The Agent Skills Directory

[SAFE]: The skill uses only the Python standard library for its processing logic, avoiding any external package dependencies or remote script execution vectors.
[SAFE]: Input validation implemented in scripts/_common.py enforces strict limits on file size (2MB), nesting depth (30), and total item counts (5000), which effectively prevents resource exhaustion attacks via malicious JSON inputs.
[SAFE]: Path validation in scripts/validate_evidence_manifest.py and scripts/_common.py prevents path traversal and symlink attacks by ensuring all evidence files are regular files, use permitted extensions (.json, .md), and are relative to a defined base directory.
[SAFE]: The JSON parsing logic in scripts/_common.py is configured with object_pairs_hook to reject duplicate keys and parse_constant to reject non-finite numbers (NaN/Infinity), mitigating common parser-level exploitation techniques.
[SAFE]: The skill explicitly defines its boundaries, prohibiting network requests, shell execution of user input, and dynamic evaluation (eval/exec), maintaining a strong security posture for a tool that processes structured data.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted JSON files supplied by users to generate reports. However, the risk of indirect injection influencing the agent is mitigated by the structural nature of the validation and the skill's design, which emphasizes that all outputs are draft preparation material for human-in-the-loop review.
Ingestion points: User-provided JSON manifests and registers are loaded via load_json() in scripts like gap_analyzer.py and audit_document_records.py.
Boundary markers: The SKILL.md and generated reports include explicit disclaimers and lane discipline rules to prevent the agent from misinterpreting structural checks as final compliance results.
Capability inventory: Capabilities are limited to local file system read/write and standard JSON/text processing; no dangerous execution or network tools are utilized.
Sanitization: The skill uses strict type checking, regex-based placeholder detection, and length limits for all text fields to ensure data remains within expected bounds.

iso-standards-readiness