linkedin-post

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill contains a hardcoded absolute file path (/Users/rami/Documents/life-os/ai-agents-config/skills/rami-voice/SKILL.md) for voice guidelines. Hardcoded absolute paths expose information about the local file system structure and user environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it is designed to process untrusted content and read user-specified local files without sanitization.
  • Ingestion points: The skill ingests raw text from articles, essays, and the full content of local files specified by the user in SKILL.md.
  • Boundary markers: There are no boundary markers, delimiters, or explicit instructions provided to the agent to treat the ingested content as data rather than instructions.
  • Capability inventory: The skill explicitly instructs the agent to "read the file first" if a file path is provided, leveraging the agent's file system access capabilities on arbitrary user input.
  • Sanitization: No sanitization, filtering, or validation is performed on the ingested content before it is processed by the model to generate the final post.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 10:16 PM
Security Audit — agent-trust-hub — linkedin-post