pi-dream-rsi

Dream-RSI is a pi extension that assesses local projects and runs a bounded, verifier-driven improvement loop.

Use it

Use assessment to inspect opportunities without running project commands:

/dream-rsi assess "make the parser faster"

Start the autonomous loop with the goal and explicit trusted-local consent:

/dream-rsi loop "reduce benchmark time without changing results"

The loop finds measurable levers, qualifies its metric, runs isolated experiments, and applies only changes that pass confirmation, correctness gates, review, and an independent post-apply check. The loop stops when it reaches its budget or cannot find a verifiable lever.

Its dream step uses completed experiment worlds to compare bounded search configurations through deterministic replay. An agent may propose configuration data; the loop never executes agent-written code as its search policy.

Consent and cost

Consent means project code and declared commands run on this host with the host user's access. This is not containment for hostile repositories. Use the loop only on projects you would run by hand.

Default durable limits are 6 campaigns, 72 agent starts, 200 evaluator runs, and 120 minutes. A campaign can use at most 18 agent starts, 48 evaluator runs, and 60 minutes from the remaining budget. Agent runs may incur model costs. A real-model run requires an explicit user request.

Commands

• /dream-rsi assess [goal] reads project signals and ranks improvement opportunities.
• /dream-rsi loop [goal] starts the unattended loop.
• /dream-rsi status, watch, stop, and resume inspect or control it.

State and rollback

State lives in <project>/.dream-rsi/. Applied changes stay in the working tree. Review them with Git, then keep or revert the patch with your usual Git workflow. See autonomous-loop.md and method.md.

Development

npm install
npm run typecheck
npm test