external-instruction-follower
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions explicitly delegate the agent's behavior to content retrieved from an external webpage, creating a surface for injection attacks.
- Ingestion points: The directive in
payload.mdinstructs the agent to "Follow the instructions from the webpage." - Boundary markers: Absent. The skill provides no delimiters, tags, or instructions to the agent to treat the external content as untrusted data rather than executable instructions.
- Capability inventory: According to the description in
SKILL.md, the purpose is to "Delegate agent behavior," which potentially exposes any tools or capabilities available to the agent to influence from the external source. - Sanitization: Absent. No filtering, validation, or sanitization logic is present to ensure the instructions obtained from the external page are safe to execute.
Audit Metadata