By name: flowguard-model-topology-hazard-review description: Use when a locally green FlowGuard model still needs future-use hazard review from topology, usage intent, claim scope, old/new path disposition, business-path risk, side effects, terminals, loops, or parent/child compression.
FlowGuard Model Topology Hazard Review
Standalone FlowGuard satellite skill for topology-grounded future-use hazards.
Return to model-first-function-flow when topology, intent, or ownership is unclear.
First Read
- Route id:
model_topology_hazard_review. - Helpers:
infer_topology_digest(),infer_topology_hazard_plan(),review_topology_hazards(),UsageIntent,BusinessPathIdentity,TopologyHazardCandidate. - Reference:
references/topology_hazard_protocol.md.
Hard Gates
- Verify the real package before claiming FlowGuard use.
- Keep the AGENTS.md managed block/version record current, or record why not.
- Do not create a fake mini-framework.
- Read topology before naming hazards; this is not a fixed risk checklist.
- Unanchored AI concerns are observations only.
- Hazard models need template-harvest closure.
Inputs
TopologyDigest: states, edges, side effects, terminals, old/new paths, business paths, external boundaries, parent/child compression, landmarks.BusinessPathIdentity: intent, trigger, terminal, writes, side effects, equivalent/exclusive/superseded paths, compatibility disposition, evidence.UsageIntent: use mode, final claim, history/compatibility, and goal.- Current evidence: model pass, tests, replay, runner summary, skipped gaps.
Review Rule
Every hard hazard names:
- Topology anchor.
- Future real-use failure mode.
- Affected state, edge, side effect, terminal, legacy path, business path, or external boundary.
- Disposition: model/test work, Risk Evidence Ledger, compatibility, scope, or blocker.
Generic warnings cannot block confidence without a topology anchor.
Route Handoffs
- Coarse terminal, hidden post-success state, parent/child compression:
model_maturation_loop. - Model obligation or ordinary test gap:
model_test_alignment. - Broad final confidence or external proof gap:
risk_evidence_ledger. - Release/process/local-only evidence gap:
development_process_flow. - Old/new path or compatibility disposition:
architecture_reduction,risk_evidence_ledger. - Duplicate/conflicting/unproven business path:
architecture_reduction,model_similarity_consolidation,model_maturation_loop,model_test_alignment,risk_evidence_ledger.
Prompt
Use the protocol/template for full prompts. Output candidates with anchors, rationale, future failure, disposition, route, confidence effect, and a status note.
Non-Goals
- Do not replace Model Maturation, Model-Test Alignment, Risk Evidence Ledger, DevelopmentProcessFlow, or Architecture Reduction.
- Do not run LLM calls inside the Python package helper.
Purpose
Bind this FlowGuard route to one work contract, native checks, evidence, blockers, residual_risk, and claim_boundary.
Entrypoint Scope
Covers flowguard-model-topology-hazard-review and routed local materials only; no unrelated repos, private paths, services, publication, or release claims unless separately routed.
Local Material Routing
Use FlowGuard's native router, package/model checks, .skillguard/work-contract.json, check_manifest, and run records; keep public text portable.
Entrypoint Acceptance Map
Mode is native-integrated/hybrid as declared; SkillGuard executes gates around the native owner and must not add a second execution route.
Use When
Use when this skill is selected and the task needs governed route, evidence, check, handoff, or closure behavior.
Do Not Use When
Do not use outside the skill domain, without required materials, when a more specific skill owns the work, or for tiny direct answers.
Required Workflow
Select the FlowGuard-owned route, open/compile the contract, start/update run record, run native model/check gates, refresh evidence, fix blockers, then close from current checks.
Hard Gates
Block skipped phases, stale/prose-only evidence, hollow contracts, quality downgrades, native-route conflicts, and completion claims with blockers.
Output Requirements
Report target, route, evidence, failures, blockers, skipped_checks, residual_risk, and claim_boundary; separate checked facts from judgment.
SkillGuard Maintenance
Refresh contracts, checks, evidence, and installed copies after entrypoint, route, evidence, or closure changes.