Clean Code AI Collaboration Skill v0.3.0
Highlights
- Adds Lightweight, Standard, and Full Audit paths with one-way risk escalation.
- Separates delivery maturity into Prototype and Production-Ready without weakening authorization or risk gates.
- Routes validation through repository-native Formatter, Linter, Static Analysis, Build, Test, Security, Coverage, and Mutation gates according to the actual risk.
- Splits review output into a concise Standard contract and a traceable Full Audit contract.
- Narrows the activation scope and sets the Codex adapter to explicit-only with allow_implicit_invocation: false.
- Documents external-validity expansion across legacy .NET, TypeScript and React, Python, Java and Spring, repository instructions, test quality, models, clients, and blind review.
Evidence boundary
The published benchmark numbers remain the fixed v0.2.0 results. v0.3.0 introduces routing, delivery-readiness, toolchain, and documentation changes, but does not claim new cross-language success rates or Token savings.
Verification
- 38 repository contract and benchmark tests passed.
- Agent Skills validator passed.
- SKILL.md body remains within the 500-word limit at 490 words.
- Existing benchmark manifests and result files were not rewritten.