methodology-work
Pass
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by design. It transforms user-provided corrections and feedback into foundational 'principles' (grundsatz) and 'distinctions' (sinn) that are intended to shape the behavior of future agent sessions.
- Ingestion points: The skill processes user-supplied rules, corrections, and agent feedback files to update the methodology graph (e.g., in the 'Recording a principle' and 'Processing agent feedback' sections).
- Boundary markers: The instructions do not specify the use of data delimiters or specific prompts to ignore malicious instructions embedded within user feedback.
- Capability inventory: The skill utilizes specialized tools (
nks_orient,nks_search,nks_look,nks_semantic_search) to read and modify the knowledge graph. While these tools are vendor-specific and provide high capability, they are necessary for the skill's primary function. - Sanitization: No explicit sanitization or validation steps are defined for the user-provided text before it is committed to the graph as a guiding principle. This architectural risk is inherent to the use case of a methodology-evolving agent.
Audit Metadata