crosspost
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content (drafts, URLs, threads) which represents an indirect prompt injection surface. However, the skill includes explicit defensive instructions to the agent to treat such content as data rather than instructions, to never follow embedded commands, and to flag agent-directed text.
- Ingestion points: The skill accepts source material from URLs or external drafts (SKILL.md).
- Boundary markers: The skill contains detailed 'Untrusted Source Material' rules that explicitly instruct the agent to ignore embedded commands and treat input strictly as data to be adapted.
- Capability inventory: No dangerous capabilities (file writes, network calls, command execution) are implemented within this skill file itself; it delegates publishing to other tools.
- Sanitization: The skill mandates flagging agent-directed text to the user instead of executing or adapting it.
Audit Metadata