ask-peer
Fail
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to bypass security sandboxes and user approval prompts. In references/ask-codex.md, it recommends using -s danger-full-access and -c 'approval_policy="never"' for write operations. In references/ask-fable.md, it suggests using --permission-mode dontAsk and bypassPermissions.
- [COMMAND_EXECUTION]: The skill uses local shell commands to drive the codex and claude CLI tools, providing them with broad access to the workspace and the ability to execute code in separate processes.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and forwards it to external agents. • Ingestion points: User-supplied question or task argument defined in SKILL.md. • Boundary markers: None observed; input is written to temporary files and passed directly to peer agents. • Capability inventory: The skill utilizes CLI tools with full filesystem write access and bypassed security permissions. • Sanitization: No evidence of input filtering or escaping before data is interpolated into the prompt for the secondary agent.
Recommendations
- AI detected serious security threats
Audit Metadata