FlowGuard: Skill for Codex

FlowGuard UI Flow Structure

Purpose

Admit content, model UI event x state, and prove runnable visible surfaces.

Entrypoint Scope

This standalone FlowGuard satellite skill owns ui_flow_structure (public_owner) UI admission/evidence, not styling or code structure.

Local Material Routing

Read references/ui_flow_structure_protocol.md for surface, journey, structure, text hierarchy, operability, evidence, and geometry.

Entrypoint Acceptance Map

Accept admission, UI state/task/control coverage, and current evidence; block leaks, blindspots, or stale proof.

Use When

• Use for UI content, controls, tasks, runnable evidence, hierarchy, or product language.

Do Not Use When

• Do not style first, invent roles, or treat observation as permission; return non-UI work to flowguard.

Required Workflow

1. Declare greenfield, source_based, or mixed; observation never grants display permission.
2. Classify only user_visible, user_on_demand, or internal; require task/state/recovery need and model journey, structure, text, and operability.
3. For product claims compare canonical typography, components, navigation, feedback, recovery, and transitions. Reuse existing business ids and keep them internal.
4. For complete coverage, freeze surface-inventory revision/fingerprint/discovery evidence and each visible item's stable id/content fingerprint before mapping.
5. Give every visible control, display, transition, overlay, recovery path, and blindspot exactly one modeled or scoped disposition; a visible item cannot be both modeled and scoped.
6. Bind claims to inventory, screenshot/DOM/event/result evidence, click chains, blindspots, and tests.

Hard Gates

• Model-purpose gate: pre-build/change freeze task-specific failure(s)/boundary; bind candidate to native good/bad-per-failure/oracle/current evidence. Reusable types are not fixed-purpose: no mode/fallback; only FlowGuard-declared checks may support completion claims.
• Require the real FlowGuard check engine and AGENTS.md managed record; forbid fake mini-frameworks.
• Unclassified/internal content cannot render; do not add audience/role/persona categories.
• On-demand stays hidden until an accessible task-owned reveal/return; hover has keyboard/focus parity.
• Every reachable enabled action needs control -> event -> owner -> function -> UI update -> evidence, pure-UI disposition, or blindspot.
• Missing inventory/item fingerprints, discovery evidence, unmapped visible items, and conflicting modeled/blindspot dispositions block complete UI coverage.
• Repeated surfaces for one intent reuse its commitment, path, and semantic grammar. Presentation exceptions never change intent, path, visibility, or result.
• Runnable UI needs current source/recovery/freshness evidence; design alone cannot prove it.
• Only explicit reuse/publication or proven cross-project use triggers strict risk_template_library closure.
• Planning context becomes UI behavior only through explicit behavior-source admission.