The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to build, validate, and install app extensions from user-provided source code and manifests (canvas-extension.json). This creates an attack surface where malicious instructions or code could be embedded in the data processed by the agent.
Ingestion points: The scripts/validate-extension.mjs script reads and parses user-supplied canvas-extension.json files and JavaScript entrypoints. The agent is also instructed to inspect repository content, including build configurations and source files.
Boundary markers: The skill provides a static validator script that enforces a specific schema and blocks forbidden patterns. The instructions also recommend manual review of the final production bundles before enablement.
Capability inventory: The skill executes the provided validator using node. It also involves running standard build tools (e.g., Vite) and test frameworks (e.g., Vitest) within the user's project environment.
Sanitization: The validator script uses regular expressions to detect and reject dangerous constructs such as CommonJS require, dynamic import(), re-exported dependencies, and assets that escape the package root.
[COMMAND_EXECUTION]: The skill requires the agent to execute local scripts and development tools. Specifically, it uses node to run the validation script (scripts/validate-extension.mjs) and its associated tests. It also guides the agent through standard development workflows, including npm install and build commands, which are necessary for the skill's primary purpose of application development.

canvas-extension-api