The Agent Skills Directory

[SAFE]: The skill operates primarily on local files to perform meaning-preserving rewrites of policies and reports. It follows security best practices by requiring the recording of source file hashes (SHA-256) and creating owner-only backups before any modifications occur, as detailed in the 'Baseline and ledger contract'.\n- [SAFE]: All included scripts (check_code_rules.py, check_evals.py, lint_writing.py, and validate_skill.py) are focused on static analysis and schema validation. These scripts analyze local files for length, nesting, and writing style without performing network operations or executing external code.\n- [SAFE]: Although the 'Complete documentation inventory' procedure allows for fetching remote content, it is restricted to official documentation sources and requires recording metadata (hashes and sizes) to prevent untrusted data from silently influencing the process. The skill explicitly warns against storing credentials in remote URLs.\n- [SAFE]: No evidence of prompt injection, obfuscation, persistence, or privilege escalation was found. The instructions emphasize 'secret safety' and the use of 'boundary markers' to prevent accidental exposure of sensitive information during the rewrite process.

meaning-preserving-rewrite