lism-css-refactor
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill utilizes a multi-step task delegation pattern defined in SKILL.md (Pass 10) and references/output-format.md. This involves generating a prompt for a 'sub-agent' to verify the refactoring results, creating a surface where instructions from one agent session influence a subsequent invocation.\n
- Ingestion points: The 'references/output-format.md' file contains the evaluation sub-agent instruction template (評価サブエージェントへの指示テンプレ).\n
- Boundary markers: The template includes behavioral constraints such as 'read-only' and 'do not change code' to limit the sub-agent's actions.\n
- Capability inventory: The skill environment allows for reading and writing files (via diff application) and executing npm scripts such as 'nr lint', 'nr typecheck', and 'nr test'.\n
- Sanitization: No specific sanitization or filtering logic is defined for the content passed between agent instances.
Audit Metadata