opl-incident-root-cause-triager

name	opl-incident-root-cause-triager
description	Use when triaging OPL stalls, currentness drift, runtime/readback mismatch, heartbeat alerts, provider failures, owner-route gaps, repeated blocker/nonprogress loops, conflict/blocker envelopes, duplicate tasks, human gates, retry/dead-letter states, or false progress/readiness incidents. Applies L0-L4 root-cause depth and routes repair without writing authority surfaces.

OPL Incident Root Cause Triager

Use this skill to move an incident report from surface status to an owner-routable root-cause brief. It is for diagnosis and routing only.

Boundary

• Treat runtime/readback, provider receipts, owner receipts, typed blockers, domain truth, artifact authority, and release authority as owned by their real surfaces.
• Use this skill only to classify evidence, identify the failing boundary, and prepare repair or owner-route options.
• Do not write owner receipts, typed blockers, domain truth, artifact authority, runtime queues, provider state, readiness claims, release claims, or production claims.
• Do not close an incident by restating labels such as blocked, idle, queue_empty, no_live_session, stale, duplicate_task, human_gate, or dead_lettered without the boundary and owner route.

Optional helper: kernel.py provides stdlib-only deterministic incident-ref normalization, L0-L4 triage skeleton/checklist builders, and authority-phrase linting. It is local support only: no file/network/subprocess use and no owner receipt, typed blocker, runtime queue, release, readiness, or production authority.

Depth Ladder

• L0_symptom: visible state, alert, failure message, stalled lane, stale projection, or missing heartbeat.
• L1_direct_boundary: command, gate, queue, provider, projection, owner route, contract, dependency, artifact, or evaluator that emitted the symptom.
• L2_cross_surface_evidence: neighboring source that proves whether the boundary is current, stale, conflicting, or missing.
• L3_owner_repair_path: owner surface, allowed and forbidden write set, legal next action, verification/readback, and stop condition.
• L4_prevention: prompt, runbook, skill, contract, validator, automation, or workflow change that would prevent recurrence.

Workflow

1. Classify incident kind: stall, currentness_drift, runtime_mismatch, heartbeat_alert, provider_failure, owner_route_gap, blocker_loop, nonprogress_loop, conflict_envelope, duplicate_task, evidence_blocker, quality_blocker, human_gate, retry_or_dead_letter, or false_progress_claim.
2. Record the L0 symptom exactly, then find the L1 boundary that produced it.
3. Check at least one L2 neighboring surface when the incident concerns currentness, runtime, owner route, readiness, repeated stalls, or heartbeat behavior.
4. Classify root cause as one or more of:
   • target_artifact_gap;
   • gate_or_evaluator_defect;
   • read_model_currentness_drift;
   • owner_route_or_authority_gap;
   • runtime_or_control_plane_defect;
   • provider_or_environment_failure;
   • legitimate_human_gate.
5. For conflict/blocker envelopes, review subject identity, idempotency key, classification, owner, allowed next action, forbidden action, retry/dead-letter semantics, and operator question without resolving the blocker by AI.
6. For stop-loss or nonprogress cases, name the repeated action pattern, why it is not progress, the minimal legal next action, and the stop condition that prevents another loop.
7. For L3, name the next owner, legal action, expected artifact or readback, verification command/ref, and stopping condition.
8. Add L4 prevention only when the incident is repeated, caused a false claim, or exposed a reusable workflow gap.

Legacy Coverage

This triager covers the retired opl-conflict-blocker-resolution-reviewer and opl-stop-loss-and-nonprogress-reviewer entries. Conflict envelopes, blockers, stop-loss loops, and nonprogress loops should route through the same L0-L4 owner-routable root-cause review.

Output Shape

Return:

• incident_kind;
• root_cause_depth: L0 through reached level;
• evidence_refs;
• root_cause_class;
• blocker_to_owner_map;
• conflict_or_stop_loss_review;
• legal_next_action;
• verification_or_readback;
• stop_condition;
• prevention_delta;
• no_authority_caveat: no owner receipts, no typed blockers, no domain truth, no artifact authority, no runtime queues, no readiness/release/production claims.