prose-crafting
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user-provided text for editing, rewriting, and auditing purposes. The instructions include a specific 'Security Model' section that provides a mitigation strategy for potential injection attacks.\n
- Ingestion points: User-provided prose text in English or Ukrainian (SKILL.md).\n
- Boundary markers: Explicit instructions to treat instruction-shaped text within the prose as data, preventing it from overriding the agent's mode, scope, or repository-level instructions.\n
- Capability inventory: The skill is restricted to editing specific target files as requested by the user and lacks network or shell capabilities.\n
- Sanitization: The agent is instructed to report the presence of instruction-shaped text to the user rather than executing it.
Audit Metadata