Ralph Batched Autonomous Executor

Execute Ralph iterations in batches – implement up to 5 user stories automatically in a single chat, then stop.

This is the batched autonomous mode skill. Use this when you want:

• Faster execution (5 stories per chat)
• Autonomous operation within batch
• Balance between speed and context management

For maximum context cleanliness, use jpralph-iterate instead (one story per chat).

Your Task

1. Read the PRD at prd.json (in project root)
2. Read the progress log at progress.txt (check Codebase Patterns section first)
3. Check you’re on the correct branch from PRD branchName. If not, check it out or create from main.
4. Loop through incomplete stories (up to batch limit):
   • Pick highest priority story where passes: false
   • Implement that story completely
   • Run quality checks (typecheck, lint, test)
   • Update AGENTS.md if you discover reusable patterns
   • If checks pass, commit with message: feat: [Story ID] - [Story Title]
   • Update PRD to set passes: true for completed story
   • Append progress to progress.txt
   • Continue to next story (up to batch limit)
5. Output completion signal based on status

Batch Configuration

• Default batch size: 5 stories
• Maximum batch size: 10 stories
• User can specify: “Execute batch of 3 stories” or “Run 7 stories”
• Purpose: Prevents context overflow by limiting iterations per chat

Progress Report Format

For EACH story in the batch, APPEND to progress.txt:

## [Date/Time] - [Story ID]
Chat: [Current chat ID] - Batch iteration [N of batch_size]
- What was implemented
- Files changed
- **Learnings for future iterations:**
  - Patterns discovered
  - Gotchas encountered
  - Useful context
---

Consolidate Patterns

Same as jpralph-iterate – add reusable patterns to ## Codebase Patterns section at top of progress.txt.

Update AGENTS.md Files

Same as jpralph-iterate – update AGENTS.md files with genuinely reusable knowledge discovered during implementation.

Quality Requirements

• ALL commits must pass quality checks (typecheck, lint, test)
• Do NOT commit broken code
• Keep changes focused and minimal per story
• Follow existing code patterns

Browser Testing

For any story that changes UI, verify in browser using Antigravity’s browser tools.

Stop Conditions

Batch Complete (more stories remain)

If you’ve completed the batch limit BUT there are still stories with passes: false:

<promise>BATCH_COMPLETE</promise>

Batch complete! ✅

Completed in this batch:
- US-001: [title]
- US-002: [title]
- US-003: [title]
- US-004: [title]
- US-005: [title]

Remaining stories: [count]
Next story: US-006 - [title]

To continue:
1. Create a NEW CHAT (for fresh context)
2. Load the jpralph-auto skill
3. Execute next batch of 5 stories

All Complete

If ALL stories have passes: true:

<promise>COMPLETE</promise>

All user stories complete! 🎉

Total stories completed: [count]
Branch: [branchName]

Next steps:
1. Review all changes
2. Run final quality checks
3. Merge branch to main

Error Handling

If a story fails quality checks after 3 attempts:

1. Mark story with notes in prd.json explaining the issue
2. Skip to next story in batch
3. Report all failures at end of batch

Example notes:

{
  "id": "US-003",
  "passes": false,
  "notes": "Failed after 3 attempts. Typecheck error: Property 'foo' does not exist on type 'Bar'. Needs manual review."
}

Loop Control

• Maximum attempts per story: 3
• Stop if: Critical error (e.g., git conflicts, missing dependencies)
• Continue if: Individual story fails (mark and skip)
• Report: All failures and successes at end

Progress Tracking During Batch

Provide status updates as you work:

Starting batch execution...
Batch size: 5 stories

[1/5] Implementing US-001: Add priority field to database
✅ US-001 complete

[2/5] Implementing US-002: Display priority indicator
✅ US-002 complete

[3/5] Implementing US-003: Add priority selector
✅ US-003 complete

[4/5] Implementing US-004: Filter tasks by priority
✅ US-004 complete

[5/5] Implementing US-005: Sort by priority
✅ US-005 complete

Batch complete! All 5 stories passed.

Important

• Work through stories in priority order
• Commit after each story (not at end of batch)
• Keep CI green throughout
• Read Codebase Patterns section before starting
• Provide clear continuation guidance when batch completes

Batch Size Recommendation

• Small PRDs (3-5 stories): Use full batch size
• Medium PRDs (6-15 stories): Use batch size of 5 (default)
• Large PRDs (16+ stories): Use batch size of 3-5, split into multiple batches

This keeps context manageable and allows for fresh chats between batches.