delegate-to-agent
Pass
Audited by Gen Agent Trust Hub on Mar 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill establishes a pattern for delegating tasks from the UI or scripts to the AI agent using the
sendToAgentChatandagentChat.submitfunctions. This creates a surface for indirect prompt injection where untrusted data can be passed into the agent's processing flow. - Ingestion points: External data is ingested via the
messageandcontextparameters of thesendToAgentChatfunction as shown in the examples inSKILL.md. - Boundary markers: The provided code snippets do not include or recommend the use of boundary markers (such as XML tags or delimiters) to wrap untrusted content.
- Capability inventory: The skill documentation indicates the agent has high-privilege capabilities including reading/writing any file, running scripts via
pnpm, and performing self-modifying code operations. - Sanitization: There is an absence of instructions for sanitizing, escaping, or validating the contents of the
contextormessagefields before they are submitted to the agent.
Audit Metadata