ship-and-monitor

Ship and Monitor

Commits changes, pushes, opens a PR, and monitors CI/CD checks — auto-fixing obvious failures and asking the user about non-obvious ones.

Workflow

Step 1: Confirm Branch

Before committing, verify the current branch is correct. In long conversations that span multiple branches, never add commits to a branch that was already merged.

git status
git log --oneline -5
git branch --show-current

Check for these red flags:

• Current branch has already been merged (check with gh pr list --state merged --head <branch>)
• Branch name doesn't match the work that was just done
• Unexpected uncommitted changes from unrelated work

If anything looks off, stop and confirm with the user using the AskUserQuestion tool before proceeding.

Step 2: Commit and Push

1. Stage relevant files only — never use git add -A or git add .. Review changes and stage specific files:

   git diff --name-only
   git diff --staged --name-only
   

2. Create a descriptive commit — summarize the "why", not just the "what":

   git add <specific-files>
   git commit -m "<descriptive message>"
   

3. Push with upstream tracking:

   git push -u origin <branch-name>
   

Step 3: Open PR

Open a pull request targeting develop by default, or a user-specified branch.

gh pr create --title "<concise title>" --body "$(cat <<'EOF'
## Summary
<bullet points describing the changes>

## Test plan
- [ ] CI checks pass
- [ ] <relevant test items>

Generated with [Claude Code](https://claude.ai/code)
EOF
)"

If the user specified a target branch, use --base <branch>. Otherwise default to develop. If develop doesn't exist, fall back to main.

Show the PR URL to the user after creation.

Step 4: Monitor CI/CD with /loop

Start monitoring CI/CD checks at 1-minute intervals:

/loop 1m check CI status for PR

Each loop iteration should:

1. Check PR status:

   gh pr checks <pr-number> --watch --fail-fast
   

   Or poll with:

   gh pr checks <pr-number>
   

2. On all checks passing: report success and stop the loop.

3. On failure: classify the failure and either auto-fix or stop to ask.

---

Failure Classification

Auto-fix (obvious failures)

Fix these automatically, commit, push, and continue monitoring:

• TypeScript/build errors — type mismatches, missing imports, unused variables
• Lint/format violations — ESLint, Prettier, Biome, etc.
• Test failures clearly caused by our changes — wrong assertion values, missing mock updates, snapshot mismatches
• Missing documented environment variables — variables referenced but not set in CI config

Auto-fix workflow:

1. Read the CI log to identify the error
2. Fix the issue locally
3. Stage only the changed files
4. Commit with a message like fix: resolve lint error in <file> or fix: update test assertion for <change>
5. Push
6. Continue monitoring

Stop and ask (non-obvious failures)

Stop the loop and report these to the user with full context:

• Test failures in code we didn't touch — may indicate a pre-existing issue or an unexpected side effect
• Infrastructure/deployment failures — Docker build errors, Terraform issues, cloud provider errors
• Flaky or intermittent test failures — tests that passed locally or in previous runs
• Permission/auth errors — missing secrets, expired tokens, insufficient permissions
• Fixes that would require architectural changes — anything beyond the scope of the original intent

When stopping, provide:

• The specific check that failed
• Relevant log output (trimmed to the useful parts)
• Your assessment of the likely cause
• Suggested next steps

---

Branch Safety Rules

• Always verify the current branch before committing
• Check if the branch was already merged — in long conversations, the user may have moved on
• Never blindly commit to whatever branch happens to be checked out
• If in doubt, ask — a 10-second confirmation is better than committing to the wrong branch

Examples

User says: "ship it" → Confirm branch → stage & commit → push → open PR → start /loop 1m monitoring

User says: "commit push and watch CI" → Same flow

User says: "ship it to main" → Same flow but target main as the PR base branch

CI lint fails: → Read log → fix lint error → commit → push → continue monitoring

CI test fails in unrelated code: → Stop loop → report failure details → ask user for guidance

Branch was already merged: → Stop → tell the user → ask what branch to use