code-batch

Cloud Batch — Parallel Multi-Repo Execution

Run the same task across multiple GitHub repos simultaneously. Each repo gets its own ephemeral sandbox.

Process

1. Get the repo list. Ask the user for repos if not provided. Accept:

   • Comma-separated: owner/repo1, owner/repo2, owner/repo3
   • One per line
   • GitHub org pattern: "all repos in org-name" (list with gh repo list org-name)

2. Get the prompt. The same instructions apply to all repos unless the user specifies per-repo tasks.

3. Dispatch in parallel. Call the remote_exec_batch MCP tool:

   • tasks: Array of { repo, prompt, branch?, model? } objects

   Example:

   {
     "tasks": [
       { "repo": "owner/repo1", "prompt": "Update dependency X to v2.0" },
       { "repo": "owner/repo2", "prompt": "Update dependency X to v2.0" },
       { "repo": "owner/repo3", "prompt": "Update dependency X to v2.0" }
     ]
   }
   

4. Report aggregated results. Show per-repo:

   • Success/fail status
   • Exit code
   • Diff summary (truncated)
   • Errors if any

Example Usage

User: "Update the README copyright year to 2026 across my-org/app, my-org/lib, and my-org/docs"

Call remote_exec_batch with 3 tasks, same prompt. Present: 3/3 succeeded, diffs showing year change.

Notes

• All tasks run in parallel — total time is roughly the slowest task
• Each task is independent; one failure doesn't affect others
• Results include per-repo diffs but changes are NOT auto-pushed
• For large batches (10+ repos), consider running in groups