sentinel-loop

Pass

Audited by Gen Agent Trust Hub on Mar 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local bash script (validate-launch.sh) to verify the integrity, structure, and safety of generated launch commands before they are presented to the user for execution.\n- [PROMPT_INJECTION]: The skill implements an orchestration pipeline that processes untrusted project documentation (indirect prompt injection surface). This is managed through the following mandatory security controls:\n
  • Ingestion points: Reads project state and requirements from PRD.md, progress.yaml, and the rules section of CLAUDE.md.\n
  • Boundary markers: Generates unique per-run nonces with a SENTINEL_ prefix and uses <promise> tags to ensure completion signaling is not accidentally triggered by model dialogue.\n
  • Capability inventory: The skill has the capability to write to local temporary files (.sentinel/tmp/) and execute local validation scripts via the shell.\n
  • Sanitization: Explicitly performs shell-safe escaping for special characters (backticks, dollar signs, quotes) and validates the final payload size to prevent execution failures or injection via shell argument limits.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 30, 2026, 12:58 AM