write
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill displays a potential surface for indirect prompt injection due to its workflow of ingesting external data and user input for subsequent documentation generation.
- Ingestion points: External data enters the context via web search results (Agent 3) and user-provided instructions (Step 1).
- Boundary markers: The skill uses placeholder markers (e.g.,
<writing plan>,<key findings>) to structure prompts for the writing agent in Step 4, though it does not use explicit boundary delimitation or "ignore instructions" wrappers. - Capability inventory: The skill possesses the capability to write and update files (
src/content/,src/sidebar*.json) and execute subprocesses via sub-agent calls. - Sanitization: While no automated sanitization is described, the workflow includes a mandatory "Synthesize Writing Plan" step (Step 3) where the agent must present the plan for user confirmation before writing, acting as a critical human-in-the-loop safety checkpoint. Additionally, the automated review process (Step 5) provides a second layer of verification against structural and stylistic anomalies.
Audit Metadata