rami-voice

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill explicitly instructs the agent to read a specific local file path: /Users/rami/Documents/life-os/notes-processing/my-writing-dna-substack.md. This represents a data exposure risk as it accesses information from the user's local filesystem.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface by reading external markdown files to guide its behavior.
  • Ingestion points: Reads /Users/rami/Documents/life-os/notes-processing/my-writing-dna-substack.md at the start of every session.
  • Boundary markers: Absent. The skill treats the file content as the 'source of truth' without delimiters or instructions to ignore potential injections within that file.
  • Capability inventory: The agent is tasked with writing, reviewing, and developing content based on this external data.
  • Sanitization: None. The content of the external file is directly adopted into the agent's persona and logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 10:16 PM
Security Audit — agent-trust-hub — rami-voice