mirror of
https://github.com/wassname/pi-plan.git
synced 2026-09-25 14:00:15 +08:00
- Default loop statement is wassname's 2026-09-24 text, spelling fixed only. - Planning prompt states the aim: the agent starts uncertain and reduces that uncertainty. - Planning now allows normal tools (bash, subagent scouts); only file edits outside the goals file and CompleteGoal are blocked. - README: design intent quote, and what the agent versus the user sees at each step. Co-Authored-By: Claude <288921227+claudypoo@users.noreply.github.com>
1.1 KiB
1.1 KiB
pi-goals contributor notes
Keep the design small and close to how models are trained (see README, Design intent): one agent with its normal tools, one goals file, the stock scheduler for wakes, pi-subagents for the judge. Do not add a private timer, task store or second agent runtime. Keep the user's loop statement wording.
All model-facing text is in src/prompts.ts, in conversation order.
Tests
Run npm test, npm run typecheck and npm run lint before committing.
test/harness.tsfakes Pi, the scheduler commands and the pi-subagents event owner.test/real-parsers.test.tschecks the fakes against the real scheduler core and pi-subagents parsers. Update it when those dependencies change.- Test data flow, ownership and lifecycle. Do not assert exact prompt wording.
For a real load check without a model:
cd "$(mktemp -d)" && echo '{"id":"1","type":"get_commands"}' | pi --mode rpc --no-extensions \
-e <repo>/src/index.ts -e <repo>/node_modules/pi-subagents/index.ts \
-e <repo>/node_modules/@jl1990/pi-scheduler/extensions/scheduler/index.ts