co-validate

Pass

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability by reading external plan files and directly interpolating their contents into a prompt for the 'codex' validation tool without sanitization.
  • Ingestion points: The contents of the plan file specified in the $ARGUMENTS path are read and processed in Step 1.
  • Boundary markers: While the original request is wrapped in tags, the plan content is appended directly after a label without clear delimiters or instructions to the model to ignore embedded commands.
  • Capability inventory: The skill utilizes mcp__validate-plans-and-brainstorm-ideas__codex and mcp__validate-plans-and-brainstorm-ideas__codex-reply to communicate with an external service and instructs the agent to modify local files based on that service's feedback.
  • Sanitization: No input validation, escaping, or instruction-filtering is performed on the plan file contents before they are passed to the subagent or the external tool.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 21, 2026, 01:01 PM
Security Audit — agent-trust-hub — co-validate