principle-separate-before-serializing-shared-state

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a process for the agent to ingest information from untrusted local project files, creating a potential surface for indirect prompt injection.\n
  • Ingestion points: Project files including AGENTS.md, CLAUDE.md, TODO.md, and the codebase (referenced in SKILL.md).\n
  • Boundary markers: Absent. The instructions do not specify delimiters or tell the agent to ignore instructions embedded in the data it reads.\n
  • Capability inventory: The agent is instructed to perform targeted searches using rg and read codebase files to generate findings.\n
  • Sanitization: Absent. No filtering or escaping is specified for the ingested content.\n- [COMMAND_EXECUTION]: The skill recommends using rg (ripgrep) for searching the codebase.\n
  • This is standard developer tool usage intended for context discovery and does not involve spawning shell commands based on untrusted network input.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 11:52 AM
Security Audit — agent-trust-hub — principle-separate-before-serializing-shared-state