4d-v21

Pass

Audited by Gen Agent Trust Hub on Jun 5, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface Risk.
  • Ingestion points: The main instructions in SKILL.md direct the agent to check for a local/ directory and prioritize reading any markdown files within it (local/*.md) before providing developer guidance.
  • Boundary markers: There are no specified delimiters or "ignore embedded instructions" warnings for the content read from this directory, which might be populated with untrusted data in collaborative environments.
  • Capability inventory: The skill leverages capabilities to read the file system, use search tools like grep, and provides documentation for executing host-level shell commands via the SystemWorker class.
  • Sanitization: No sanitization or validation of the local content is implemented.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 5, 2026, 03:36 PM