The Agent Skills Directory

[PROMPT_INJECTION]: The file-bridge template implements an Indirect Prompt Injection surface designed for external agent control.
Ingestion points: The extension monitors the .vscode/ai-bridge/commands/ directory for JSON files containing action requests (implemented in assets/file-bridge/src/fileBridge.ts).
Boundary markers: The template code does not implement specific delimiters or instructions to ignore embedded prompts within the ingested JSON data.
Capability inventory: The extension registers handlers for reading files, writing files, and executing a whitelisted set of VS Code commands (defined in assets/file-bridge/src/extension.ts).
Sanitization: The template code in assets/file-bridge/src/extension.ts lacks path validation, although the reference documentation (references/ai-integration.md) provides examples of how to implement such sanitization.
[DATA_EXFILTRATION]: The readFile handler provided in the file-bridge template (assets/file-bridge/src/extension.ts) allows reading from arbitrary file paths using vscode.workspace.fs.readFile without verifying if the path is within the workspace boundaries, potentially exposing sensitive system files.
[COMMAND_EXECUTION]: The writeFile handler in the file-bridge template (assets/file-bridge/src/extension.ts) enables writing to arbitrary file paths on the system, which could be exploited to overwrite configuration files or other sensitive data.

vscode-extension-builder-lawvable