.
- The orchestrator handles test failures. Push after lint+pkg pass, CI runs full tests,
diagnose-ci catches failures.
Input
- No arguments: run default checks (lint + pkg/package).
--with-tests: also run unit tests. Use when the fix changed
source code (.ts or .py files), not just lockfile/config.
Entry Gate
Verify you are on a branch (not detached HEAD) and the working tree is
clean (all changes committed):
git branch --show-current
git status --short
The working tree should reflect what you intend to push. Unstaged
changes are allowed — the checks run against the working tree. But
if you see unexpected results, commit first and re-run.
Step 1 — Detect project toolchain
Check what exists in the repo root:
if Taskfile.yml exists AND package.json exists -> TypeScript project
elif tox.ini exists -> Python/tox project
elif pyproject.toml exists (no tox.ini) -> Python/uv project
else -> Unknown, stop
15 of 17 devtools repos are Python/tox. Only vscode-ansible is
TypeScript. ansible/actions is a special case (mixed).
Step 2 — Install dependencies
Use frozen/locked install to match CI:
TypeScript:
If pnpm-lock.yaml was modified (e.g., lockfile regen fix), use:
pnpm install
Otherwise:
pnpm install --frozen-lockfile
The frozen check is CI's job. Verify-local's job is "will build/lint/pkg
pass," not "is the lockfile in sync."
Python/tox:
uv sync --no-progress -q
Python/uv:
uv sync --no-progress -q
If install fails, report it and stop — nothing else will work.
Step 3 — Run checks
Run each check sequentially. For each check, capture the exit code
reliably:
COMMAND_HERE; exit_code=$?
echo "CHECK_NAME: exit_code=$exit_code"
Do NOT pipe the command through other commands when capturing exit codes
— that captures the exit code of the pipe's last segment, not the check.
Default checks (always run)
TypeScript projects (~2 min total):
| # | Check | Command | Time | What it catches |
|---|
| 1 | build | task build | ~30s | TypeScript errors, missing exports, tsup bundling |
| 2 | package | task package | ~45s | Knip dead code, npm packaging, vitest list |
| 3 | lint | npx prek run --all-files | ~30s | ESLint, biome, ruff, codespell, cspell, yamllint, shellcheck |
task package is the critical one — it runs knip which catches most
lockfile-related breakage.
Python/tox projects (~1 min total):
| # | Check | Command | Time | What it catches |
|---|
| 1 | lint | tox -e lint | ~30-60s | ruff, mypy, pydoclint, pre-commit hooks |
| 2 | pkg | tox -e pkg | ~10-20s | Package builds correctly |
Python/uv projects (~30s total):
| # | Check | Command | Time | What it catches |
|---|
| 1 | lint | uvx pre-commit run --all-files | ~30s | All pre-commit hooks |
With --with-tests (only when source files changed)
TypeScript: add task test (+5-15 min)
Python/tox: add tox -e py (+2-10 min)
Python/uv: add uv run pytest (+1-5 min)
Never run locally
These are CI's job — slow, flaky, or environment-dependent:
| Check | Why skip |
|---|
task e2e / task wdio | Needs VS Code binary, Xvfb, 30+ min |
tox -e integration / tox -e eco | Needs external services |
tox -e docs | Only relevant if docs changed |
tox -e milestone | Ecosystem integration tests |
Step 4 — Check if tests are warranted
If --with-tests was NOT passed, check what files the current branch
changed compared to the base:
git diff origin/main --name-only
If the diff contains ONLY these file types, tests are NOT needed:
pnpm-lock.yaml, uv.lock, yarn.lock*.json (package.json, tsconfig.json, knip config)*.toml (pyproject.toml)*.yaml / *.yml (CI config, renovate config)
If the diff contains .ts, .py, .js, or .vue source files, note
it in the output: "Source files changed — consider --with-tests." But
still do NOT run tests by default.
Step 5 — Validate commit message
If there are commits on this branch that are not on the base branch,
validate the latest commit message follows conventional commits:
msg=$(git log -1 --format="%s")
echo "$msg" | grep -qE "^(fix|feat|chore|docs|style|refactor|test|build|ci)(\(.+\))?: .+" \
&& echo "COMMIT_MSG: valid" || echo "COMMIT_MSG: invalid — must follow conventional commits"
Step 6 — Output
## Local Verification
**Toolchain:** TypeScript / Python-tox / Python-uv
**Branch:** branch-name
**Verdict:** ALL PASSING / FAILED
**Duration:** ~Ns total
### Check results
| # | Check | Status | Duration |
|---|---------|--------|----------|
| 1 | build | pass | 12s |
| 2 | package | pass | 45s |
| 3 | lint | pass | 30s |
### Failure details (if any)
**Check:** lint
**Exit code:** 1
**Error:** <last 10 lines of output from the failing check>
### Commit message
**Message:** fix(deps): resolve CI failure from knip update
**Valid:** yes / no
### Source file advisory
Source files changed: yes / no
Tests run: no (default mode)
Advisory: <"Source files changed — consider --with-tests" or "Lockfile/config only — tests not needed">
If all checks pass and commit message is valid: safe to push.
If any check fails: do NOT push. Fix the issue and run verify-local
again.
