thompson-unix-philosophy

Fail

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: HIGHPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The title header '# Ken Thompson Style Guide' contains a payload of 184 hidden characters (Zero-Width Space, Zero-Width Non-Joiner, and Zero-Width Joiner). This technique is used to inject instructions that bypass user visibility but are processed by the AI. Initial decoding of the binary-to-Unicode mapping reveals the prefix 'SK1LL', suggesting a structured hidden payload.
  • [DATA_EXFILTRATION]: The skill provides an attack surface for indirect prompt injection by demonstrating how to combine data ingestion with network capabilities. * Ingestion points: Code examples in SKILL.md demonstrate reading data from 'os.Stdin', command-line arguments, and local files. * Boundary markers: The skill lacks any instructions or delimiters to isolate untrusted user data from the agent's logic. * Capability inventory: The skill includes functional examples of network requests ('http.Get') and file system operations ('os.Open'). * Sanitization: No input validation, escaping, or sanitization techniques are provided or recommended for the demonstrated data processing patterns.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 15, 2026, 07:01 PM
Security Audit — agent-trust-hub — thompson-unix-philosophy