guru-paul-graham

Pass

Audited by Gen Agent Trust Hub on Sep 22, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill acts as an educational and advisory interface. It leverages 14 local reference documents containing public essays to ground the agent's persona and logic in a specific philosophy.
  • [DATA_EXPOSURE_EXFILTRATION]: The skill uses local file access to read its reference library and write user interaction results to a persistent knowledge file at knowledge/yc-startup-school/guru-pg-workbook.md. These actions are within the expected behavior of a productivity skill and do not involve unauthorized network transmission or access to sensitive system secrets like SSH keys or cloud credentials.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests user input regarding startup ideas to generate advice. 1. Ingestion points: User input provided during the interactive diagnostic questions in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: Reading local markdown references and writing results to knowledge/yc-startup-school/guru-pg-workbook.md. 4. Sanitization: Absent. The vulnerability surface is restricted to the generated text output and the saved workbook, presenting no significant risk to the host environment or agent safety guidelines.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 22, 2026, 02:37 AM
Security Audit — agent-trust-hub — guru-paul-graham