AX Audit — Agent Approval and Control UX

When to use it

Supply an agentic feature diff or explicit product paths. Produces evidence-linked findings and a designer/PM relationship summary, with unknowns distinguished from passes. Useful for approval dialogs, tool-execution dashboards and streaming controls, not generic forms or a security certification. Includes architecture checks to verify UI promises; runtime-only checks require actual interaction evidence. Upstream shipping tiers are suggested team policy. Documentation reviewed only: not installed or runtime-tested. GitHub stars are repository-wide, not skill-specific.

Compatible agents

CodexClaude Code

Installation

From an environment with Node.js, run `npx skills add mblode/agent-skills --skill ax-audit --agent codex claude-code`. Select the intended host and scope, then start a fresh agent session. Preserve the complete skill folder including rules, references and probes where present; do not copy SKILL.md alone. The upstream README documents installing the complete collection too.

Review the source and instructions before installing in your agent.

License

MIT

Last checked Nov 3, 2026.

Pair this skill with a visual reference

A skill guides how your agent works; a design reference shows the direction you want. Choose an example for a specific job, such as the headline hierarchy or form layout, and keep your project's existing design rules explicit.

Start with design references you can inspect, then use our guide to briefing an AI coding agent to turn observations into requirements. If you need implementation components, compare UI libraries for your stack.

These pages are public. Browsing a skill does not install it, and TastefulKit's design-reference MCP tools do not search the AI-skill directory.