investigate-pi-only — WorksGood live mirror

Metadata

Status	abandoned
Assigned	`agent-824`
Created	2026-07-26T06:54:22.017021517+00:00
Started	2026-07-26T07:15:24.085539805+00:00
Completed	2026-07-26T07:15:46.495016374+00:00
Tokens	0 in / 0 out
Failure reason	Cleanup: zero-token/no-output run incorrectly marked pending-eval; superseded by direct diagnosis.

Description

Investigation

Determine exactly what changed when WG moved to a Pi-only model/configuration plane, with special attention to chat sessions and worker execution.

Answer with source- and history-backed evidence:

Were direct claude: and codex: executors actually removed from dispatch, or only removed from the supported/authoritative configuration surface?
What does commit/task make-pi-the-2 (and any follow-ups) change, and why?
How are Claude and OpenAI Codex models used now through Pi (for example pi:anthropic:... and pi:openai-codex:...)?
What is the current live project route for chat, workers, and agency roles, and how does that differ from this current Pi session environment?
Identify compatibility/dead-code residue and documentation drift, especially the conflict between wg quickstart / docs/COMMANDS.md and root AGENTS.md / older source paths.
State whether the observed shift is intentional, partial, or accidental, and list concrete corrective options if the product intent is unclear.

Do not modify product code or configuration. Produce a concise report in the task log/session summary with relevant commit hashes and file/line references.

Validation

Inspect the implementation and git history around 295486e6 / task make-pi-the-2 and follow-up commits.
Distinguish handler/executor, provider/model, chat route, worker route, and one-shot agency route.
Verify current effective routes from WG graph/config state without changing them.
Reconcile apparently contradictory CLI/docs/source evidence and explicitly label stale compatibility surfaces.
Report findings only; no source or config files changed.

## Investigation
Determine exactly what changed when WG moved to a Pi-only model/configuration plane, with special attention to chat sessions and worker execution.

Answer with source- and history-backed evidence:
- Were direct `claude:` and `codex:` executors actually removed from dispatch, or only removed from the supported/authoritative configuration surface?
- What does commit/task `make-pi-the-2` (and any follow-ups) change, and why?
- How are Claude and OpenAI Codex models used now through Pi (for example `pi:anthropic:...` and `pi:openai-codex:...`)?
- What is the current live project route for chat, workers, and agency roles, and how does that differ from this current Pi session environment?
- Identify compatibility/dead-code residue and documentation drift, especially the conflict between `wg quickstart` / `docs/COMMANDS.md` and root `AGENTS.md` / older source paths.
- State whether the observed shift is intentional, partial, or accidental, and list concrete corrective options if the product intent is unclear.

Do not modify product code or configuration. Produce a concise report in the task log/session summary with relevant commit hashes and file/line references.

## Validation
- [ ] Inspect the implementation and git history around `295486e6` / task `make-pi-the-2` and follow-up commits.
- [ ] Distinguish handler/executor, provider/model, chat route, worker route, and one-shot agency route.
- [ ] Verify current effective routes from WG graph/config state without changing them.
- [ ] Reconcile apparently contradictory CLI/docs/source evidence and explicitly label stale compatibility surfaces.
- [ ] Report findings only; no source or config files changed.

Depends on

(none)

Required by

abandoned .flip-investigate-pi-only

✉ Messages 1 message

#1user2026-07-26T07:01:22.322122753+00:00delivered

Please explicitly assess the architectural distinction between a Pi-owned/default model plane and Pi as the only executor/runtime. Evaluate a Pi-first design that keeps chat on Pi but permits direct Claude/Codex worker adapters only by explicit per-task/profile selection, with no implicit cross-system fallback or WG-owned model catalog.

Log

2026-07-26T06:54:22.006565013+00:00 Task paused
2026-07-26T06:54:27.773801176+00:00 Task published
2026-07-26T07:15:24.211994062+00:00 Spawned by coordinator --executor pi --model zai:glm-5.2 --isolation required-worktree
2026-07-26T07:15:46.495021415+00:00 Task pending required evaluation gate (agent reported done; awaiting exact attempt-bound `.flip-*`/`.evaluate-*` verdicts)
2026-07-26T07:15:54.193547845+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:16:17.537134180+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:16:43.433855215+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:17:06.951179521+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:17:30.630216203+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:17:52.796365644+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:17:57.798052544+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:18:14.610303387+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:18:19.991372490+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:18:39.765628849+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:18:44.712163298+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:19:03.247331004+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:19:08.234077849+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:19:23.899683519+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:28:04.759674345+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:28:09.766841651+00:00 Repaired evaluation pipeline drift for authoritative source attempt 1 (evalp-dea85b96a430b107387bb7e1)
2026-07-26T07:57:06.579999494+00:00 Pinned historical PendingEval as a required gate at evaluator threshold 0.70
2026-07-26T07:58:03.547105606+00:00 Task abandoned: Cleanup: zero-token/no-output run incorrectly marked pending-eval; superseded by direct diagnosis.