Tech-debt-skill automates codebase debt audits

This is a sharp answer to a real failure mode in AI code review: shallow, uncited feedback that sounds thorough and gets ignored. The value is less “AI writes a report” and more “AI is forced to ground the report in the codebase and preserve it as an artifact.”

• –It scopes audits across architecture, tests, dependencies, performance, security, observability, and docs, which makes it materially more useful than a one-pass lint summary
• –Stack-aware tool selection is the right move: JS/TS, Python, Rust, and Go each get their own analyzer set, so the report should be more credible than language-agnostic prompts
• –The repeat-run baseline is the strongest part operationally; marking findings as `NEW` or `RESOLVED` makes debt review trackable over time
• –The tradeoff is obvious: this is only as good as the analyzers and the repository conventions it can infer, so it works best as a disciplined audit assistant, not an oracle
• –The source video frames it as one of many Claude Code skills, which makes it feel like an ecosystem play rather than a standalone product with a broad user-facing launch