flowguard-development-process-flow

name	flowguard-development-process-flow
description	Use for staged development, multi-skill work, freshness, sync, release, or final process claims.

FlowGuard Development Process Flow

Purpose

Simulate lifecycle order and freshness without replacing specialists.

Entrypoint Scope

Front-door FlowGuard satellite skill; development_process_flow (public_owner) owns execution_freshness and conditional internal process optimization, with plan_detailing and agent_workflow delegates.

Local Material Routing

Read references/development_process_flow_protocol.md; load references/process_optimization_protocol.md only for an optimization reason and references/failure_triage_protocol.md only for non-pass triage.

Entrypoint Acceptance Map

Order modes, keep ordinary work light, compare hard-equivalent candidates conditionally, enforce freshness, and delegate gaps.

Use When

• Use for rough plans, staged work, multiple skills, background checks, artifact versions, payload schemas, sync, release, or broad claims.

Do Not Use When

• Do not replace specialists; return unclear routing to model-first-function-flow.

Required Workflow

1. Register modes, artifacts, actions, targets, writes, evidence, peer changes, and a bounded provider package.
2. Activate optimization only for explicit_request, multiple_equivalent_routes, material_rework_risk, or diagnostic_boundary_choice; otherwise return not_needed with no ceremony.
3. When active, prove equal outcome, obligation/evidence, safety, protected side effects, dependency authority, and execution owner. Choose targeted|declared_complete|budgeted diagnosis plus sequential|safe_parallel execution.
4. Reuse TestMesh and Finding Ledger evidence; group only related findings, repair the primary owner/root cause, and revalidate affected obligations. Hard blockers stop invalid descendants; material evidence stales the decision.
5. Route non-pass results, scan owners, preserve peer writes, track freshness domains separately, freeze final inputs, and use one owner per heavy check.

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.
• Use the real FlowGuard check engine and AGENTS.md managed record; never create a fake mini-framework.
• Keep provider tasks and sibling semantics with native owners; progress is not proof.
• Estimated comparison supports a preference; measured finite comparison may support a bounded minimum. Never claim global optimality.
• Ordinary work has no optimizer records; repairs need relation, owner, and affected-revalidation evidence.

Output Requirements

• Return evidence, failures, blockers, skipped_checks, residual_risk, claim_boundary, typed_next_actions, modes, freshness, and required revalidation. Include optimization details only when active; diagram edges mean order, invalidation, or required revalidation.

SkillGuard Maintenance

• Edit current source, regenerate current outputs, and remove former authority. SkillGuard cannot manufacture evidence.