skills/emilkowalski/skills/break-ui/Gen Agent Trust Hub

break-ui

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: No malicious prompt injection patterns were found. The static detection for "ignore previous instructions" refers to a defensive rule (Hard Rule 6) that instructs the agent to disregard such instructions if they are found within the source code being analyzed, which is a security best practice for agents processing untrusted files.
  • [DATA_EXFILTRATION]: The skill does not perform any network operations or access sensitive credentials. It operates entirely within the local repository context to analyze UI components and generate test data.
  • [REMOTE_CODE_EXECUTION]: No remote code execution patterns, external script downloads, or untrusted package installations were detected. All data values used for testing are sourced from the local CATALOG.md file.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface as it reads component source code and schemas. However, it implements a robust mitigation by explicitly instructing the agent to treat repository content as data only and to flag/ignore any embedded instructions aimed at steering the agent's behavior.
  • [DYNAMIC_EXECUTION]: While the skill modifies local code to insert a "Worst Case" toggle, this is its primary stated purpose as a development utility. It does not use eval(), exec(), or other unsafe dynamic execution methods on untrusted data. The inclusion of XSS patterns like <script>alert(1)</script> in CATALOG.md is for the legitimate purpose of testing the UI's escaping mechanisms.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 08:33 PM