From spec to tickets without filing twice

Product and designPM and two engineers turning a spec into GitHub issues with agents

I asked Claude Code to break the export spec into issues. The same morning an engineer asked Codex to 'create tickets for the export work'. We ended up with 23 issues, nine of them duplicates.

How do I stop two AI agents from filing duplicate GitHub issues from the same spec?

Have the PM's agent break the spec into parts with ArcRoom's split_task, then have each agent claim a part with claim_task before filing its issue with gh. A second agent sees which parts are claimed or done, so nobody files the same ticket twice. ArcRoom doesn't sync with GitHub, Linear or Jira; the tracker stays the record of the work.

Also asked as: Two of us asked agents to break down the same spec and got duplicate tickets. How do we avoid that? · How do I split a spec into tickets across several AI agents without overlap? · Can Codex see which parts of a spec Claude Code already claimed?

The work

The PM's agent runs split_task on the spec; the parts are stored in the playbook.

ArcRoom keeps
  • Claim
  • Split into parts
  • Model router
The next agent gets

split_task → bulk export (6 parts)

Step by step

  1. The PM's agent runs split_task on the spec; the parts are stored in the playbook.
  2. Each agent claims a part with claim_task before filing its issue with gh.
  3. A second agent sees which parts are claimed or done.
  4. split_task suggests a cheaper model per part through route.model.

What the agent sees

split_task → bulk export (6 parts)
✓ CSV writer (issue #412) ✓ size limit (#413) ○ email when ready: claimed by codex-arun
route.model: 'progress bar copy' → haiku

Compared with today's tools

Manual breakdown in Linear or Jira

The PM writes the tickets by hand or one person runs the agent.

Setup
None
Ongoing
30–60 min per spec

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
2 min (npx arcroom init)
Ongoing
None

The tracker is the right record; Mem0 and Zep have no task claims, so nothing stops two agents filing 'CSV writer' at the same time.

Where it stops

ArcRoom doesn't sync with GitHub, Linear or Jira. The tracker stays the record of the work.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in product and design