better-interface
Pass
Audited by Gen Agent Trust Hub on Aug 21, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface by ingesting and relying on content from local project documentation files (e.g.,
CONTRIBUTING.md,Storybookdocs,ADRs) to define its review standards and consolidation logic. An attacker who controls these project files could influence the agent's reasoning or the final consolidated verdict. \n - Ingestion points: Files like
CONTRIBUTING.md,CODING_STANDARDS.md,AGENTS.md,CLAUDE.md, Storybook documentation, and interface ADRs as specified in SKILL.md. \n - Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are used when processing these external sources. \n
- Capability inventory: Includes conditional source code modification (if requested by the user) and the execution of project-specific verification commands. \n
- Sanitization: No specific sanitization or filtering of external content is mentioned. \n- [COMMAND_EXECUTION]: The instructions direct the agent to identify and run "any preview or test command" found within the project environment. This capability to execute arbitrary project-defined commands during the verification phase represents a potential risk if the repository contains malicious scripts or configurations. \n- [DATA_EXFILTRATION]: The skill reads various project-level documentation and source code to identify styling systems, component libraries, and design tokens. While functional, this involves targeted reading of local filesystem data which contributes to the agent's overall data exposure surface.
Audit Metadata