Quick Start
New to Trinity? Paste this into Claude Code, then run /trinity:start-here:
/plugin marketplace add abilityai/abilities
/plugin install trinity@abilityai
/trinity:start-here is the guided journey — what Trinity is → get an instance → connect MCP (with a live smoke test) → your first agent alive. Resumable at every step; nothing is deployed until you choose it.
# Add the abilities marketplace (one-time)
/plugin marketplace add abilityai/abilities
# List available plugins
/plugin list abilityai
# Install the core plugins
/plugin install create-agent@abilityai
/plugin install agent-dev@abilityai
/plugin install trinity@abilityai
Or from the terminal:
claude plugin add abilityai/abilities
claude plugin install create-agent@abilityai
Creating Agents
The fastest way to get started is /create-agent:create — a single entry point that shows all available creation paths:
/create-agent:create
Available Wizards
All agent creation wizards are consolidated into the create-agent plugin. Use /create-agent:[wizard] syntax:
| Wizard | Command | What it creates |
|---|
| custom | /create-agent:custom | Any agent, from an interview — role, skills, schedules, credentials, Trinity wiring |
| website | /create-agent:website | Single website scaffold (no agent, just a site) |
| clone | /create-agent:clone | Clone an existing agent repository as starting point |
| adjust | /create-agent:adjust | Review and improve an existing agent |
Every wizard-created agent includes:
/onboarding — Persistent checklist that tracks setup progress across sessionsdashboard.yaml — Domain-specific metrics dashboard for Trinity/update-dashboard — Schedulable skill that keeps dashboard metrics current
All Plugins
The abilities marketplace contains 5 focused plugins:
create-agent
Create new Claude Code agents with domain-specific wizards or from scratch.
/plugin install create-agent@abilityai
| Skill | Description |
|---|
/create-agent:create | Entry point — shows all available wizards |
/create-agent:custom | Interview-driven agent for any domain |
/create-agent:website | Single website (no agent) |
/create-agent:clone | Clone existing agent |
/create-agent:adjust | Modify existing agent |
agent-dev
Develop and extend existing agents with skills, memory systems, and workflows.
/plugin install agent-dev@abilityai
| Skill | Description |
|---|
/agent-dev:create-playbook | Create a new skill/playbook for the agent |
/agent-dev:adjust-playbook | Modify an existing skill/playbook |
/agent-dev:add-memory | Add memory system (file-index, brain, json-state, workspace) |
/agent-dev:add-backlog | Add GitHub Issues backlog workflow |
/agent-dev:add-git-sync | Add auto-commit hooks for durable cross-session state |
/agent-dev:add-orchestrator | Make the agent a system-aware orchestrator of other agents |
/agent-dev:add-pipeline | Add a long-running multi-stage pipeline |
/agent-dev:add-pipeline-instance | Add an instance to a pipeline |
/agent-dev:add-pipeline-stage | Add a stage to a pipeline |
/agent-dev:validate-pipeline | Lint a pipeline definition |
/agent-dev:backlog | View GitHub Issues backlog |
/agent-dev:groom | Tag issues with skill:* labels, set priorities |
/agent-dev:roadmap | View issues grouped by affected skill |
/agent-dev:claim | Claim the next issue to work on |
/agent-dev:autoplan | Plan a claimed issue against its SKILL.md |
/agent-dev:commit | Commit skill changes and close the issue |
/agent-dev:close | Close an issue without a commit |
/agent-dev:sprint | Guided full cycle: roadmap → claim → autoplan → commit |
/agent-dev:work-loop | Run autonomous work loop |
/agent-dev:plan | Plan multi-session work |
Memory systems (installed via /agent-dev:add-memory):
- file-index — Workspace file awareness and search
- brain — Zettelkasten-style knowledge graph
- json-state — Structured JSON state with jq updates
- workspace — Multi-session project tracking
trinity
Connect, deploy, and sync agents to the Trinity platform.
/plugin install trinity@abilityai
| Skill | Description |
|---|
/trinity:connect | One-time: authenticate and configure MCP connection |
/trinity:onboard | Per-agent: compatibility check, file creation, deploy |
/trinity:sync | Ongoing: sync changes between local and remote |
After connection, Trinity MCP tools are available directly:
mcp__trinity__list_agents — List all remote agentsmcp__trinity__chat_with_agent — Send messages to remote agentsmcp__trinity__create_agent — Deploy an agent from its GitHub repo (template: "github:owner/repo") — the default pathmcp__trinity__deploy_local_agent — Deploy from a local archive — the fallback path
dev-methodology
Documentation-driven development methodology for any codebase.
/plugin install dev-methodology@abilityai
| Skill | Description |
|---|
/dev-methodology:init | Scaffold methodology into your project |
/dev-methodology:read-docs | Load project context at session start |
/dev-methodology:implement | End-to-end feature implementation |
/dev-methodology:validate-pr | Validate PR against methodology |
/dev-methodology:commit | Create well-formatted commits |
/dev-methodology:security-check | Quick security scan |
/dev-methodology:security-analysis | Deep security analysis |
/dev-methodology:add-testing | Add tests to existing code |
/dev-methodology:tidy | Clean up code |
/dev-methodology:roadmap | Generate project roadmap |
utilities
General-purpose ops and productivity skills.
/plugin install utilities@abilityai
| Skill | Description |
|---|
/utilities:save-conversation | Save conversation as structured markdown |
/utilities:investigate-incident | Structured incident investigation |
/utilities:bug-report | Create sanitized GitHub issue |
/utilities:safe-deploy | Safe deployment with backup/rollback |
/utilities:docker-ops | Docker container management |
/utilities:sync-ops-knowledge | Update ops docs from commits |
/utilities:batch-claude-loop | Batch headless Claude Code calls |
The Agent Development Workflow
Abilities supports a four-step workflow for building agents that appreciate over time:
1. Scaffold 2. Develop 3. Deploy 4. Iterate
/create-agent:* /agent-dev:create-playbook gh repo create ... git push
/agent-dev:add-memory /trinity:onboard /trinity:sync
/agent-dev:add-backlog (deploys from the /create-agent:adjust
repo Trinity clones)
Scaffold — Use /create-agent:custom to get a fully configured agent with CLAUDE.md, skills, Trinity files, and an onboarding tracker.
Develop — Use /agent-dev:create-playbook to add capabilities, /agent-dev:add-memory to add persistence, /agent-dev:add-backlog for task management, and /agent-dev:add-orchestrator to make an agent aware of — and able to drive — other agents.
Deploy — from the agent's GitHub repository. Run /trinity:connect once to authenticate, add a GitHub token to your instance (Settings → GitHub token — a fine-grained PAT with Contents: Read), push each agent to a repo, then run /trinity:onboard in its directory. Trinity clones the repo and tracks the branch, so the deployed agent is a named commit, its template.yaml schedules are materialized at creation, and every later change ships with git push. Deploying from local files remains available as the fallback for an agent that has no repo yet — /trinity:onboard offers to move it onto the repo path afterwards.
Iterate — Push changes with git push, then /trinity:sync (or mcp__trinity__git_pull) to advance the deployed agent to the new commit. Use /create-agent:adjust to audit and improve.
Plugin Details
create-agent
Agent creation — the custom interview builds a fully configured, Trinity-compatible agent for any domain; website scaffolds a site. (The eight pre-built domain wizards were retired in create-agent 2.0.0, 2026-09-14 — every shape is reachable through custom.)
/create-agent:create # Discovery — shows all wizards
/create-agent:custom # Interview-driven agent wizard
/create-agent:website # Single website scaffold
/create-agent:adjust # Improve existing agent
Generated agents include:
- CLAUDE.md — Identity, behavioral instructions, artifact dependency graph
- Initial skills — 2-4
.claude/skills/ playbooks based on purpose
- Onboarding system —
onboarding.json + /onboarding skill
- Dashboard —
dashboard.yaml + /update-dashboard skill
- Trinity files —
template.yaml, .env.example, .mcp.json.template
- Git repo — Initialized and committed
agent-dev
Development tools for extending existing agents. Memory systems are copied directly into agents — no plugin dependency at runtime.
/agent-dev:create-playbook # Create new skill/playbook
/agent-dev:adjust-playbook # Modify existing skill/playbook
/agent-dev:add-memory # Add memory system
/agent-dev:add-backlog # Add GitHub Issues workflow
/agent-dev:add-orchestrator # Make the agent a system-aware orchestrator
/agent-dev:work-loop # Autonomous work loop
Memory options:
- file-index — Agent needs awareness of workspace files
- brain — Connected notes, knowledge graph (Zettelkasten)
- json-state — Structured state, counters, config
- workspace — Multi-session project tracking
trinity
Simplified deployment to the Trinity platform — 3 skills covering the complete workflow.
/trinity:connect # One-time authentication
# then: Settings → GitHub token on your instance (fine-grained PAT, Contents: Read)
# then: gh repo create <agent> --private --source=. --push
/trinity:onboard # Deploy the agent — from its GitHub repo
/trinity:sync # Ship changes: push locally, remote pulls
What is Trinity? The operating system for the AI-native company — open source, self-hosted, that you own:
- Autonomous operation — Agents run 24/7 with cron-based scheduling
- Multi-agent orchestration — Coordinate teams of specialized agents
- Human-in-the-loop — Approval gates where decisions matter
- Enterprise controls — Audit trails, cost tracking, Docker isolation
- Your infrastructure — Self-hosted, data never leaves your perimeter
dev-methodology
Documentation-driven development for any codebase. Enforces a 5-phase cycle: context loading, development, testing, documentation, and PR validation.
/dev-methodology:init # Scaffold methodology
/dev-methodology:read-docs # Load context at session start
/dev-methodology:implement #42 # End-to-end feature implementation
/dev-methodology:validate-pr 123 # Validate PR
utilities
General-purpose ops and productivity skills for SSH-accessible services and daily workflows.
/utilities:investigate-incident # Structured incident investigation
/utilities:safe-deploy update # Safe deployment with backup
/utilities:safe-deploy rollback # Revert to previous commit
/utilities:docker-ops logs [service] # View container logs
/utilities:save-conversation # Save conversation as markdown
Installation
From Marketplace (Recommended)
# Add marketplace (one-time)
/plugin marketplace add abilityai/abilities
# Install plugins
/plugin install create-agent@abilityai
/plugin install agent-dev@abilityai
/plugin install trinity@abilityai
/plugin install dev-methodology@abilityai
/plugin install utilities@abilityai
Manual Installation
# Clone the repository
git clone https://github.com/abilityai/abilities.git
# Install a plugin directly
/plugin add ./abilities/plugins/create-agent
Contributing
We welcome contributions! See CLAUDE.md for development guidelines.
Adding a Plugin
- Create a new directory in
plugins/
- Add
.claude-plugin/plugin.json with plugin metadata
- Add your skills in
skills/[skill-name]/SKILL.md
- Add a
README.md with usage documentation
- Register in
.claude-plugin/marketplace.json
- Submit a pull request
Creating Install Wizards
To add a new agent creation wizard:
- Create a skill directory in
plugins/create-agent/skills/[wizard-name]/
- Add a
SKILL.md with the guided question flow
- Update the
/create entry point skill to include the new wizard
- Update
marketplace.json to register the new skill
License
MIT
Support
