launch-subagent
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a process for delegating tasks to sub-agents by constructing detailed prompts from current context. This creates a surface where instructions embedded in processed data could potentially influence sub-agent behavior.
- Ingestion points: Instructions in
SKILL.mdrequire writing a "full brief into the prompt" for sub-agents, incorporating scope and needed context. - Boundary markers: The skill does not explicitly instruct the use of delimiters or 'ignore embedded instructions' markers when passing context to sub-agents.
- Capability inventory: The instructions suggest providing sub-agents with tools such as
deepapifor web search and scraping, giving them network access capabilities. - Sanitization: There are no instructions provided for sanitizing or filtering context data before interpolation into the sub-agent brief.
Audit Metadata