The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts such as audit_manual.py and seal_manual.py using the subprocess module. These executions are limited to the skill's own internal logic and the local Python interpreter, without using shell=True or processing unvalidated external commands.
[DATA_EXPOSURE]: The skill reads repository files and technical manuals to perform its function. It implements comprehensive directory traversal protections, specifically using the _safe_source function in check_manual.py and the _within function in _skill_protocol_runtime.py, which ensure that file access is restricted to the intended project directories and cannot access sensitive system paths.
[PROMPT_INJECTION]: The skill is designed to process untrusted source documents and Markdown manuals. It mitigates indirect prompt injection risks by using a deterministic validation protocol and explicit grounding instructions that treat source data as evidence rather than instructions. It also uses SHA-256 integrity checks for all bound sources to detect unauthorized changes to inputs.

manualize