karpathy-llm-wiki

Pass

Audited by Gen Agent Trust Hub on Apr 13, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external sources and integrates it into a persistent knowledge base. \n
  • Ingestion points: External content is fetched from URLs or user-pasted text during the Fetch (raw/) step in SKILL.md.\n
  • Boundary markers: The instructions do not define boundary markers or warnings for the agent to ignore embedded instructions in the source material.\n
  • Capability inventory: The skill allows the agent to read and write files within the project's raw/ and wiki/ directories and perform network requests to fetch data.\n
  • Sanitization: There is no explicit requirement for the agent to sanitize or escape the external content before synthesizing it into wiki articles.\n- [EXTERNAL_DOWNLOADS]: The skill fetches source material from arbitrary external URLs to populate the knowledge base, which is an intended core function but represents an interaction with untrusted external content.\n- [COMMAND_EXECUTION]: The skill performs automated file system operations, including directory creation and file modification (auto-fixing links and index entries) during the Ingest and Lint operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 13, 2026, 11:22 AM
Security Audit — agent-trust-hub — karpathy-llm-wiki