NewsBlur: Command for Claude Code

Your task

Commit all staged/unstaged changes, push to remote, create or update the PR, then watch CI until it finishes and fix any failures.

Arguments provided: {{ arguments }}

Steps to execute

1. Check git status to see what changes exist:

   git status
   git diff --stat
   git diff --staged --stat
   

2. Check for existing PR for the current branch:

   gh pr view --json number,title,url,state 2>/dev/null
   

   • If a PR exists and is OPEN, note its URL and number - we'll update it
   • If no PR exists or it's closed/merged, we'll create a new one

3. Stage all changes (if there are unstaged changes):

   git add -A
   

4. Create the commit following the commit message conventions:

   • Run git log --oneline -5 to see recent commit style
   • Run git diff --staged to understand the changes
   • Write a concise commit message summarizing the changes
   • Use the format:

     git commit -m "$(cat <<'EOF'
     <commit message here>
     
     Co-Authored-By: Claude Opus 4.5 <[email protected]>
     EOF
     )"
     

5. Push to remote:

   git push -u origin HEAD
   

6. Create or update the PR:

   If a PR already exists:

   • The push already updated the PR
   • Note: "Updated existing PR: "

   If no PR exists:

   • Create a new PR:

     gh pr create --title "<title>" --body "$(cat <<'EOF'
     ## Summary
     <1-3 bullet points summarizing the changes>
     
     ## Test plan
     - [ ] <testing checklist items>
     
     ---
     Generated with [Claude Code](https://claude.com/claude-code)
     EOF
     )"
     

7. Watch CI and auto-fix failures — do NOT stop after pushing. Monitor the PR's checks until every check has a terminal status (pass, fail, skipped, cancelled), and fix any failures before handing back to the user.

   a. Poll for status. After pushing, wait ~30s for CI to register, then:

   gh pr checks <PR_NUMBER>
   

   The first column is the check name, the second is status (pending, pass, fail, skipping, cancelled). Re-run this periodically. Do not chain sleep commands to wait — prefer run_in_background so you don't block, and poll by calling gh pr checks again on a reasonable cadence (30–60s). Keep going until no check shows pending.