Sikaru plugins and skills

The official Sikaru plugin for Claude Code and Codex, plus the same skills for any agent that loads SKILL.md files: Cursor, Devin, Capy, Pi, OpenCode, Gemini CLI, GitHub Copilot, Amp, Cline, Windsurf, Goose and more. Your coding agent learns to sign you up for Sikaru, choose a project, and then work with it: call catalog models, ask System One decisions, write and publish managed agents, evaluate them in RL environments, and upload traces.

The plugin also connects the public Sikaru docs over MCP (sikaru-docs, https://docs.sikaru.ai/_mcp/server). That server is read-only and needs no credentials.

Install

Claude Code

claude plugin marketplace add aadi-labs/sikaru-plugins
claude plugin install sikaru@sikaru

Codex

codex plugin marketplace add aadi-labs/sikaru-plugins
codex plugin add sikaru@sikaru

Any agent with skills

npx skills add aadi-labs/sikaru-plugins

The installer detects the agents you have and asks where to install. To pick one, pass its name with -a (add -g to install for your user instead of the project):

Any other agent that reads SKILL.md files works too: copy a folder from skills/ into its skills directory. See the full list of supported agents.

Capy

Capy installs this repository as a plugin, skills and docs server together: open Settings → MCP & plugins, choose From GitHub repository, and enter https://github.com/aadi-labs/sikaru-plugins. Click Sync on the plugin page to pick up new versions. To keep the skills in one repository instead, install them where Capy reads repository skills:

npx skills add aadi-labs/sikaru-plugins -a universal

That writes to .agents/skills/, which Capy, Cursor, Codex, OpenCode, Pi and other agents all read.

Docs over MCP in other agents

Claude Code, Codex and Capy get the docs server from the plugin. Elsewhere, add https://docs.sikaru.ai/_mcp/server as a remote MCP server (Streamable HTTP, no credentials):

• OpenCode, in opencode.json: {"mcp": {"sikaru-docs": {"type": "remote", "url": "https://docs.sikaru.ai/_mcp/server"}}}
• Gemini CLI, in ~/.gemini/settings.json: {"mcpServers": {"sikaru-docs": {"httpUrl": "https://docs.sikaru.ai/_mcp/server"}}}
• Cursor, in .cursor/mcp.json: {"mcpServers": {"sikaru-docs": {"url": "https://docs.sikaru.ai/_mcp/server"}}}
• GitHub Copilot in VS Code, in .vscode/mcp.json: {"servers": {"sikaru-docs": {"type": "http", "url": "https://docs.sikaru.ai/_mcp/server"}}}

Agents without MCP, such as Pi, read the same docs as Markdown: start at https://docs.sikaru.ai/llms.txt.

Then ask your agent

• "Set me up with Sikaru."
• "List the Sikaru models and chat with the cheapest one that supports reasoning."
• "Write a Sikaru agent for this repository, validate it and run one task."
• "Run my eval dataset against the live agent and show me the failures."

Skills

Requirements

The skills drive the sikaru CLI. They need a build with the decision and credits groups (sikaru decision --help should work):

curl -fsSL https://app.sikaru.ai/install.sh | sh   # macOS and Linux
npm install -g @aadi-labs/sikaru-cli               # any platform with Node.js 20+

Code uses the published SDKs, which share the CLI's version:

Signing up and signing in both start with sikaru login, which also saves your default project. Agents without the CLI follow https://sikaru.ai/auth.md. Code, servers and CI use a project API key in SIKARU_API_KEY, created in the dashboard at https://app.sikaru.ai.

Work is prepaid with credits, and an agent can fund it: sikaru credits add --amount 20 creates an offer that an agent wallet pays, even for a brand-new organization. People can instead save a card in Billing at https://app.sikaru.ai, which also adds a one-time $5 starting credit.

Repository layout

Development

python3 scripts/validate.py --cli "$(which sikaru)"
claude plugin validate .

The --cli check runs every sikaru … command the skills mention with -h. A renamed or removed CLI command fails the check.

License

MIT