Pr Check

---
description: Run all local CI checks before creating a PR
allowed-tools: AskUserQuestion, Bash, Read, Edit, Write, Glob, Grep
argument-hint: (no arguments)
---

# PR Check Skill

Run all local CI checks before creating a PR. On failure, offer to fix issues interactively.

## Philosophy: Ask Early, Ask Often

**This skill should liberally use `AskUserQuestion` at decision points.** CI check failures require judgment about whether to auto-fix, skip, or investigate. The user should approve each fix strategy.

- **When** a check fails — ask how to handle before auto-fixing
- **When** multiple checks fail — let the user prioritize which to fix first
- **After** all checks pass — ask about next steps (PR, review, etc.)

## Pre-flight

Before running checks, verify the Makefile exists:

```bash
ls -la ods/Makefile

If the Makefile doesn't exist, inform the user and stop.

Checks Overview

Display to user before running:

Running 8 ODS checks:

1. Shell lint - bash -n syntax check on all .sh files
2. Python compile - py_compile check on dashboard-api modules
3. Tier map + contract tests - Unit and contract test suite
4. BATS unit tests - Shell library tests via BATS
5. Smoke tests - Platform-specific smoke tests
6. Installer simulation - Installer simulation harness
7. Dashboard build/lint - ESLint + Vite production build
8. Secret scan - Pre-commit hooks (gitleaks, private keys)

Execute Script

Run the full gate (covers checks 1-6):

cd ods && make gate

Then run dashboard checks:

cd ods/extensions/services/dashboard && npm run lint && npm run build

Then run secret scan:

pre-commit run --all-files

IMPORTANT: Stream output to user in real-time.

On Success

If all checks pass:

AskUserQuestion:
  question: "All 8 PR checks passed! What would you like to do next?"
  header: "Next"
  multiSelect: false
  options:
    - label: "Create PR"
      description: "Run /pr to create a pull request with these changes"
    - label: "Run code review"
      description: "Run /code-review before creating the PR"
    - label: "Done"
      description: "Checks passed — I'll handle the rest manually"

On Failure - Interactive Fix Loop

When a check fails, you MUST:

1. Identify Which Check Failed

Parse the output to determine which step failed:

From make gate:

• Shell syntax check failed = Check 1
• Python compile check failed = Check 2
• Tier map / contract tests failed = Check 3
• BATS unit tests failed = Check 4
• Smoke tests failed = Check 5
• Installer simulation failed = Check 6

From dashboard:

• npm run lint failed = Check 7 (lint)
• npm run build failed = Check 7 (build)

From pre-commit:

• gitleaks / private key detected = Check 8

2. Ask User How to Handle Each Failure

Use AskUserQuestion for each failed check rather than text-based offers:

AskUserQuestion:
  question: "Check <N> (<check name>) failed. How should I handle it?"
  header: "<check>"
  multiSelect: false
  options:
    - label: "Auto-fix (Recommended)"
      description: "<Specific fix strategy for this check type>"
    - label: "Show error details"
      description: "Display the full error output before deciding"
    - label: "Skip this check"
      description: "Move on — I'll handle it manually"
    - label: "Abort"
      description: "Stop running checks — I need to investigate first"

Fix strategies by check type:

Check	Auto-Fix Strategy
[1] Shell lint	Read the failing file, fix syntax error (missing quotes, bad conditionals, etc.)
[2] Python compile	Read the failing .py file, fix syntax/import error
[3] Tier map / contracts	Read failing test and source, fix the logic mismatch
[4] BATS tests	Read failing .bats test and source lib, fix the assertion or implementation
[5] Smoke tests	Read the smoke test script, fix platform-specific issue
[6] Simulation	Read simulation harness output, fix installer phase issue
[7] Dashboard lint/build	Run npx eslint --fix in dashboard/, manually fix remaining; fix build errors
[8] Secret scan	Identify the secret, remove it from code, add to .gitignore if needed

If multiple checks fail, ask about priority order:

AskUserQuestion:
  question: "<N> checks failed: <list>. Which should I fix first?"
  header: "Priority"
  multiSelect: false
  options:
    - label: "Fix in order (Recommended)"
      description: "Address failures in check order: <ordered list>"
    - label: "Lint first"
      description: "Fix lint/syntax issues first — they may resolve other failures"
    - label: "Tests first"
      description: "Fix test failures first — they indicate real bugs"
    - label: "Let me choose"
      description: "I'll tell you which check to fix first"

3. After Fix - Re-run Checks

After applying any fix:

cd ods && make gate

Continue the fix loop until:

• All checks pass, OR
• User decides to stop

4. Track Fixes Applied

Keep a mental note of all fixes applied during the session. On final success, summarize:

All PR checks passed!

Fixes applied:

• Fixed shell syntax error in installers/lib/detection.sh
• Fixed BATS assertion in tests/bats-tests/tier-map.bats
• Removed accidental API key from config/litellm.yaml

Error Recovery

If a check fails in an unexpected way (not a code issue):

• Missing dependency: Offer to install it (npm ci for dashboard, pip install pre-commit)
• BATS not found: Check tests/bats/ submodule: git submodule update --init
• Timeout: Suggest re-running or checking system resources

Notes

• make gate runs: lint + test + bats + smoke + simulate (checks 1-6 in one command)
• Dashboard checks must be run separately (not in Makefile gate)
• Secret scan requires pre-commit to be installed
• The Makefile uses `set -