How to use Claude to coordinate auto-merged features
The author reports 562 merged PRs in two weeks, mostly for Sightline, and a custom-roles goal that ran for over a day with periodic check-ins.
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Accomplish the plan. Ship code using auto-merge. You're the coordinator. Don't do work directly. Use sub-agents for everything, and parallelize when possible.
Reach people exploring AI agents
See placements →What happened
The author reports 562 merged PRs in two weeks, mostly for Sightline, and a custom-roles goal that ran for over a day with periodic check-ins.
Reported by alexeyindeev on BlogHow the agent did it
- Plan the feature with Claude and review scope, risks and previews.
- Set a goal assigning implementation to coordinator-managed sub-agents.
- Require tests and review checks before merge-queue admission.
Where a person stepped in
The author reviewed plans and progress reports, checked in periodically, and nudged agents; tests and bots gated merge-queue admission.
A person set the goal and reviewed the result.
Alexey Indeev describes moving from reviewing agent-written code to reviewing plans and progress reports while building Sightline, an internal operations platform at Spare. He reports merging 562 pull requests in two weeks, with most of that work going into Sightline.
Each session started with planning in Claude: breaking down the work, discussing challenges and risks, and sometimes requesting a UX preview. Once the plan was agreed, the author set a goal using a coordinator prompt. The coordinator held the plan and status while sub-agents performed the work, allowing delegated tasks to come and go without filling the coordinator's context.
Sightline pull requests had to pass unit tests, end-to-end tests, CI, Bugbot, and Strix before Mergify admitted them to the merge queue. The author describes small increments using feature flags, followed by end-to-end verification in staging or production. Instead of reading diffs, he requested an HTML report showing completed work, upcoming work, and needs for human input.
A custom-roles goal ran for over a day. The author checked reports periodically and nudged the agent as needed, while keeping concurrent goals on unrelated features. Who it's for: experienced teams already equipped with test coverage, review bots, merge queues, and feature flags who can actively review plans and exceptions.
Prompt
Original prompt
Source: Blog · alexeyindeev
Frequently asked questions
Which checks blocked Sightline merges?
The source lists unit tests, end-to-end tests, CI, Bugbot and Strix. Mergify would not admit a PR to the merge queue until those checks passed.
What did the author review instead of diffs?
The author reviewed the plan and HTML progress reports describing what was done, what came next and what the agent needed. They also checked in periodically and nudged long-running goals.
More details
Remaining steps
- Merge small feature-flagged increments and verify them.
- Review HTML progress reports and nudge ongoing goals.
About this account
A coordinator tracked the plan and status while sub-agents did the work and shipped increments through an enforced auto-merge workflow.
Platform: Blog. Attribution: https://alexeyindeev.substack.com/p/i-stopped-reviewing-my-agents-code. Prompt: Original (quoted from source). Sourced by the Agents In Use pipeline.
Credit: alexeyindeev on Blog
Source published 4 Oct 2026
Source checked 14 Oct 2026
Tools and topics
Claude · Bugbot · Strix · Mergify · CI
Auto-merge · Coordinator agents · Progress reports
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.