Outcomes
- Eight audit commands plus /audit-fix, public and installable via the Clownware plugin marketplace
- Assessment-only by design: ranked findings with file:line citations, verified against the code before reporting
Stack
Context
Universal dev-workflow skills packaged as a Claude Code plugin: an audit
suite — /arch-audit, /security-audit, /test-audit, /a11y-audit,
/deps-audit, /perf-audit, /skill-audit, /design-audit — plus
/audit-fix, ADR authoring, PR descriptions, root-cause debugging, and test
scaffolding. One marketplace add puts the whole toolkit in your agent.
The Problem
AI-assisted codebases ship fast and rot fast. The gap isn’t generation — it’s assessment: knowing what an agent-built app actually contains before you trust it. This plugin is the productized form of the discipline layer I run in the day job — hardening AI-built apps other people started — packaged so any repo gets the same scrutiny.
My Role
Designer and builder — the audit prompts encode the checklist I actually run.
Approach
- Assessment-only by design. Audits report and rank; they don’t patch.
Fixes are left to you — or to
/audit-fix, which applies the mechanical ones and re-verifies using the report’s own methods. - Findings are verified before they’re reported. Every finding carries file:line citations and is checked against the code before it reaches the report — plausible-but-wrong findings die in verification, not in your backlog.
- Probe, don’t assume. The skills inspect the repo they run in and degrade gracefully — they work on codebases that don’t follow Clownware conventions.
- One marketplace, four plugins. Distribution rides the Clownware plugin marketplace: universal tools here, product-coupled plugins in their own repos.
The discipline running through this approach is the site’s thesis: Architecture Is the Interface Now.