| name | run-checks |
| description | Run the project quality gate through the auto-detected package manager and test runner, then summarize failures concisely. |
| when_to_use | Use WHEN asked to run checks, verify the build, make sure it passes, or verify work before a PR — including the verification step of `/implementation` after realizing To-Do items. Safe for Claude to run before declaring work done. This is the lightweight author-time gate; the final pipeline gate is `/ci-gate`, which also fixes failures and records the result. |
| argument-hint | (no args) |
| model | inherit |
Run the quality gate
Run every step through the detector. Never hardcode the package manager or runner.
source "${CLAUDE_PROJECT_DIR}/.claude/lib/detect-toolchain.sh"
echo "Toolchain: PM=$PM, tests=$TEST_RUNNER, root=$PROJECT_ROOT"
run_typecheck
run_lint
run_tests
run_build
Order: typecheck, lint, tests, build. After a failure, capture the signal and
continue independent steps. Mark missing scripts or binaries as skipped.
Relationship to /ci-gate. This skill runs the four steps individually
and keeps going after a failure to surface every signal at once — ideal for
author-time feedback inside /implementation. /ci-gate runs the chained
ci script (typecheck && lint && test && build) fail-fast, then fixes
failures and writes the result to the task doc. A green run-checks does not
guarantee /ci-gate passes (ordering differs), so the pipeline still ends on
/ci-gate. Use this skill to catch problems early, not to replace that gate.
Report
- ✅/❌ per step: typecheck, lint, tests, build.
- For failures, report the file, the exact error, and the smallest useful fix.
- Group related failures, and do not paste full logs.
- If everything passes, say so in one line.
Fix requested failures without unrelated changes.
Checklist
