recipe-prepare-implementation
Pass
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill reads work plans from external sources such as
$ARGUMENTSor files indocs/plans/to determine its execution flow. These inputs are used to generate task files and instruct subagents. There are no explicit boundary markers or sanitization steps to prevent malicious instructions within a work plan from influencing the agent's behavior (Indirect Prompt Injection). - [COMMAND_EXECUTION]: The skill orchestrates the creation of task files and the spawning of subagents to execute them. While these actions are targeted at project management within the repository, the logic relies on data parsed from untrusted work plans, which could be manipulated to trigger unintended file operations or subagent tasks.
Audit Metadata