Push and Create PR

This skill pushes the current branch to origin (fork) and creates a PR to upstream (main repo).

Workflow

1. Check for uncommitted changes

   bashCopy

   git status --porcelain
   

   If there are changes, ask the user if they want to commit them first.

2. If committing, follow the commit style (see below), then continue.

3. Push to origin

   bashCopy

   git push -u origin HEAD
   

4. Create PR to upstream

   bashCopy

   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:

• Short, imperative sentences (e.g., "Fix wrong type", "Add feature X")
• Use backticks for code references (e.g., "Unify GrootActionDecoder and GrootObservationEncoder")
• No trailing period for short messages
• Examples from this repo:
  • Avoid loading object dtype in SimpleSignal
  • Fix groot metadata so that lerobot dataset can work with multiple keys
  • 6D rotation representation
  • Unify GR00T action decoding and observation encoding

Analyzing Changes for PR

Important: Before writing the PR title and description:

1. Look at ALL commits on the branch (not just the latest):

   bashCopy

   git log main..HEAD --oneline
   

2. Review the full diff from main:

   bashCopy

   git diff main...HEAD --stat
   

3. Identify 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

4. Title should capture the major change, not enumerate commits

Remotes

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):

1. Fetch and display the comments:

   bashCopy

   gh api repos/OWNER/REPO/pulls/NUMBER/comments
   

2. Analyze 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?

3. 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

4. Present findings to user - do NOT make code changes until user approves

5. 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