create-plan-by
Fail
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to invoke external CLI tools using flags designed to bypass security protocols and user confirmation. Specifically, it specifies the use of
--dangerously-skip-permissionsfor theclaude(Claude Code) tool and--yolofor theagent(Cursor) tool. These flags suppress safety prompts and human-in-the-loop verification, which is a high-risk autonomy abuse pattern. - [COMMAND_EXECUTION]: The skill directs the AI to search the user's terminal configuration files for command aliases and execute them if a specific sub-agent is not found. This introduces a risk where the agent might execute unverified or malicious local commands without proper validation of the alias's actual function.
- [PROMPT_INJECTION]: The skill creates a surface for Indirect Prompt Injection (Category 8). It ingests untrusted data (plans generated by external sub-agents) and processes this content to synthesize a revised plan. The instructions lack boundary markers (delimiters) or sanitization logic to ensure that instructions embedded within a sub-agent's generated plan do not override the primary agent's logic during the synthesis phase.
Recommendations
- AI detected serious security threats
Audit Metadata