The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from arbitrary web pages and browser extensions, creating a surface for indirect instructions to be ingested from external sites.\n- Ingestion points: Target URLs, screen content, API contracts, and browser extension manifests (SKILL.md).\n- Boundary markers: The instructions mandate verifying environment overrides and candidate revisions before testing, but lack explicit delimiters for external content.\n- Capability inventory: The skill uses browser automation tools, file system writes for notes, and can execute repository scripts (SKILL.md).\n- Sanitization: Relies on human review of screenshots and evidence rather than programmatic content sanitization.\n- [COMMAND_EXECUTION]: The instructions direct the agent to run project-specific commands like clean install, build, and lint when source access is available.

ui-regression