Get Started/Installation & Usage

Usage

Discover plugins

Marketplace registrations are in:

  • .claude-plugin/marketplace.json for Claude Code.
  • .agents/plugins/marketplace.json for Codex.
  • .cursor-plugin/marketplace.json for Cursor.

Each entry points to a source directory under plugins/.

Install and use a skill

Install one skill without cloning the repository or installing its plugin:

bash
gh skill install mehdihadeli/agent-toolkit durable-plans
npx skills add mehdihadeli/agent-toolkit --skill durable-plans

Choose one command and select your coding tool when prompted. GitHub CLI 2.90+ is required for gh skill; npx skills requires Node.js and npm. See skills-only installers for agent selection, scope, pinning, and the difference from clone-based installation.

After installation, give your coding tool a concrete task:

text
Use durable-plans to track this migration, record verification evidence, and leave a clear next action for the next session.

The host discovers and activates installed skills according to its own rules. Reading the catalog does not install a skill, and installing a skill does not start an MCP server, .NET agent, or plugin-level hook.

Run the search MCP

bash
dotnet run --project plugins/search/mcps/search/AgentSkillsMcp.Search.csproj

Default transport is stdio. Pass --http, --transport=http, or --transport http to host HTTP at http://127.0.0.1:6243. Configure Tavily enablement and TAVILY_API_KEY through the environment. Tavily Search and Extract are implemented by the shared SDK under plugins/search/shared/tavily.

Run the quality agent

bash
dotnet run --project plugins/dotnet-quality/agents/dotnet-quality-agent/AgentSkillsMcp.DotnetQuality.CopilotAgent.csproj

Set QUALITY_AGENT_WORKING_DIRECTORY to the workspace to inspect. The agent is read-only.

Validate the repository

bash
mise run validate
mise run waza-eval
dotnet test --solution agent-toolkit.slnx
vally lint .

Run one Waza suite during skill development:

bash
WAZA_SKILL=skill-guide mise run waza-eval

Use Waza for fast mock-based trigger routing, Vally evaluations for real agent behavior and prompt contracts, unit tests for deterministic logic, and integration tests for hosted boundaries. See authoring.md before adding a plugin or host surface.

View source on GitHub
Agent ToolkitOpen source. Plugin first.