content-hooks
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [DATA_EXFILTRATION]: The skill hardcodes an absolute local file path (
/Users/rami/Documents/life-os/ai-agents-config/skills/rami-voice/SKILL.md) for voice guidelines. While this appears to be a functional reference to a shared configuration, it discloses internal directory structures and user-specific naming conventions. - [INDIRECT_PROMPT_INJECTION]: The skill processes external transcript data, which presents an attack surface for indirect prompt injection if the source content contains malicious instructions.
- Ingestion points: Reads a transcript markdown file and a
research.mdfile as the primary source for content generation. - Boundary markers: The instructions include a preprocessing step to strip YAML frontmatter and specific markdown sections, which acts as a rudimentary boundary, but it lacks explicit delimiters to distinguish data from instructions for the main content body.
- Capability inventory: The skill performs web searches and writes a new file (
hooks.md) to the local filesystem. - Sanitization: Performs basic markdown cleaning (stripping wiki-links and specific headers) but does not perform content-level sanitization to filter out potential embedded instructions.
Audit Metadata