Claude Code coordinates edits through Foremerge
GPTree reports using Foremerge with parallel coding agents to expose conflicting edit intentions before code changes; a replay flagged one conflict and revealed a scope blind spot.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Before editing, publish your intended changes and scopes through Foremerge. Check related work, negotiate conflicting plans, log the resolution, and run the configured acceptance check before handing over.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- naw103 / GPTree
- Original account
- Read the source ↗
- Published
- 21 Sept 2026
- Source link checked
- 15 Oct 2026
- Reported result
- GPTree reports using Foremerge with parallel coding agents to expose conflicting edit intentions before code changes; a replay flagged one conflict and revealed a scope blind spot.
- Human involvement
- Humans configure acceptance checks and can override a conflict with a logged reason. Replay results and stress-test claims are builder reports.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Parallel coding agents publish intended changes and scopes before editing their isolated worktrees.
- Foremerge returns related work and conflicts from the shared SQLite coordination store.
- Agents coordinate resolutions with a recorded rationale rather than relying on file locks.
At GPTree, naw103 ran coding agents in parallel worktrees and found that plans could conflict even when their patches did not touch the same lines. Replacing a class while another agent extended it was the motivating example. The team built Foremerge and reports using it in its development workflow.
Agents publish their intended change and affected scope before editing. A shared SQLite store returns related work and declared conflicts, while each agent retains its own worktree. Claude Code and Codex can use the same coordination store. When a high-severity conflict appears, agents exchange messages and record how it was resolved; claims are advisory, so the tool does not lock every overlapping scope.
The most useful result is qualified. The author replayed 76 intentions from a previous build, found one conflict and says Foremerge flagged it. Review also exposed a missed relationship when one agent claimed a class and another an internal method. That is evidence of both a working check and a real coverage gap.
A human configures the executable acceptance check. An agent's claim that tests passed does not satisfy that gate. The published account also notes missing push notifications and limits in comparing actual changes, so coordination still depends on agents declaring useful scopes and checking status.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · naw103 / GPTree
Frequently asked questions
Does Foremerge lock every overlapping edit?
No. Claims are advisory leases, with high-severity unresolved conflicts handled at acceptance.
What did the replay find?
The builder reports one conflict in 76 intentions, which was flagged, plus a blind spot involving class and method scopes.
Can an agent merely say its tests passed?
No. The configured acceptance check must run against the change state; a reported pass is not enough.
More details
Remaining steps
- The builder replays 76 intentions, reports one flagged conflict and identifies a class-versus-method scope blind spot.
- A human-configured executable check gates acceptance of the resulting change.
About this account
The agents declare plans, respond to coordination findings and log resolutions while coding in separate worktrees.
Firsthand account opened and read, with linked project documentation where available. Outcomes are author reports, not independently reproduced. Suggested prompt written by Agents In Use. The builder discusses using Claude Code and alternating with Codex. The replay does not identify every participating client, so its result is not attributed solely to Claude. No claim that Foremerge itself is an LLM agent.
Credit: naw103 / GPTree
Source published 21 Sept 2026
Source checked 15 Oct 2026
Tools and topics
Claude Code · Foremerge · SQLite · Git worktrees
Claude Code · Foremerge · SQLite
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.