M87 Governed AgentChain

Early prototype of the M87 governed orchestration architecture. Built July 2025 as part of the original Filament governance research.

---

What This Is

An early implementation of governed Claude sub-agent orchestration — modular task delegation, test integrity enforcement, and output safety handling via Claude's ~/.claude/commands/ system.

This repo represents the first working version of patterns that evolved into the current M87 governance stack. The core ideas here — proposal-gated execution, artifact-backed output, structured audit loops — are now implemented more fully in Spine Lite and m87-audit-agent.

---

Structure

.
├── .claude/
│   └── commands/
│       └── m87_governed_agentchain.md   # Main orchestrator file
├── examples/
│   ├── TASK.md                          # Task spec file
│   └── emotion_signal.json              # Emotional state input
├── docs/
│   └── governance-overview.pdf          # Architecture overview
└── README.md

---

Quick Start

1. Copy the command file into your Claude commands folder:

mkdir -p ~/.claude/commands/
cp .claude/commands/m87_governed_agentchain.md ~/.claude/commands/

2. Run inside Claude:

/m87_governed_agentchain

---

Governance Patterns (Early Implementation)

• SPOT — Output policy violation detection
• FORT — Test integrity enforcement, prevents test modification or leakage
• PARCEL — Safe output delivery and containment

These became the SPOT and FORT rule taxonomy in the current audit agent ruleset.

---

Current Work

For production-ready governed agent infrastructure:

• M87-Spine-lite — Runtime governance for Claude Code. Policy-gated execution, cryptographic receipts, posture escalation.
• m87-audit-agent — Static code auditor. Deterministic enforcement, SHA-256 receipt chain, CI-ready.

---

License

MIT — © 2025 M87 Studio LLC