| name | flowguard-model-topology-hazard-review |
| description | Use when a locally green FlowGuard model needs topology-grounded future-use hazard review for broad claims, business paths, old/new disposition, side effects, terminals, loops, external boundaries, or parent/child compression. |
FlowGuard Model Topology Hazard Review
Purpose
Infer actionable future-use hazards from the actual model topology and usage intent; keep unanchored AI concerns observation-only.
Entrypoint Scope
Route id: model_topology_hazard_review; role: public_owner; native owner: model_topology_hazard_review. This standalone FlowGuard satellite skill owns topology-anchored risk routing, not generic brainstorming.
Local Material Routing
Read references/topology_hazard_protocol.md for TopologyDigest, UsageIntent, business-path identity, anchors, dispositions, and completion rules.
Entrypoint Acceptance Map
Accept a current topology digest, usage intent, and evidence boundary; promote only anchored hazards; block unresolved high-impact paths/loops/side effects; hand model, test, reduction, process, and risk work to typed owners.
Use When
- Use before broad done/release/publish confidence when local green may hide duplicate/conflicting paths, broad terminals, repeatable side effects, compatibility paths, or closure/liveness hazards.
Do Not Use When
- Do not use for generic risk lists, unmodeled systems, or as a replacement for maturation, alignment, Risk Evidence Ledger, or Architecture Reduction; return unclear topology to
model-first-function-flow.
Required Workflow
- Record usage, scope, topology, business paths, evidence, and gaps. Portable temporal claims bind the digest to the exact
flowguard.portable_model.v1 fingerprint and executable obligations.
- For each candidate, name the topology anchor, real-use failure, affected element, confidence effect, and disposition.
- Resolve, scope with rationale, or issue typed owner-route handoffs and maintenance obligations.
Hard Gates
- Model-purpose gate: before build/change, freeze this instance's task-specific failure(s) and boundary; then bind candidate plus native good/bad-per-failure/oracle/current evidence. Reusable types are not fixed-purpose; no mode/fallback; SkillGuard only supervises FlowGuard-declared checks.
- Verify the real FlowGuard check engine and AGENTS.md managed record; never create a fake mini-framework.
- Unanchored concerns cannot block confidence; anchored hazards need current evidence, owner route, or explicit scoped disposition.
- Important path conflicts, loop liveness, compatibility/history, and template harvest closure must remain visible before broad confidence.
- Portable liveness/fairness requires canonical checker evidence for the same graph; prose/metadata and stale or truncated reports cannot pass.
Output Requirements
- Return
evidence, failures, blockers, skipped_checks, residual_risk, claim_boundary, and typed_next_actions, plus anchored candidates and confidence effects.
SkillGuard Maintenance
- Edit
.skillguard/contract-source.json, then regenerate derived contracts; SkillGuard validates hazard structure and cannot promote unanchored prose into proof.
