The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides patterns for downloading language server binaries from GitHub releases using the zed::download_file function. It also covers the installation of Node.js packages via zed::npm_install_package. These operations are standard for extension development and require explicit capability grants in the user's settings.
[COMMAND_EXECUTION]: The documentation describes the use of zed::Command to launch external processes, such as Language Server Protocol (LSP) binaries and Model Context Protocol (MCP) servers. The implementation snippets demonstrate how to locate binaries in the system PATH using worktree.which and how to define execution environments.
[INDIRECT_PROMPT_INJECTION]: The skill identifies attack surfaces related to AI Slash Commands and Model Context Protocol (MCP) servers. These features ingest external data (LSP outputs) or user input (slash command arguments) that could potentially influence agent behavior. The guide includes registration and implementation patterns for these capabilities.
Ingestion points: Slash command arguments (run_slash_command), project workspace data, and MCP server responses.
Boundary markers: Not explicitly defined in the basic implementation snippets.
Capability inventory: Uses process:exec, download_file, and npm:install capabilities.
Sanitization: Not demonstrated in the simplified code examples; developers are expected to handle validation within their Rust implementation.
[DYNAMIC_EXECUTION]: The skill covers the creation of Rust crates compiled to WebAssembly (Wasm). These extensions are executed within a sandboxed Wasmtime environment provided by the Zed editor, which limits the potential impact on the host system.

zed-editor