skill-eval

Pass

Audited by Gen Agent Trust Hub on Sep 22, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes instructions from external skill files and task execution artifacts, writing feedback back to the target skill's SKILL.md. This ingestion-and-write loop provides an attack surface where malicious content in an evaluated skill could influence the agent.
  • Ingestion points: Reads target SKILL.md files and execution logs stored in .scratch/neonforge-v1/skill-eval/.
  • Boundary markers: Absent. No markers are defined to isolate untrusted skill content during processing.
  • Capability inventory: The process requires file writing to SKILL.md and executing shell commands for task verification (per AC requirements).
  • Sanitization: Absent. There is no mechanism described for validating or sanitizing feedback before it is written into skill metadata or instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 22, 2026, 02:56 PM
Security Audit — agent-trust-hub — skill-eval