The Agent Skills Directory

[SAFE]: The skill operates entirely within the local environment using the Python standard library, adhering to the security model declared in the frontmatter and documentation.
[DATA_EXPOSURE_AND_EXFILTRATION]: No network libraries or credential-handling logic were found. Input validation in scripts/_common.py explicitly rejects URL-like paths and symlinks, preventing unauthorized data access or exfiltration.
[REMOTE_CODE_EXECUTION]: The scripts perform deterministic structural validation and do not utilize dangerous functions like eval(), exec(), or subprocess calls with external data.
[INDIRECT_PROMPT_INJECTION]: While the skill processes local JSON, CSV, and Markdown data, it implements strict schema validation and string sanitization (HTML and Markdown escaping) in its scaffolding tools to mitigate content-based injection risks.
[OBFUSCATION]: Analysis of all scripts and assets revealed no signs of Base64 encoding, zero-width characters, homoglyphs, or other obfuscation techniques.
[PRIVILEGE_ESCALATION]: File operations are restricted to existing local directories, and the skill uses atomic writes with private file permissions (0600) for generated reports, preventing unauthorized file system modification.

hypothesis-generation