superskill

Multi-agent skill, command, subagent, magent, hook, and MCP config distribution and authoring CLI. Install a Claude Code plugin to any coding agent, and create / validate / evaluate / refine / evolve agent-facing content from the command line.

Install

npm i -g @gobing-ai/superskill

# or go with bun
bun add -g @gobing-ai/superskill

No clone, no checkout, no extra flags. The published package bundles the cc plugin and the team-stark-children main-agent config, and the CLI locates them at runtime, so install works from any directory:

superskill install cc --magent team-stark-children

That distributes cc (skills, commands, subagents, hooks, rules) to every configured target and emits the team-stark-children magent. Want only the plugin, or only specific agents? superskill install cc (no --magent) and superskill install cc --targets codex,pi work too. See cmd_install.md for the full surface, including installing from a GitHub marketplace (--marketplace gobing-ai/superskill).

From source (contributors):

proto use          # install pinned tool versions (Bun, etc.)
bun install
bun run build
cd apps/cli && bun link

Requires Bun ≥ 1.3.14. See Installation guide for details and troubleshooting.

Quick start

# Install the bundled cc plugin to every supported target
superskill install cc --targets all

# Install cc plus the team-stark-children main-agent config
superskill install cc --magent team-stark-children

# Author a skill: scaffold → validate → evaluate → refine
superskill skill scaffold my-skill --description "Deploy a Cloudflare Worker"
superskill skill validate my-skill
superskill skill evaluate my-skill --save
superskill skill refine my-skill --auto

# Package a skill for distribution, or merge skills
superskill skill package my-skill
superskill skill migrate ./source-skill-1 ./source-skill-2 ./dest-skill

# Emit a hook to a single target agent
superskill hook emit my-hook --target pi

Full walkthrough: Quick start guide.

Supported agents

| Agent | Skills | Commands | Subagents | Hooks | |-------|:------:|:--------:|:---------:|:-----:| | Claude Code | ✓ | ✓ | ✓ | ✓ | | Grok | ✓ | ✓ | ✓ | ✓ | | Codex | ✓ | ✓ | ✓ | ✓ | | Pi | ✓ | ✓ | ✓ | ✓ | | omp | ✓ | ✓ | ✓ | ✓ | | OpenCode | ✓ | ✓ | ✓ | ✓ | | Antigravity IDE | ✓ | ✓ | — | ✓ | | Antigravity CLI | ✓ | ✓ | — | ✓ | | Hermes | ✓ | ✓ | — | ✓ |

Both Antigravity targets get skills, commands, and hooks from rulesync (codexcli/antigravity-cli/antigravity-ide adapters); subagents are not part of the rulesync→antigravity map, so they don't ship there. omp, Grok, and Claude Code install plugins natively (no rulesync pass); the remaining targets route through rulesync's per-target generators. Hermes copies skills/commands from OpenCode's rulesync output and adds a canonical hooks.json copy at ~/.hermes/hooks.json.

Pi loads extensions natively from a plugin's plugin.json extensions.pi array (v0.3.12+), installing them to .pi/agent/plugins/<plugin>/ with generated package.json and settings.json registration — no @vahor/pi-hooks dependency required. Plugins without extensions.pi fall back to the @vahor/pi-hooks hook-emit path.

Grok loads the Claude-format plugin natively — slash form is /plugin:command (e.g. /cc:skill-add). Codex/Pi-installed skills under ~/.agents/ may also appear in Grok as /plugin-command (hyphen form) when Grok discovers the shared skills root; prefer the colon form for plugin commands.

See entity locations for the exact install directories per agent.

Agents that don't natively support some entity types still get them. superskill install adapts commands and subagents as Skills 2.0 skill directories for targets that lack them — so every agent receives the full plugin surface, regardless of native feature set.

Commands

| Command | What it does | Docs | |---------|-------------|------| | install | Distribute a plugin's skills, commands, subagents, magents, hooks, and MCP config to target agents | cmd_install.md | | agent | Manage subagent definitions (scaffold / validate / evaluate / refine / evolve) | cmd_agent.md | | skill | Manage skill definitions (lifecycle + package, migrate) | cmd_skill.md | | command | Manage slash command definitions | cmd_command.md | | hook | Manage hook definitions (+ emit, run) | cmd_hook.md | | magent | Manage main-agent configurations | cmd_magent.md | | script | Run / resolve / build portable twins for plugin scripts (run, path, convert) | how to organize |

The five type commands share a common lifecycle: scaffold → validate → evaluate → refine → evolve, with type-specific quality dimensions and rubrics.

Further reading

| Topic | Document | |-------|----------| | Full help index | docs/help/index.md | | Quality system (rubrics, scoring, evolve gate) | docs/help/quality_system.md | | Entity locations per target agent | docs/help/entity_locations.md | | Bundled cc plugin | docs/help/bundled_plugin.md | | Development guide (stack, build, tests) | docs/help/development.md | | Architecture decisions (authoritative) | docs/00_ADR.md | | Product scope | docs/01_PRD.md | | CLI surface reference | docs/04_DESIGN.md |

License

Apache 2.0