The Agent Skills Directory

[SAFE]: The skill uses Python scripts that rely exclusively on the standard library. No external dependencies are required or installed.
[SAFE]: Data processing is limited to structural validation of local JSON and Markdown files. The scripts include explicit protections against directory traversal (checking for absolute paths and '..' segments) and symbolic link exploitation.
[SAFE]: The CLI tools implement resource constraints, such as maximum file size (2MB), maximum JSON nesting depth (30 levels), and maximum item counts (5,000), preventing resource exhaustion attacks.
[SAFE]: The skill explicitly forbids and avoids dangerous functions like eval(), exec(), pickle, or any form of shell command execution. All file operations are performed using safe pathlib abstractions.
[SAFE]: The instructions and scripts emphasize a 'human-in-the-loop' model, with clear boundaries stated in SKILL.md and references/ files to ensure the AI does not attempt to make unauthorized legal or compliance determinations.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill handles potentially sensitive organizational evidence, but it does so entirely within the local environment. It has no network capabilities and provides no mechanism for exfiltrating data. (Internal Severity: SAFE)
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied JSON manifests and records. While this represents an attack surface, the scripts only perform deterministic structural checks and do not use the data in a way that could influence agent behavior beyond reporting on the data's structure. (Internal Severity: LOW)

iso-standards-readiness