Dev run tests

Run the Gates

Cheapest-first verification with scope detection. Invoke per task slice, at phase boundaries, and before push.

Steps

1. Resolve commands: profile verify set - run exactly that single gate-runner command and report what it ran and skipped; the remaining steps are its internals and this skill adds only the failure analysis. Else profile typecheck / lint / test; else detect (package.json scripts, *.sln - dotnet test, pyproject.toml - pytest, go.mod - go test ./..., Makefile targets). Confirm detected commands in one line on the first run.
2. Detect scope: git status --porcelain + git diff --name-only - map changed files to their test areas. Docs/markdown-only change - report "no tests to run" and stop. An explicit argument overrides.
3. Typecheck first - it fails fastest. On error: report the file and the first error, STOP.
4. Lint second. On error: report the finding, STOP.
5. Tests: scoped by default; full suite when shared code changed, at phase boundaries, and before push. Enforce the profile coverage_gate when set.
6. On failure: failing test name, the assertion, trimmed relevant output; classify the cause (production change / the test itself / environment); propose the fix. Re-run after fixes with the same scope discipline.
7. Report per area: pass/fail counts, duration, coverage vs the gate.

Verify

• All gates green for the declared scope, or a precise failure analysis delivered - never "mostly passing".

Scope / hand-off

• Fixing production code - the executing skill/subagent; CI-level failures - git-finalize-pr.

CRITICAL

• Never modify tests, configs, or thresholds to make a failing gate pass.
• A flaky test (passes on rerun) is reported as flaky with evidence; never silently retried into green.
• Missing env/dependency - report the precondition; never stub the environment to force a pass.