building-chat-widgets

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • COMMAND_EXECUTION (MEDIUM): The documentation instructs users to execute python3 scripts/verify.py to validate the skill. Executing local scripts provided within a skill package poses a risk of arbitrary code execution if the script's contents are not explicitly audited.
  • INDIRECT PROMPT_INJECTION (MEDIUM): The skill implements functionality to fetch and render external data in UI widgets, which can influence agent behavior or trigger actions.
  • Ingestion points: Untrusted data enters via the onTagSearch function (SKILL.md) which fetches results from /api/search.
  • Boundary markers: None. Data from the API (e.g., item.name) is directly interpolated into widget properties without delimiters or 'ignore' instructions.
  • Capability inventory: The skill possesses the ability to navigate() the browser, execute chatkit.sendUserMessage() to influence the conversation, and perform server-side data mutations via the action() handler.
  • Sanitization: No sanitization or schema validation is performed on the data returned from the search API before it is rendered into interactive components.
  • UNVERIFIABLE DEPENDENCIES (LOW): The skill references external files in the references/ directory and a scripts/ directory that were not provided for analysis, preventing a complete security audit of the execution chain.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 10:55 AM