How to use OpenClaw to run dev and QA loops
DevClaw uses OpenClaw to run issue-based DEV and QA loops for isolated Telegram projects, reusing sessions and selecting models by task.
news.ycombinator.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Configure [OpenClaw project] with DevClaw using [Telegram group] and [GitHub or GitLab issues]. Isolate its queue and state, dispatch coding tasks to DEV, and send results to QA with failures returned to DEV. Reuse sessions and select models according to [task policy].
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- laurentenhoor
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author uses DevClaw for all development and claims roughly 70% token savings versus one large model with fresh context each time.
- Human involvement
- The post explicitly says there is no human in the dev/QA loop; QA is performed by an agent.
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
- Built DevClaw to isolate development projects by Telegram group.
- Dispatched issue-based tasks to DEV workers to write code.
- Sent implementations to QA and returned failures to DEV automatically.
The author built DevClaw, an OpenClaw plugin that treats each Telegram group as an isolated development team, and says they use it for all their development. Its orchestrator, development workers and QA workers run independently. DEV writes code, QA reviews it, and failures return automatically to DEV without a human in the loop.
Each group's project has separate queues, workers, sessions and state. GitHub or GitLab issues act as the source of truth, with no internal database. A CLI-based heartbeat scans queues and dispatches work without consuming LLM tokens. The plugin combines label transitions, state changes, session dispatch and audit logging in an atomic operation with rollback on failure.
The system chooses models by task, using Haiku for typos, Sonnet for features and Opus for architecture. Workers retain sessions between tasks rather than rereading the codebase each time. The author claims these choices compound to roughly 70% token savings compared with one large model using fresh context for every task; the post does not supply a benchmark methodology.
Who it's for: developers exploring issue-driven autonomous development with separate project state and automated QA feedback.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · laurentenhoor
Frequently asked questions
How are projects separated?
Each Telegram group has its own project queue, workers, sessions and state; GitHub or GitLab issues serve as the source of truth.
Is a human involved in the development loop?
The author explicitly describes autonomous DEV and QA loops without a human in the loop: failed reviews go back to DEV automatically.
More details
About this account
The author explicitly describes autonomous orchestration, development and QA loops without human participation in the loop.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47016188. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: laurentenhoor on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
OpenClaw · DevClaw · Telegram · GitHub Issues · GitLab Issues · CLI
Development automation · QA · Telegram
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.