Skip to content

plugins.md

Latest commit

 

History

History
41 lines (29 loc) · 1.69 KB

plugins.md

File metadata and controls

41 lines (29 loc) · 1.69 KB
Raw

Agent Plugins

agentkit-plugins parses and validates Agent Plugins 1.0 packages without turning a plugin into a parallel capability system.

plugin/
├── plugin.json
├── skills/*/SKILL.md
└── mcp.json

Information-first loading

use agentkit_plugins::AgentPlugin;
use agentkit_tool_skills::SkillRegistry;

let plugin = AgentPlugin::load("./plugins/acme")?;

let skills = SkillRegistry::from_skill_dirs(plugin.skill_directories())
    .discover_skills()
    .await;

for diagnostic in plugin.diagnostics() {
    eprintln!("{:?}: {}", diagnostic.kind, diagnostic.message);
}

for server in plugin.mcp_servers() {
    // The host chooses how and whether to map this portable declaration into
    // an agentkit_mcp::McpServerConfig.
    println!("{}: {:?}", server.name, server.transport);
}

The loader returns validated manifest metadata, exact skill directories, raw portable MCP declarations, opaque client-extension data, and structured diagnostics. Invalid skills and MCP entries are isolated according to the standard.

Deliberate boundaries

The crate does not install or update plugins, choose permissions, create a SkillRegistry, launch MCP servers, expand MCP placeholders, or choose a PLUGIN_DATA directory. Those are host/runtime decisions. PLUGIN_DATA is only needed if a host later launches a plugin's stdio MCP server; it is a writable persistent directory selected for that installed plugin instance.

Use the umbrella crate's plugins feature to access the parser as agentkit::plugins. The feature remains independent of tool-skills and mcp, allowing consumers to inspect packages without pulling in either runtime.