The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automates the installation of its core tool and browser dependencies from external package registries.
Evidence: The Auto-Setup section in SKILL.md includes pip install aat-devqa and playwright install chromium.
[COMMAND_EXECUTION]: The skill performs testing tasks by executing the aat CLI tool and managing local project configurations.
Evidence: SKILL.md and references/cli-reference.md document commands such as aat scan, aat run, aat init, and aat dashboard.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external web applications, which serves as a potential vector for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context via URL scanning, the generated scan_result.json file, and OCR or Vision AI results processed during test execution.
Boundary markers: The skill implements a mandatory 4-step workflow in SKILL.md that requires explicit user approval after scanning the application and before running any generated test scenarios.
Capability inventory: The agent has the capability to execute the aat CLI, write scenario files to the local filesystem, and take screenshots of web applications.
Sanitization: The skill relies on human-in-the-loop validation, requiring the agent to present all scan results and scenario details to the user for review before proceeding.

awt