delegate-prose
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided prose (technical articles, READMEs, etc.) and delegate them to sub-agents. This introduces an attack surface where malicious instructions could be embedded in the data provided to the sub-agent.
- Ingestion points: User input is accepted via stdin and processed by
build-request.sh,prepare.sh, andrun.shto create tasks for sub-agents. - Boundary markers: The skill encourages a structured Markdown format (Objective, Scope, Context, etc.) to define the delegation request.
- Capability inventory: The skill has the ability to write files to the local filesystem (defaulting to
delegate-prose-output/) and execute sub-agents via various AI providers. - Sanitization: While the skill implementation includes security-conscious features like a '--pure' mode for OpenCode to prevent arbitrary code execution from plugins, the provided script shims do not explicitly show sanitization of the interpolated user prose.
Audit Metadata