Mission Seed — M3: Operating-Procedures Validate → Triage → Data-Drive

Status: seed (not yet a mission). Feed this to /spec-kitty.specify in a fresh session. Part of: charter-resolution program (see ../program-brief.md). Closes: #2994, #3352, and the operating-procedures channel of #3488. Effort: L. Depends on: M1 (single-authority resolution parity) landed — needs the validated node universe. Independent of M2.

Problem

collaboration.operating-procedures on an agent profile is a schema-validated list[str], but its values are never checked against real doctrine nodes: measured across 16 built-in profiles, 36 of 50 declarations name no node (8 name a wrong-kind node; only 6 name a real procedure node). Because the edge-extractor ignores the field entirely, the real agent_profile → procedure edges are hand-pinned one profile at a time in _CURATED_ARTIFACT_EDGES. So authored procedure references are silently inert, and #3352 (data-drive those edges) cannot be done by blind emission because it would mint 36 dangling edges → assert_valid failure.

Fix approach (hard internal order — do NOT reorder)

  1. Validate loud (first). Add a load-time validator: every operating-procedures entry must resolve to a real doctrine node. Converts the 36 silent fictional refs into loud failures. Cheap, high-value, independent. (Empty-allowlist gate precedent: WP09.)
  2. Triage the 36. Author / repoint / delete each fictional entry (per-artefact review). Decide the field's direction (project into edges vs retire/rename).
  3. Data-drive (last). Teach the extractor to emit agent_profile --requires--> procedure from the field, guarded to targets that resolve to an existing node; retire the hand-pinned _CURATED_ARTIFACT_EDGES entries sourced from operating-procedures. Also complete the still-unwired RECONCILE third trigger edge (tactic:change-apply-smallest-viable-diff).

Open operator decisions (resolve at this mission's discovery)

  • Wire vs deprecate: should operating-procedures become a first-class edge source, or be deprecated/renamed in favour of DRG edges? (Either way, ship the dead-entry diagnostic.)
  • Triage dispositions: the 36 fictional refs need per-entry author/repoint/delete calls — likely a review artifact, not a single decision.

Scope

  • In: the operating-procedures validator, the 36-entry triage, the data-driven extractor emission, retiring the related hand-pins.
  • Out: the broader kind-complete cascade (#2829, that is M5); delivery/render of procedures to the agent (that is M4's procedures[]/step-description work).

Key seams (from the investigation)

  • doctrine/agent_profiles/profile.py (CollaborationContract.operating_procedures)
  • doctrine/drg/migration/extractor.py (extract_artifact_edges; the hand-pin at ~:463)
  • _CURATED_ARTIFACT_EDGES (the per-profile hand-pins to retire)
  • validator precedent: the WP09 empty-allowlist gate

Risk

Data-driving before triage mints 36 dangling edges and fails assert_valid. The order above is load-bearing.

Edit this page