Commit Graph
20 Commits
Author SHA1 Message Date
wassname2 f9bb2aec2e docs: record supervision preferences and human journal 2026-09-14 14:13:18 +08:00
wassnameandPi/OpenAI 9cce6a6bee Remove abandoned runtime and keep current entry points and tests
Replace historical docs, trial output and dead runtime with current source, agent definition and reusable scripts. Bundle pinned worker, Intercom and scheduler dependencies via Pi manifest. Retain 71 current tests, including real RPC proposal/editor/Ready checks and evidence edge cases; lint/typecheck pass. Historical material remains available at f22d83c.

Co-Authored-By: Pi/OpenAI <288921227+claudypoo@users.noreply.github.com>
2026-09-10 18:40:23 +08:00
wassname2 cb35fbf1fe Research overnight supervision across harnesses and Pi packages 2026-09-10 11:43:48 +08:00
wassname2 fb5503f083 Prototype main-chat supervision with interactive edxeth workers 2026-09-10 10:03:40 +08:00
wassname 17a3b25a82 Batch high-impact planning questions 2026-08-26 13:43:42 +08:00
wassname 2362073d04 Make planning research conditional 2026-08-26 13:42:44 +08:00
wassname c1bf91f3db Resolve planning uncertainty before approval 2026-08-26 13:32:06 +08:00
wassname 389af540d1 Test review flow through Pi RPC 2026-08-26 12:12:05 +08:00
wassname fa7195eafb Make planning goals concrete 2026-08-26 12:05:04 +08:00
wassname 0d972e81c3 Align planning state with review flow 2026-08-26 10:46:05 +08:00
wassname b173d145db Version plans and expose judge review 2026-08-24 21:38:45 +08:00
wassnameandClaudypoo 0a349b056e review: deepseek approves the ready menu after 2 rounds
Both round-1 findings were withdrawn once the reviewer had the plan-mode
facts. Comment the state-flip order, which is the part that reads like a bug
and is not.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
2026-08-21 08:25:54 +08:00
wassnameandClaudypoo e6af6db3d9 release 0.2.0: docs and description follow the per-session plan path
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
2026-08-14 16:07:26 +08:00
wassnameandClaudypoo 7e427d2ca6 spec: one plan file per session, .pi/plan/<session_id>.md
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
2026-08-14 16:07:26 +08:00
wassname 6df9390d2f drop pi-lgtm references from description, README and spec 2026-08-10 18:07:04 +08:00
wassname 14e757cfe1 Make goal judge use explicit session model 2026-06-29 05:28:17 +08:00
wassname 9a4da96d56 Fail forward when goal judge is inconclusive 2026-06-29 05:14:23 +08:00
wassname 489f9b8c35 Clean pi-plan references, add judge timeout, fix heading format
- Rename spec doc to 2026-06-15_pi-goals.md, update title
- Update review.md spec reference
- Rename piPlanExtension -> piGoalsExtension in src/index.ts
- Add 120s timeout to judge subprocess (was unbounded, caused hang)
- Change planInjection heading from 'Goals (goals.md):' to '.pi/goals.md:'
- Add FIXMEs for tool label, progress visibility, heading format
2026-06-17 18:09:03 +08:00
wassnameandClaudypoo c74f625b65 Fix newSession context bug + mark unwired loop prompts (external review)
External review (deepseek-v4-pro, docs/reviews/review.md) found:
- Real bug: reviewLoop cast agent_end's ExtensionContext to ExtensionCommandContext
  and called newSession on it, which would crash the "fresh context" path. Now the
  /plan command handler's context (which has newSession) is saved and reused, with a
  graceful in-place fallback (burneikis pattern).
- Dead code: continuation + loopJudge prompts are unused (the autonomous loop is
  intentionally cut). Marked NOT-YET-WIRED in the prompts.ts flow header rather than
  removed, so the full intended flow stays reviewable.
Review also confirmed: no comment bloat, no over-engineering.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
2026-06-15 18:40:19 +08:00
wassnameandClaudypoo d97b532d7b pi-plan: plan-mode goals + evidence in one plan.md, subagent sign-off
Small, guide-not-gate plan/goal tracker for pi. The agent edits plan.md with
its normal Edit tool; CompleteGoal is the one blessed path that runs verify +
a read-only judge and records the result. Plan mode drafts goals (done_when +
failure_modes + subtasks), a per-turn injection keeps the active goal alive
through compaction, and a reminder drives upkeep + autonomy.

- src/plan-file.ts: pure parse + the two writes CompleteGoal needs + recordSignOff
- src/index.ts: plan mode, review menu, injection, reminder, widget, CompleteGoal, oracle spawn
- src/prompts.ts: all model-facing text in flow order
- test/: 15 unit tests (parser, disambiguation, sign-off record logic)

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
2026-06-15 18:15:03 +08:00