cw-panel
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external content and processes it through a chain of subagents, posing a risk of indirect prompt injection.
- Ingestion points: Untrusted data enters the agent context via the draft text and associated metadata (brief, outline, project instructions, feedback) as described in the "Load Context" and "Run Synthesizer" steps.
- Boundary markers: While the skill uses Markdown headers (e.g.,
## Draft,## Reviewer Outputs) as delimiters when passing content to the synthesizer agent, it lacks explicit "ignore embedded instructions" warnings or strict boundary protocols to prevent malicious instructions within a draft from influencing the reviewer or synthesizer subagents. - Capability inventory: The skill utilizes file-reading capabilities (
../../references/context-contract.md) and orchestrates the execution of multiple parallel subagents (e.g.,compound-writing:review:*). - Sanitization: There is no evidence of input validation, filtering, or escaping for the draft content before it is interpolated into the prompts for the reviewer or synthesizer agents.
Audit Metadata