opl-stage-candidate-portfolio-reviewer

name	opl-stage-candidate-portfolio-reviewer
description	Use when reviewing OPL stage candidate portfolio refs-only projections, candidate ref coverage, assumption/provenance/negative-path/advisory-metric refs, human-review route-back, missing refs, and domain-authority overclaim risks.

OPL Stage Candidate Portfolio Reviewer

Use this skill to review stage_candidate_portfolio read models and projection packets. Keep the name and scope precise: this is not a research hypothesis portfolio skill. It reviews OPL-owned body-free stage candidate refs, routing, and advisory signals while the domain agent keeps candidate body, domain truth, quality verdict, artifact authority, and owner receipt authority.

OPL Owner Boundary

• Treat stage-candidate-portfolio.schema.json, Foundry Lab read models, stage packs, domain owner refs, human-review refs, and owner receipts as the authority for recorded state.
• Treat this Skill as the AI-first reviewer for candidate-ref coverage, assumption/provenance/negative-path readability, advisory metric restraint, human-review route clarity, and overclaim prevention.
• Do not read or reconstruct candidate bodies, write domain truth, accept or reject candidates, rank candidates as truth, sign owner receipts, create typed blockers, mutate artifacts, or declare domain/source/publication/runtime readiness.
• A stage candidate portfolio review is not a hypothesis quality verdict, research novelty verdict, owner acceptance, human-gate decision, stage admission, or production readiness claim.

AI-first / Contract-light Semantics

• Use portfolio schemas, indexes, and modules only for candidate identity, body-free refs, owner routes, receipts, recovery, and verification.
• Keep elastic portfolio review in this Skill: judge ref coverage, advisory metric restraint, negative-path preservation, human-review fit, and route-back quality.
• If the read model cannot support a claim, classify the missing or stale ref instead of encoding candidate judgment in a module.

Cross-Domain Failure Patterns

Use the shared pattern ids in ../../references/stage-cross-domain-failure-patterns.md as portfolio review heuristics, not new portfolio body rules. Map them to candidate-ref coverage, advisory-metric restraint, human-review route clarity, and owner-route gaps.

Inputs

• stage_candidate_portfolio packets or stage_candidate_portfolio_summary read models.
• Related stage context refs, evidence-pack refs, candidate refs, assumption refs, provenance refs, negative-path refs, advisory metric refs, human-review refs, owner refs, and route-back refs.
• Domain policy, Stagecraft, Foundry Lab, Atlas, runtime, or owner-evidence refs when they explain a missing or stale projection edge.

Workflow

1. Identify portfolio_id, target_domain_id, portfolio_status, candidate count, owner refs, and authority boundary.
2. Confirm the packet stays body-free:
   • every candidate, evidence, assumption, provenance, negative-path, advisory, and human-review entry carries refs/status only;
   • body_included is false everywhere;
   • no candidate_body, domain_body, evidence_body, body, content, payload_body, or artifact_body field is present.
3. Review candidate coverage:
   • each candidate has candidate_ref, origin_ref, owner_ref, and evidence refs when evidence is required;
   • rationale refs are present when route selection depends on them;
   • candidate statuses do not imply domain acceptance unless an owner receipt ref says so.
4. Review assumption, provenance, and negative-path refs:
   • material assumptions have owner refs and support or contradiction refs;
   • provenance checks name source refs, source status, and missing/failed checks;
   • negative paths preserve failed evidence, duplicate/adverse findings, and route-back reasons instead of hiding them behind advisory ranking.
5. Review advisory metrics:
   • ranking, proximity, novelty, coverage, or score refs are treated as ordering or briefing signals only;
   • advisory metrics never suppress missing evidence, missing owner review, stale provenance, failed paths, or human gate requirements.
6. Review human-review route:
   • required review requests, decisions, owner routes, and route-back refs are present;
   • missing refs are named as evidence gaps, not converted into quality verdicts.
7. Recommend the smallest legal repair: add missing ref, refresh stale ref, route to domain owner, request human review, update Stagecraft/Foundry Lab projection, or hold as operator attention.

Finding Classes

• domain_body_leak;
• authority_boundary_overclaim;
• candidate_ref_gap;
• assumption_ref_gap;
• provenance_ref_gap;
• negative_path_gap;
• advisory_metric_overclaim;
• human_review_route_gap;
• owner_route_gap;
• stale_or_missing_read_model_ref;
• critique_as_repair_hint_overclaim;
• source_or_receipt_stale;
• candidate_body_reconstruction_forbidden;
• no_issue_found.

Output Shape

Return:

• stage_candidate_portfolio_ref;
• stage_candidate_portfolio_summary_ref;
• reviewed_candidate_refs;
• finding_class;
• missing_or_stale_refs;
• advisory_metric_review;
• human_review_route;
• owner_route;
• recommended_delta;
• severity_recommendation: operator_attention, route_back, or owner_gate_needed;
• authority_boundary: no candidate bodies, no domain truth, no candidate acceptance, no quality verdict, no owner receipts, no typed blockers, no artifact mutation, no readiness claim.