improve-ui-components
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes content from external websites and repository files, presenting a surface for indirect prompt injection attacks. Ingestion points: Page URLs, accessibility trees, DOM content, and repository files. Boundary markers: Hard Rule 8 explicitly warns the agent to treat fetched content as data, not instructions, reducing accidental obedience. Capability inventory: File writes to the repository, local server management, and GitHub CLI execution. Sanitization: The skill instructs the agent to report any instructions found in processed data as findings rather than executing them.
- [COMMAND_EXECUTION]: The skill executes shell commands to manage local development environments and interact with the GitHub CLI. This includes repository-defined scripts like pnpm dev, pnpm test, or npm lint, and GitHub commands such as gh issue create. It also uses lsof to identify processes on local network ports.
- [EXTERNAL_DOWNLOADS]: The skill fetches a knowledge base from the author's infrastructure. It downloads components.json from oekazuma.github.io at the start of every run and mentions using npx skills add for updates.
- [DYNAMIC_EXECUTION]: The skill uses browser automation to execute a JavaScript function for UI analysis. It uses page.evaluate() to run a provided script that calculates target sizes on the audited page to check compliance with accessibility standards.
Audit Metadata