knowledge-graph-hardened

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests and processes user-supplied text that is stored and later rendered back into the agent's context through entity summaries.\n
  • Ingestion points: The add and supersede commands in scripts/kg.py take user-provided fact strings and store them in items.json.\n
  • Boundary markers: Facts are interpolated directly into the summary.md file without delimiters, isolation blocks, or warnings to ignore embedded instructions.\n
  • Capability inventory: The skill utilizes scripts/kg.py to write facts and summaries to the local filesystem; the agent is instructed to perform these updates based on conversational input.\n
  • Sanitization: No sanitization, validation, or escaping of the fact content is performed in scripts/kg.py beyond basic whitespace trimming.\n- [COMMAND_EXECUTION]: The skill uses a bundled Python script (scripts/kg.py) to manage local file operations. The script is dependency-free and includes logic to validate entity names, effectively preventing path traversal through character restriction on input parameters.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 06:31 PM
Security Audit — agent-trust-hub — knowledge-graph-hardened