The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests and processes untrusted data from local extension projects.
Ingestion points: The skill reads configuration and metadata from vscode_extensions/<name>/package.json and other project files.
Boundary markers: No explicit delimiters or instructions are used to prevent the agent from obeying instructions embedded within the extension source code.
Capability inventory: The skill can execute shell commands via npm, npx, and the code CLI, allowing for subprocess creation.
Sanitization: There is no validation or filtering of scripts defined in package.json before they are executed via npm run compile.
[COMMAND_EXECUTION]: The skill performs several command-line operations to compile, package, and install extensions.
Executes npm run compile to build the TypeScript source.
Executes npx @vscode/vsce package to create extension bundles.
Executes code --install-extension to modify the local VSCode environment.
[EXTERNAL_DOWNLOADS]: The skill uses npx to fetch and run the @vscode/vsce package. This is an official utility provided by Microsoft for VSCode extension management and is considered a well-known, trusted source.

vscode-extension-dev