The Agent Skills Directory

[SAFE]: The skill instructions and reference documents focus on standard QA practices, such as choosing appropriate test levels (unit vs. integration), reusing existing project frameworks, and performing visual regression checks.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection because it is designed to ingest and process project documentation (such as AGENTS.md or API docs) and user-provided screenshots.
Ingestion points: Project documentation, API schemas, and user-provided images/mockups are read to determine expected behavior (referenced in references/test-strategy.md and references/ui-visual-verification.md).
Boundary markers: The instructions do not define specific delimiters for untrusted data, but they prioritize existing project tests and explicit user approval for agent actions.
Capability inventory: The skill can execute project test commands (shell) and interact with a built-in browser or Playwright.
Sanitization: No specific sanitization logic for external data is mentioned, which is typical for development-oriented agents.
[COMMAND_EXECUTION]: The skill identifies and runs existing project test commands (e.g., via package.json, pytest, or go test). This behavior is restricted to the smallest useful target to verify specific changes and is core to the skill's intended purpose.

testing-verification