skos-taxonomy
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions involve processing and classifying external data such as agent interactions, documents, and memory. This introduces a potential surface for indirect prompt injection where malicious instructions hidden in the data could influence the classification logic or downstream agent actions.
- Ingestion points: Interaction data, documents, and event logs processed for taxonomy classification as described in the 'Agent Classification Contract' section of SKILL.md.
- Boundary markers: No explicit prompt boundary markers or 'ignore instructions' directives are specified for the ingested content.
- Capability inventory: The skill documents the use of shell commands (curl) and infrastructure tools (kubectl).
- Sanitization: While label normalization is mentioned, there is no comprehensive sanitization logic for the content of the data being classified.
- [COMMAND_EXECUTION]: The documentation includes functional shell command examples for interacting with local services and managing infrastructure.
- Evidence: Examples in SKILL.md include curl requests to a local Typesense instance and a kubectl port-forward command for service access.
Audit Metadata