cs-loop
Pass
Audited by Gen Agent Trust Hub on Jun 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements a secure 'Draft-Approve-Execute' architecture that enforces role separation between planning, approval, and execution roles. Planning and approval agents operate in read-only sandboxes.
- [COMMAND_EXECUTION]: Orchestration is performed via a shell script (tools/codex-loop.sh) that manages AI tasks using the platform's execution utility. The script follows secure coding standards for shell variable handling.
- [PROMPT_INJECTION]: The skill manages potential indirect prompt injection through its multi-agent verification workflow. Ingestion points: tools/codex-loop.sh interpolates user objectives, human-decision files, and workspace artifacts. Boundary markers: Present (the skill mandates Context Boundary and Subtask Summary sections for worker contexts). Capability inventory: codex exec utility with workspace-write permissions. Sanitization: Absent.
- [DATA_EXFILTRATION]: No unauthorized network operations or exfiltration patterns were detected in the instructions or orchestration scripts.
Audit Metadata