Run a task-manager sub-agent with atomic commits
A poster built a Claude Code sub-agent that fetches the next task, marks it complete, and enforces atomic commits with a mandatory commit id.
Done withClaude Code
- What they did
- In the thread, one commenter says they built a Python script to maintain a YAML todo file with task descriptions, then added a task-manager sub-agent with two actions: get_next_task and mark_task_completed. They required a commit ID when marking tasks complete so Claude Code would make atomic commits, while the person still had to watch for cases where it ignored the setup and used TodoWrite instead.
- How it went
- The reported result was better task-by-task control and enforced atomic commits, but the setup was not fully reliable because Claude Code sometimes tried to bypass the sub-agent workflow.
- Worth knowing
- The main catch is reliability: even with the sub-agent and mandatory commit ID, the poster says Claude Code still sometimes bypassed it and used TodoWrite.
Try it yourself with Claude Code
Create a Claude Code sub-agent called task-manager for [project]. It should fetch the next open task from [task file or tracker], hand it to me to work on, mark it complete when done, and require an atomic commit with a commit ID before closing any task. Finish by running it on one sample task and showing me the result.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
2 more accounts of this
Source: Reddit · Mar 1, 2025
Related use cases
Turn tickets into merged pull requests
Pulls tasks from GitHub Issues or Linear, then runs Claude Code or Codex in isolated Kubernetes pods to carry each one through to a PR.
Debug frontend UI bugs with a browser agent
A frontend developer pairs Claude in Chrome with Claude Code to open the app, click through it, read the DOM, check console errors and network requests, and fix UI bugs.
Track investments over time via Claude Code and Robinhood MCP
Poster asks how to get Claude Code connected to Robinhood via MCP to keep tracking investments after researching or trading; a question, not a finished result.
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Run production deploys and port scans from Telegram with OpenClaw
A security professional uses an OpenClaw agent with read-only prod access, PR-only changes and manual approval for restarts to deploy changes and scan a VPS for exposed ports.
Test features on screen and merge fixed PRs
Ryan Carson uses Devin's Test App feature to record and narrate feature clicks, fix bugs, and merge PRs with a !land command.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.