analyze-run

Pass

Audited by Gen Agent Trust Hub on Jul 11, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses npx convex run --prod to execute queries against a production environment. This allows for arbitrary execution of vendor-specific logic in a sensitive context.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it retrieves evaluation data (like failure reasons and error messages) from an external database and interpolates it into sub-agent prompts without adequate protection.\n
  • Ingestion points: Evaluation failure details fetched from the debugQueries:getFailedEvalsForRun and debug:getEvalDebugInfo commands in SKILL.md.\n
  • Boundary markers: No clear delimiters or system instructions are provided in the sub-agent prompt template to prevent the model from obeying instructions embedded in the evaluation data.\n
  • Capability inventory: The skill and its sub-agents have the ability to execute shell commands (npx convex) and perform filesystem write operations in the reports/ directory.\n
  • Sanitization: There is no evidence of sanitization, escaping, or validation of the failure data before it is passed to the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 11, 2026, 03:02 PM
Security Audit — agent-trust-hub — analyze-run