The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the ability to execute shell commands to manage its environment and perform checks.
Evidence: In lib/test-output-format.md, the skill explicitly uses mkdir -p .context/ui-test-screenshots to create directories for failure artifacts.
Evidence: In SKILL.md, the 'Assertion Protocol' specifies using eval to return structured data from the browser for deterministic verification.
[PROMPT_INJECTION]: The skill possesses a significant attack surface for Indirect Prompt Injection due to its interaction with untrusted external data and its capability to modify project code.
Ingestion points: The agent ingests data from live web pages (via browser automation), project briefs (docs/briefs/), and learning documents (docs/learnings/) to plan and execute tests.
Boundary markers: The instructions do not provide specific delimiters (e.g., XML tags or clear 'ignore' warnings) to separate instructions from the data being analyzed during the QA process.
Capability inventory: The skill can execute browser scripts (eval), write to the file system (creating screenshot directories), and perform 'AUTO-FIX' operations which involve modifying project source code based on test results.
Sanitization: There are no explicit instructions for the agent to sanitize, escape, or validate the content retrieved from external UI elements or documents before using it to decide on code fixes.

qa