global-config

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFE
Full Analysis
  • Prompt Injection (SAFE): The instructions focus on interaction style, language policy, and routing logic. There are no attempts to bypass safety filters or ignore system instructions.
  • Data Exposure & Exfiltration (SAFE): The skill identifies local project files (e.g., package.json, tsconfig.json) to detect technology stacks. This is a standard functional behavior with no evidence of sensitive data being transmitted externally.
  • Indirect Prompt Injection (LOW): The skill reads data from untrusted local files like .agents/CONTEXT.md. 1. Ingestion points: package.json, tsconfig.json, .agents/CONTEXT.md, GEMINI.md, next.config., vite.config., nest-cli.json. 2. Boundary markers: None. 3. Capability inventory: File read (for context detection), File write (to .agents/skills/local/). 4. Sanitization: None mentioned. The risk is low as the data is used for stack inference rather than execution.
  • Persistence Mechanisms (SAFE): The 'Local Skill Capture' feature writes project-specific guidance to a local directory. This is an intended feature for agent workflow persistence and does not target system startup or user shell profiles.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:30 PM