Managing a fleet of AI agents with a git repo of markdown plans and coordinator agents
bibryam · x · 2026-10-05
Engineer Fatih Arslan shares the workflow he's used for months to run a fleet of AI agents:
- Plans as index: every agent task is a single markdown file (with area prefixes like hub-service-500.md) stored in a dedicated git repo — human-readable, shareable with coworkers or agents, with history for free.
- Structure: a one-time /plan-init skill scaffolds a plans/ repo with drafts/next/open/done/discarded folders and a README.md index.
- Coordinators: he moved from random agents creating plans to long-running coordinator agents that read/write the repo and drive progress.
- Entry point: e.g., a Slack alert gets captured via @cursor /plan-add investigate why the service is getting 500... as a draft plan that agents pick up.
The core idea: use human-readable plan files as shared memory and a coordination layer between agents instead of letting them work in isolation.
More from coding & agent
- The viral vibe-coder SEO prompt: AI fixes the tech, but backlinks still take grinding — sujingshen · 2026-10-05
- Blender + Claude + Polyxd: swapping prompts for sliders to drive a 3D island scene live — sidahuj · 2026-10-05
- Borrowing aviation's controlled English (ASD-STE100) to strip AI fluff and fake completions — sujingshen · 2026-10-05
- Use Magpie CLI to check quotas and route sub-agents by urgency to save tokens — lxfater · 2026-10-05
- Give Claude real designer assets: 40 vetted resources to fix vibe-coded UI, says founder — PrajwalTomar_ · 2026-10-05
- Claude Opus 5.5 builds an interactive lens lab in one 1h26m shot for $25.66 — nikola_mr64990 · 2026-10-05