The Agent Skills Directory

Local Command Execution: The skill employs standard development commands including npm install, npm run compile, and code-insiders. These are used to manage dependencies, build extension source code, and launch the editor with remote debugging enabled. This behavior is typical for setting up and testing extension environments.\n- Browser-Based Dynamic Execution: The automation workflow uses agent-browser eval to execute JavaScript directly within the VS Code UI context. This is a necessary technique for interacting with the Monaco Editor and other webview-based components that do not respond to standard high-level interaction events. It allows the agent to manage focus and verify visible text content.\n- Indirect Data Ingestion Surface: The skill interacts with content generated or displayed within the VS Code Copilot Chat interface, which represents a surface for untrusted data.\n
Ingestion points: Data enters the context from the VS Code UI via agent-browser snapshot and eval commands (SKILL.md).\n
Boundary markers: Absent. The skill does not currently specify delimiters or instructions to ignore potential commands within the retrieved UI content.\n
Capability inventory: The skill uses Bash to execute agent-browser and code-insiders, and agent-browser eval for dynamic code execution (SKILL.md).\n
Sanitization: Absent. The skill does not mention specific filtering or escaping of content retrieved from the webview targets.

launch