The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses pnpm exec tessl to run a local linter and review tool. This is a standard development practice for validating skill structure and quality within this specific platform ecosystem. Evidence: SKILL.md, references/tessl.md.
[EXTERNAL_DOWNLOADS]: The skill mentions external documentation and scripts (e.g., from Vercel Labs or Anthropics) as part of its 'Progressive Disclosure' guidelines. These are treated as trusted references for design patterns rather than runtime code dependencies. Evidence: references/best-practices.md.
[PROMPT_INJECTION]: The skill explicitly warns against and provides tests to detect indirect prompt injection in the skills it audits. It encourages the use of 'boundary markers' and 'sanitization' to prevent agents from obeying instructions embedded in untrusted data. Evidence: evals/discovery-audit-and-skill-handoff-routin/criteria.json.
[DATA_EXFILTRATION]: No patterns of sensitive data exfiltration were detected. The skill's network operations are restricted to authenticated Tessl reviews, which are gated by credit-aware policies and user approval. Evidence: references/tessl.md.
[CREDENTIALS_UNSAFE]: The skill correctly instructs users NOT to hardcode secrets and instead rely on repository-owned authentication or configuration for its audit tools. Evidence: SKILL.md, references/tessl.md.

skill-audit