principle-encode-lessons-in-structure
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and follow instructions from a relative file path (
../saltmode/references/host-adapter.md), creating a surface where instructions from that file enter the agent's context. - Ingestion points:
SKILL.mddirects the agent to read the content of../saltmode/references/host-adapter.mdonce per session. - Boundary markers: The instructions do not define specific delimiters or "ignore previous instructions" warnings for the content loaded from the host adapter.
- Capability inventory: The skill metadata includes
disable-model-invocation: true, which restricts the execution environment, though the agent maintains standard file reading and tool use capabilities. - Sanitization: No validation or sanitization of the external file content is specified.
Audit Metadata