How to use Claude to configure an agent team
A Reflectt submission reports Claude configuring an agent team from a bootstrap document. It also describes atomic task claims, structured reflections and presence heartbeats used to address coordination failures.
news.ycombinator.com
Original prompt
Shared by the source author. See source notes for context.
Follow the instructions at reflectt.ai/bootstrap
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source poster / publisher
- reflectt
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The submission reports end-to-end installation and a team of nine agents with 1,344 of 1,362 tasks done. It describes fixes for duplicate claims, lost context and status noise.
- Human involvement
- The operator starts installation and supplies the bootstrap instruction; the post does not describe a subsequent human configuration review.
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
- Started the installation with npx reflectt-node.
- Pasted the bootstrap instruction into Claude.
- Had Claude configure the system from the bootstrap document.
A Reflectt submission, presented as written by the team's distribution agent Spark, describes using Claude to configure an agent team from bootstrap instructions. The reported installation involves running npx reflectt-node and pasting a single instruction into Claude directing it to the bootstrap document. The submission says Claude then configures the system itself.
The account also describes coordination problems encountered while building the system. Duplicate task claims led the team to add atomic claiming: a competing claim receives a 409 response. To address agents losing context overnight, agents write a short reflection after each task, with patterns surfaced as insights for the team. Status messages moved into a heartbeat and presence system so chat could be used for substantive communication.
The submission reports nine agents running since February 28, with 1,344 of 1,362 tasks marked done. It claims installation works end to end, but the sentence about server startup is incomplete and supplies no usable startup duration. These are the submission's own operational claims; it does not describe an independent installation review.
Who it's for: People experimenting with persistent multi-agent teams that need task ownership, shared context and visible presence.
Prompt
Original prompt
Source: Hacker News · reflectt
Frequently asked questions
What instruction was pasted into Claude?
The reported instruction was: "Follow the instructions at reflectt.ai/bootstrap". The submission says Claude used it to configure the system.
How did the team prevent duplicate task ownership?
It added atomic claim semantics to the task board, returning 409 when another agent had already claimed the task.
How did agents preserve context between tasks?
Agents wrote brief structured reflections after each task, and recurring patterns surfaced as insights the team could read.
More details
About this account
The submission says Claude configured the system from bootstrap instructions, while the resulting team used task claims, reflections and heartbeats.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47289444. Prompt: Original (quoted from source). Sourced by the Agents In Use pipeline.
Credit: reflectt on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude · reflectt-node · Task board · Heartbeat system
Agent teams · Coordination · Bootstrap · Task management
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.