Task · SKT-0153.05

Model a multi-service agent pipeline with review loops and terminal outcomes

Description

doc-0004 W4: spike the seam (app service nodes plus an Agent Observability binding preferred), then build the generic pipeline: orchestrator, retrieval, draft, gated review with bounded revision, conditional stage, terminal outcome, workflow steps, ratings, per-language tag channels. Same owner as the hook-call task.

Acceptance Criteria

Definition of Done

Implementation Plan

Reviewed app-first spike and exact file inventory precede code. Deliver neutral declarations and rating ABI first, app-only execution next, then shared namespace and receipt-backed root integration. The parent remains open until actual wire, placement and outcome proofs pass.

Implementation Notes

Owner requested immediate loop closeout; work is parked and no continuation is authorised by background CI events. Preserve delivered source and partial work; resume only under an explicit new instruction, reconciling current Git, board, runtime and the recorded acceptance blockers. Engine hosted AMD64 race/e2e result remains pending at closeout; no full-estate completion or runtime/live activation claim.

References

View the source file on GitHub