The Agent Skills Directory

[SAFE]: The skill instructions are focused exclusively on software testing and quality assurance methodologies. No malicious patterns such as command injection, credential theft, or unauthorized data exfiltration were detected.
[COMMAND_EXECUTION]: The skill uses the agent-browser tool to perform UI automation tasks. This includes navigating to URLs, taking screenshots, and interacting with page elements. These actions are consistent with the skill's stated purpose and use authorized browser automation tools.
[PROMPT_INJECTION]: The skill processes untrusted data by interacting with external web content via agent-browser. This interaction represents an indirect prompt injection surface if the agent attempts to follow instructions found on the web pages being tested. However, the instructions emphasize 'staying in persona' and following a specific journey plan, which mitigates accidental obedience to third-party content.
Ingestion points: External web pages loaded via agent-browser open <url> (SKILL.md, references/web-ui-qa.md).
Boundary markers: None detected. No explicit warnings to ignore embedded instructions in the test surface are provided.
Capability inventory: UI interaction (click, fill) and file system writes for artifacts (screenshots, issues) via agent-browser and local file operations.
Sanitization: None detected.

qa-execution