Interview

Pass

Audited by Gen Agent Trust Hub on May 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local scripts (InterviewScan.ts and TelosRenderer.ts) from the ~/.claude/PAI/TOOLS/ directory using the bun runtime. These executions are used to scan context completeness and regenerate summary files.
  • [DATA_EXFILTRATION]: Status notifications are sent to a local server (http://localhost:31337/notify) using curl. This communication is confined to the local machine and does not involve external data transmission.
  • [PROMPT_INJECTION]: The skill features an indirect prompt injection surface as it reads and acts upon user-provided context files (e.g., MISSION, GOALS).
  • Ingestion points: Files in ~/.claude/PAI/ such as foundational TELOS and IDEAL_STATE files.
  • Boundary markers: None explicitly defined in the file processing workflow; utilizes natural language summaries.
  • Capability inventory: Local command execution via bun and file system access using the Read and Edit tools.
  • Sanitization: Relies on the agent's internal logic to interpret and format text from context files correctly.
Audit Metadata
Risk Level
SAFE
Analyzed
May 2, 2026, 01:03 AM
Security Audit — agent-trust-hub — Interview