The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were identified. The skill's content is instructional and intended for game development and testing workflows.
[COMMAND_EXECUTION]: The skill utilizes legitimate browser automation tools through the agent-browser interface. It demonstrates how to use agent-browser eval and agent-browser mouse commands to interact with a game's internal state and simulate user input. These commands are localized to the testing environment.
[PROMPT_INJECTION]: While the skill involves reading data from a web page (Phaser game state), which inherently creates an indirect prompt injection surface, no malicious instructions or bypass attempts were found. The risk is consistent with standard browser-based agent operations.
Ingestion points: The skill reads from the window.__TEST__ object and its properties using agent-browser eval as seen in SKILL.md and references/common-commands.md.
Boundary markers: None are explicitly defined in the provided code patterns to separate game data from agent instructions.
Capability inventory: The skill utilizes agent-browser eval, agent-browser mouse, and grep for environmental inspection and interaction.
Sanitization: Data retrieved from the browser is not sanitized before use, which is typical for developer-focused testing scripts.

phaser-test-seam-patterns