M8 — Orchestration, error handling, validation & eval #9

Closed
opened 2026-07-24 00:53:36 +00:00 by david · 0 comments
Owner

Goal: wire the stages into one autonomous run and validate end-to-end.

Tasks

  • Sequential orchestrator: Jira fetch → implement → review → remediate → MR → write-back.
  • Enforce the Error-handling policy table (fail-fast setup/fetch; bounded retries on transient calls; per-stage degrade; abort+dispose+persist on agent failure; non-fatal write-back; dispose() in finally).
  • Persist all artifacts under .pi-loop/runs/<run-id>/; on failure write error.json.
  • Final summary: MR URL, remainingFindings, degradedStages.
  • Gated E2E (PILOOP_E2E=1) against a throwaway ticket + sandbox repo/branch — assert on artifacts/exit conditions.
  • Light eval checkpoint: run the pipeline on 2–3 representative tickets, review wrapper prompts + skill composition output quality, tune as needed.
  • Document setup + usage (feeds a future README).

Depends on: M0–M7.

Validation: one command (npm start -- JIRA-123) runs all stages and yields a Draft MR + Jira write-back, with artifacts on disk; gated E2E passes; sample-ticket eval acceptable.

**Goal:** wire the stages into one autonomous run and validate end-to-end. ### Tasks - [ ] Sequential orchestrator: Jira fetch → implement → review → remediate → MR → write-back. - [ ] Enforce the Error-handling policy table (fail-fast setup/fetch; bounded retries on transient calls; per-stage degrade; abort+dispose+persist on agent failure; non-fatal write-back; `dispose()` in `finally`). - [ ] Persist all artifacts under `.pi-loop/runs/<run-id>/`; on failure write `error.json`. - [ ] Final summary: MR URL, `remainingFindings`, `degradedStages`. - [ ] Gated E2E (`PILOOP_E2E=1`) against a throwaway ticket + sandbox repo/branch — assert on artifacts/exit conditions. - [ ] Light eval checkpoint: run the pipeline on 2–3 representative tickets, review wrapper prompts + skill composition output quality, tune as needed. - [ ] Document setup + usage (feeds a future README). **Depends on:** M0–M7. **Validation:** one command (`npm start -- JIRA-123`) runs all stages and yields a Draft MR + Jira write-back, with artifacts on disk; gated E2E passes; sample-ticket eval acceptable.
david closed this issue 2026-07-25 04:06:30 +00:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
david/pi-loop#9
No description provided.