Cloud Code Plugins/Trinity Onboard

Trinity Onboard

Complete Trinity platform integration for Claude Code agents. Provides the canonical skills for adopting, syncing, and operating agents on Trinity's autonomous agent orchestration infrastructure.

Prerequisites

You need access to a Trinity instance before using this plugin.

•Self-Host (Open Source) — Deploy your own instance from github.com/abilityai/trinity

•Managed by Ability AI — Contact trinity@ability.ai for a managed instance

Item	Description
Trinity URL	Your instance URL (e.g., `https://trinity.example.com`)
API Key	From Trinity dashboard > Settings > API Keys

Credentials are resolved automatically in this order: TRINITY_API_KEY / TRINITY_URL environment variables, then Trinity CLI profile (~/.trinity/config.json), then interactive prompt. If the Trinity CLI is installed, /trinity-onboard will use trinity deploy . for deployment, which handles versioning and tracking automatically.

Installation

# One-liner (if you already have the abilityai marketplace)
claude plugin install trinity-onboard@abilityai

# First time with abilityai marketplace
claude plugin add abilityai/abilities && claude plugin install trinity-onboard@abilityai

# Or from inside a Claude Code session
/plugin marketplace add abilityai/abilities
/plugin install trinity-onboard@abilityai

How It Works: Local-Remote Pairing

Trinity uses a paired agent architecture — the same agent runs both locally and remotely, synchronized through Git:

•Same agent, two locations — Both share identical skills and instructions, synced via Git

•GitHub as state management — Your agent's identity (skills, config) lives in Git

•Local orchestrator — Your local session triggers and monitors remote execution

•Heartbeat pattern — Local agent polls remote for long-running task status

•Event-driven pipelines — Agents emit events, subscribers receive automated tasks

•Credentials stay local — .env and .mcp.json are gitignored, synced separately

Mode	Command	Use Case
Execute	/trinity-remote exec <task>	Run task on remote, get response
Deploy-Run	/trinity-remote run <task>	Sync changes first, then execute
Fan-Out	/trinity-remote fan-out <task>	Parallel self-invocation (1-50 tasks)
Scheduled	/trinity-schedules	Cron-based autonomous execution
Heartbeat	/create-heartbeat	Local-controlled polling loop
Events	/trinity-events	Inter-agent pub/sub pipelines

Core Trinity Skills

Skill	Description
/trinity-onboard	Convert any agent to Trinity-compatible format
/trinity-compatibility	Read-only audit of agent structure
/trinity-sync	Git-based synchronization with remote agent
/trinity-remote	Remote agent operations (exec, run, notify)
/trinity-schedules	Manage scheduled autonomous executions
/trinity-events	Inter-agent event subscriptions and pub/sub pipelines

Extended Skills

Skill	Description
/credential-sync	Push/pull credentials between local and remote
/create-heartbeat	Generate polling skills for long-running tasks
/create-dashboard-playbook	Generate a skill to keep dashboard.yaml updated
/request-trinity-access	Request Trinity access via email + tweet verification

Workflow

The onboarding flow has two paths:

•Deploy to Trinity — Full setup with credentials, MCP connection, and remote deployment via trinity deploy . (CLI) or MCP-based deploy (fallback)

•Adapt only — Create Trinity-compatible files without connecting to a Trinity instance

After onboarding, the workflow is: Adopt → Develop → Sync → Remote → Schedule → Events

Quick Start

# 1. Check compatibility
/trinity-compatibility

# 2. Adopt Trinity methodology
/trinity-onboard

# 3. Sync with remote
/trinity-sync status
/trinity-sync push
/trinity-sync pull

# 4. Remote operations
/trinity-remote
/trinity-remote exec <task>
/trinity-remote run <task>

# 5. Schedule autonomous tasks
/trinity-schedules status
/trinity-schedules schedule my-skill "0 9 * * *"
/trinity-schedules trigger my-skill

# 6. Event-driven pipelines
/trinity-events list
/trinity-events subscribe agent-a task.done "Process {{payload.result}}"
/trinity-events emit report.generated {"report_id": "rpt-1"}

Credential Management

# Push local credentials to remote
/credential-sync push
/credential-sync push --files=.env

# Encrypted Git storage
/credential-sync export    # Create .credentials.enc (safe to commit)
/credential-sync import    # Restore from encrypted backup

# Check status
/credential-sync status

Required Files

File	Purpose
CLAUDE.md	Agent instructions (the "brain")
template.yaml	Trinity deployment metadata
.mcp.json.template	MCP config with `${VAR}` placeholders
.env.example	Document required environment variables
.gitignore	Security-critical exclusions
.trinity-remote.yaml	Deployment tracking (auto-generated by `trinity deploy .`)

Platform Capabilities (Post-Onboarding)

Once deployed to Trinity, agents gain access to these platform features:

Persistent Setup Script

Place a script at ~/.trinity/setup.sh to persist system-level packages (apt-get, npm -g, pip) across container restarts. The script runs automatically on every container start.

Voice Chat

Agents can accept real-time voice conversations via the Gemini Live API. Enable by setting GEMINI_API_KEY and VOICE_ENABLED=true, then creating a voice-agent-system-prompt.md file on the remote agent.

Channel Adapters (Slack & Telegram)

Agents can receive and respond to messages from Slack channels and Telegram groups. Each agent gets its own dedicated channel with identity customization (name + avatar). Configure from Trinity Settings (Slack) or via ANNOUNCE_TELEGRAM_TOKEN (Telegram).

Per-User Persistent Memory

Public-facing agents automatically maintain per-user memory for email-verified visitors. Memory is scoped to (agent_name, user_email) and injected into every conversation. No configuration needed — automatic when using public links with email verification.

Execution Query Tools

Three MCP tools enable programmatic monitoring: list_recent_executions, get_execution_result, and get_agent_activity_summary. Useful for orchestrator agents monitoring worker fleets and polling async task results.

Project Planner Utilities