trinity
A coordination-first AI employee OS tuned for watched threads, TODO-first heartbeat execution, file-backed memory, and evidence-driven delivery.
| Field | Value |
|---|---|
| Status | Draft |
| Version | v0.1.0 |
| Family | manager-heavy |
| Spec | >=1.0.0 <2.0.0 |
| Persona | Coordinator-first |
| Heartbeat | Proactive |
Description
Trinity is designed for AI agents that must act like accountable operators, not passive chatbots. It emphasizes watched DM/thread environments where every action needs a short receipt, strict follow-up discipline, and evidence-driven delivery.
Use Cases
- Owner-facing AI employee workspaces
- Slack / IM-heavy execution loops
- Multi-agent coordination with strict follow-up discipline
- Environments where every outbound message requires post-send verification
Skills
Required (included_skills)
| Skill | Description |
|---|---|
| agent-manager | Employee agent lifecycle management |
| use-fractalbot | Outbound messaging via fractalbot CLI |
| turbo-frequency | High-frequency execution mode |
Workspace Files
After installation, your workspace will contain:
Core OS files
| File | Purpose |
|---|---|
SYSTEM.md | Kernel contract |
SOUL.md | Operating principles |
AGENTS.md | Session bootstrap and rules |
USER.md | Operator profile |
HEARTBEAT.md | TODO-first execution surface |
OKR.md | Active OKR control plane |
TODO.md | Temporary work tracker |
MEMORY.md | Long-term memory |
TOOLS.md | Environment notes |
Memory system
| Path | Purpose |
|---|---|
memory/index.md | Topic navigation |
memory/watched-dm-receipts.md | Watched thread receipts |
OKR system
| Path | Purpose |
|---|---|
okrs/Candidate.md | Non-active OKR parking lot |
Key Principles
- Result-oriented, not step-oriented: Focus on outcomes, not process
- Post-send verification: Not just "sent" but "delivered correctly"
- Watched DM/thread discipline: Every action change produces a short receipt
- TODO-first heartbeat: Urgent work before OKR progression
- Evidence-driven delivery: Prove completion with artifacts
Version History
| Version | Date | Highlights |
|---|---|---|
| v0.1.0 | 2026-04-12 | Initial draft — AI employee OS with watched thread discipline |
View full release notes on GitHub
Install
See the Installation Guide for step-by-step instructions.