advise
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill automates task distribution by executing shell commands via the herdr CLI, including workspace creation and agent initialization. It utilizes local configuration files, such as advise.json and harnesses.json, to define execution parameters for various agent harnesses.
- [PROMPT_INJECTION]: A reporting mechanism allows worker agents to send completion summaries to an advisor agent via the herdr agent prompt command. This summary is interpolated into the advisor's context without sanitization or boundary markers, creating a surface for indirect prompt injection if a worker processes malicious data. Evidence includes: 1) Ingestion point in SKILL.md Step 6 completion callback; 2) Absence of boundary markers in the callback prompt; 3) Advisor capabilities include agent management and prompting; 4) Lack of input sanitization for worker-generated summaries.
- [EXTERNAL_DOWNLOADS]: The skill refers to updating the herdr tool using the brew package manager, which is a well-known and trusted service.
Audit Metadata