push-pr
Push current branch to origin and create a PR to upstream. Checks for uncommitted changes first.
When & Why to Use This Skill
This Claude skill streamlines the Git development lifecycle by automating branch pushing and Pull Request (PR) creation. It intelligently manages the transition from local code to remote review by checking for uncommitted changes, enforcing project-specific commit message conventions, and generating comprehensive PR descriptions through automated diff analysis. By integrating with the GitHub CLI, it ensures a consistent, professional, and efficient contribution workflow for individual developers and teams.
Use Cases
- Automating the repetitive 'push and PR' cycle to save time and reduce manual CLI errors during feature development.
- Enforcing strict repository commit standards and 'no-AI attribution' rules to maintain a clean and professional project history.
- Generating high-quality PR titles and summaries by automatically analyzing multi-commit diffs to capture the primary intent of changes.
- Streamlining the code review process by fetching GitHub comments and providing structured analysis before proposing code modifications.
| name | push-pr |
|---|---|
| description | Push current branch to origin and create a PR to upstream. Checks for uncommitted changes first. |
Push and Create PR
This skill pushes the current branch to origin (fork) and creates a PR to upstream (main repo).
Workflow
Check for uncommitted changes
git status --porcelainIf there are changes, ask the user if they want to commit them first.
If committing, follow the commit style (see below), then continue.
Push to origin
git push -u origin HEADCreate PR to upstream
gh pr create --repo Positronic-Robotics/positronic --base main --head vertix:BRANCH_NAME \ --title "Title" --body "$(cat <<'EOF' ## Summary <bullet points describing the change> ## Test plan <how to test, or "Tested locally" if applicable> EOF )"
Commit Message Style
Follow the project's commit message conventions:
- CRITICAL: Never mention AI/Claude/assistant - no "Co-Authored-By" or AI attribution
- Keep messages concise - prefer one-line summary, use body only when necessary
- Short, imperative sentences (e.g., "Fix wrong type", "Add feature X")
- Use backticks for code references (e.g., "Unify
GrootActionDecoderandGrootObservationEncoder") - No trailing period for short messages
Good examples from this repo:
Avoid loading object dtype in SimpleSignalFix groot metadata so that lerobot dataset can work with multiple keys6D rotation representationUnify GR00T action decoding and observation encodingAdd chunked batching to migrate_remoteFix cfg/server.py imports and refactor dataset docstrings
What NOT to do:
- ❌ Don't list every single change in the message body
- ❌ Don't add "Co-Authored-By: Claude" or any AI attribution
- ❌ Don't use vague corporate-speak ("improve maintainability", "enhance code quality")
- ❌ Don't write multi-paragraph explanations in commit message
Analyzing Changes for PR
Important: Before writing the PR title and description:
Look at ALL commits on the branch (not just the latest):
git log main..HEAD --onelineReview the full diff from main:
git diff main...HEAD --statIdentify the MAJOR change - what is the primary purpose of this branch?
- Multiple small fixes supporting one feature = describe the feature
- Refactoring + new capability = focus on the new capability
- Don't list every small change; summarize the intent
Title should capture the major change, not enumerate commits
Remotes
| Remote | Repository | Purpose |
|---|---|---|
origin |
vertix/positronic-open | Push branches here |
upstream |
Positronic-Robotics/positronic | Create PRs here |
After PR Creation
Show the PR URL so user can review it.
Handling PR Review Comments
IMPORTANT: Analysis first, no code changes until approved.
When PR comments arrive (from bots or humans):
Fetch and display the comments:
gh api repos/OWNER/REPO/pulls/NUMBER/commentsAnalyze each comment - provide opinion on:
- Is the concern valid?
- Does it apply to our use case?
- What's the priority (critical / nice-to-have / not applicable)?
- Is there context (previous commits, design decisions) that explains the current code?
Check conversation history - the code may be intentional:
- Look at relevant commits:
git show COMMIT_HASH - Check if there's reasoning in commit messages or code comments
- Consider the broader architecture and data flow
- Look at relevant commits:
Present findings to user - do NOT make code changes until user approves
If changes are needed, discuss the approach first, then implement
This prevents:
- Blindly "fixing" intentional design decisions
- Breaking working code based on misunderstood context
- Wasting time on changes that aren't needed