guardian
Pass
Audited by Gen Agent Trust Hub on Mar 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The orchestrator script
scripts/guardian.pyutilizessubprocess.runto execute local Python utility scripts includingmonitor_session.py,context_filter.py, andvalidator.py. This represents an internal command execution flow used to coordinate the skill's different logic components.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes user-controlled files for code review and task planning.\n - Ingestion points:
scripts/context_filter.pyreads the contents of project files specified by the user for analysis by subagents.\n - Boundary markers: Prompt templates in the
Templates/directory include explicit 'CRITICAL CONSTRAINTS' and 'READ-ONLY' instructions intended to prevent subagents from executing actions or using tools.\n - Capability inventory: The skill utilizes the
Read,Glob,Grep, andTasktools. It uses theTasktool to spawn sub-models for analysis tasks.\n - Sanitization: The implementation relies on prompt-based constraints and the intended read-only nature of the subagents rather than performing input sanitization of the files being reviewed.
Audit Metadata