Agent Conductor

icohangar-ops/agent-conductor MCP server

MCP Registry npm PyPI Conformance

Cubiczan stack — Profile · CHP · You are here: agent-conductor

AGENTS.md in, governed agent team out.

Agent Conductor is an MCP server that turns the two conventions the coding-agent ecosystem has converged on — AGENTS.md operating manuals and SKILL.md skills — from passive documentation into an active orchestration layer, with a consensus-hardened decision engine gating high-stakes changes.

• Mirrors: Cubiczan/agent-conductor · codeberg.org/cubiczan/agent-conductor · icohangar-ops/agent-conductor
• License: MIT
• Status: v0.1 — working scaffold; see Roadmap

---

The problem

Every serious agent tool — Claude Code, Cursor, Copilot, Codex, Gemini CLI — now reads an AGENTS.md at the repo root and a catalog of SKILL.md files. But both conventions are honor-system prose:

• Nothing compiles the contract. The non-negotiable rules, layer boundaries, and verification checklists live as markdown the agent may or may not internalize.
• Nothing gates the decision. An agent that's about to rewrite your scoring model proceeds with the same confidence as one renaming a variable.
• Nothing verifies the checklist ran. "Run npm test before handing off" is a suggestion, not a gate.

Conductor makes the conventions executable — without asking any agent tool to change. It ships as a standard MCP server, so anything that speaks MCP gets contract compilation, skill discovery, and decision gating for free.

How it works

MCP client (Claude Code / Cursor / Copilot / ...)
        │  stdio (JSON-RPC, MCP)
        ▼
┌────────────────────────────────────────────────┐
│ TypeScript front end (src/)                    │
│   contract/parser.ts   AGENTS.md → contract    │
│   skills/loader.ts     SKILL.md discovery      │
│   server.ts            7 MCP tools             │
└────────────────┬───────────────────────────────┘
                 │  newline-delimited JSON, child stdio
                 ▼
┌────────────────────────────────────────────────┐
│ Python decision engine (engine/)               │
│   bridge.py → PyPI consensus-hardening-protocol│
│   R0 gates · foundation attacks · lifecycle    │
└────────────────────────────────────────────────┘

Three capability groups:

1. Contract — compile an AGENTS.md into structured mission, non-negotiable rules, layer do/don't boundaries, verification gates, skill recommendations, and an out-of-scope list.
2. Skills — discover SKILL.md skills across project and personal scopes with progressive disclosure: metadata costs ~100 tokens, bodies load only on demand.
3. Decision — gate work through the Consensus Hardening Protocol: a cheap R0 sanity gate before work starts, and an adversarial foundation-attack pass before a high-stakes change locks.

Quick start

npx -y @cubiczan/agent-conductor
pip install consensus-hardening-protocol   # required for decision_* tools

npm: @cubiczan/agent-conductor · PyPI: consensus-hardening-protocol

Requirements: Node 23+ (runs TypeScript natively) and Python 3.10+ with the published CHP package installed.

git clone https://github.com/icohangar-ops/agent-conductor.git
cd agent-conductor
npm install
pip install -r engine/requirements.txt
npm test            # TypeScript tests (parser, skills, live engine bridge)
npm run test:engine # Python bridge protocol tests
npm run build

Register with Claude Code:

claude mcp add agent-conductor -- node /path/to/agent-conductor/dist/index.js

Or in any MCP client's JSON config:

{
  "mcpServers": {
    "agent-conductor": {
      "command": "npx",
      "args": ["-y", "@cubiczan/agent-conductor"]
    }
  }
}

Set CONDUCTOR_PYTHON if your Python 3 lives somewhere other than python3.

Then, from any project that has an AGENTS.md:

"Load this project's agent contract, list its verification gates, and run a decision_adversary pass on the change I'm about to make."

Tool reference

contract_load

Compile an AGENTS.md (or CLAUDE.md) into a structured contract. Accepts a file path or a project directory; defaults to the current working directory.

// input
{ "path": "examples/pipeline-pulse" }

// output (abridged — real output from the bundled example)
{
  "source": "examples/pipeline-pulse/AGENTS.md",
  "title": "AGENTS.md — Pipeline Pulse CRM",
  "mission": "Pipeline Pulse CRM is a lightweight, local-first pipeline review dashboard...",
  "rules": [
    "Deterministic logic — same inputs → same scores, labels, and summaries...",
    "Logic in crm.js — keep main.js thin (fetch, render, events).",
    "... (6 total)"
  ],
  "layers": [
    { "layer": "src/crm.js", "role": "Domain logic",
      "do": "Deterministic scoring, filtering, summaries", "dont": "DOM manipulation" }
  ],
  "gates": [
    { "name": "Code change checklist", "commands": ["npm test"], "notes": "" },
    { "name": "Before completion", "commands": [], "notes": "npm test — all green...\n..." }
  ],
  "skills": [
    { "task": "CRM scoring / forecast changes", "skill": "obra/test-driven-development",
      "url": "https://github.com/obra/superpowers/...", "why": "Tests-first changes to deterministic logic" }
  ],
  "outOfScope": ["External CRM integrations (Salesforce, HubSpot, etc.)", "..."],
  "sectionCount": 28
}

The parser is lossless: sections it doesn't recognize are preserved verbatim, so nothing in an unconventional AGENTS.md is dropped.

contract_verification

Returns only the verification gates — the named checklists and shell commands that must pass before work is handed off. Pair it with your agent's workflow: run the commands, confirm success, then declare done.

skills_list

Discover SKILL.md skills visible from a project root. Metadata only.

// input
{ "projectRoot": "examples/pipeline-pulse" }

// output
{
  "skills": [
    {
      "name": "pipeline-scoring",
      "description": "Explain and modify scoreDealRisk weights in src/crm.js with matching test updates...",
      "version": "0.1.0",
      "scope": "project"
    }
  ]
}

Search order (first hit per skill name wins):

Priority	Path	Scope
1	<project>/.conductor/skills/*/SKILL.md	project
2	<project>/.claude/skills/*/SKILL.md	project
3	<project>/.cursor/skills/*/SKILL.md	project
4	~/.claude/skills/*/SKILL.md	personal
5	~/.cursor/skills/*/SKILL.md	personal

skill_load

Load the full SKILL.md body for one named skill — the on-demand half of progressive disclosure. Call it only when the task matches the skill's description.

decision_gate

The Consensus Hardening Protocol R0 gate: the cheapest, highest-leverage check, run before doing the work.

// input
{ "solvable": true, "scoped": false, "valid": true, "worth_it": true }

// output
{ "verdict": "HALT", "results": { "Solvable": "PASS", "Scoped": "FATAL", "Valid": "PASS", "Worth_it": "PASS" } }

Any FATAL answer halts: stop and reframe before burning tokens on a problem that isn't scoped, isn't understood, or isn't worth solving.

decision_adversary

A one-shot adversarial pass for high-stakes changes: CHP attacks the claim's foundations, scores them 0–100, and returns devil's-advocate findings plus a session status.

// input
{
  "claim": "Change scoreDealRisk stale-activity weight from 20 to 30",
  "context": "Tests updated; label distribution checked against fixture"
}

// output
{
  "status": "EXPLORING",          // or HALT / REFRAME_REQUIRED
  "foundation_score": 77,
  "findings": [
    "Treat every financial number as unverified until tied to source data.",
    "Require explicit flip criteria for any provisional recommendation."
  ],
  "verification_failures": ["PENDING third-party validation"],
  "report": "## TriangulationRunner Adversary Pass\n..."
}

Statuses map to the CHP decision lifecycle (EXPLORING → PROVISIONAL_LOCK → LOCKED, with HALT and REFRAME_REQUIRED exits): EXPLORING means the claim survived the attack and work may proceed toward a lock; HALT/REFRAME_REQUIRED mean the foundations failed.

engine_status

Health-check the Python engine subprocess. Returns { ok, engine: "chp", version }.

What the parser recognizes

contract_load is convention-based, not schema-based. It extracts the patterns AGENTS.md files in the wild actually use:

Contract field	Source convention
mission	First Mission / Purpose / Overview section
rules	List items under Non-negotiables > Engineering rules > generic rules (priority-ordered so a generic "Product rules" section never shadows explicit non-negotiables)
layers	First table with a Layer column under an architecture-like heading
gates	Shell code blocks + list items under checklist / verification / before-completion headings
skills	Tables with Task / Skill / Why columns; links resolved to text + URL
outOfScope	List under an out-of-scope / non-goals heading
sections	Everything, verbatim — the lossless fallback

Headings inside code fences are ignored; tables tolerate emphasis in headers; markdown links and emphasis are stripped from extracted text.

Writing skills

A skill is a directory containing SKILL.md with YAML frontmatter:

---
name: pipeline-scoring
description: Explain and modify scoreDealRisk weights in src/crm.js with matching test updates. Use when changing deal risk scoring, risk labels, or forecast thresholds.
version: 0.1.0
tools: [Read, Edit, Bash]
---

# Pipeline Scoring

Step-by-step instructions the agent follows when the task matches...

Quality bar (inherited from the awesome-agent-skills standards): third-person description with matchable keywords, metadata around 100 tokens, body under 500 lines, no machine-specific absolute paths, declare only the tools the skill needs.

The bundled example — examples/pipeline-pulse — is a complete real-world AGENTS.md plus a project-scoped skill, and is what the test suite compiles.

Project structure

.
├── AGENTS.md                  # This repo's own contract (compiles with itself)
├── ARCHITECTURE.md            # Design decisions and component detail
├── src/
│   ├── index.ts               # stdio entrypoint
│   ├── server.ts              # MCP server: 7 tools
│   ├── contract/              # AGENTS.md → AgentContract compiler
│   ├── skills/                # SKILL.md loader + registry
│   ├── engine/chpBridge.ts    # Python engine client
│   └── utils/logger.ts        # stderr-only logging (stdout is the transport)
├── engine/
│   ├── bridge.py              # JSON-over-stdio router → PyPI `chp`
│   ├── requirements.txt       # consensus-hardening-protocol pin
│   ├── NOTICE.md              # attribution for the published engine
│   └── test_bridge.py         # protocol tests
├── examples/pipeline-pulse/   # real AGENTS.md fixture + example skill
└── test/                      # node:test suites (run the .ts directly)

Development

pip install -r engine/requirements.txt
npm test            # TypeScript tests — includes a live engine round-trip
npm run test:engine # Python-side protocol tests
npx tsc --noEmit    # type check
npm run build       # emit dist/
npm run dev         # run the server from source (Node type stripping)

House rules (the full set is in this repo's own AGENTS.md):

1. stdout is sacred — the MCP transport owns it; all logging goes to stderr on both sides of the bridge.
2. Zero new Node runtime dependencies — only @modelcontextprotocol/sdk and zod; markdown/frontmatter stay hand-rolled. CHP is a PyPI dep.
3. Erasable TypeScript only — source must run under Node's type stripping (no enums, no parameter properties).
4. CHP via PyPI — install consensus-hardening-protocol; do not re-vendor under engine/. Protocol fixes belong upstream.
5. Python 3.10+ — required by the published package.

Roadmap

Version	Theme	Scope
v0.2	Enforcement	Execute contract_verification gates as real subprocesses and return pass/fail evidence — turning "reads the contract" into "enforces the contract"
v0.3	Orchestration	Expose decision_lock + mesh session tools over MCP (multi-agent deliberation on top of published CHP)
v0.4	Registry	Install vetted skills from remote catalogs (awesome-agent-skills format) with source-review prompts

Provenance

Conductor deliberately reuses proven components rather than rewriting them:

Component	Source	License
Decision engine (PyPI)	consensus-hardening-protocol	MIT
MCP server + registry shape	onchainmind	MIT
Skill quality standards	VoltAgent/awesome-agent-skills	—
Example fixture	Pipeline Pulse CRM operating manual	fixture

See engine/NOTICE.md and ARCHITECTURE.md for the two-language design.

---

Cubiczan stack

| Governance | consensus-hardening-protocol · agent-conductor · compliance-as-code-agent · cleanmandate | | Platform | cubiczan-mcp-server · operational-intelligence · software-factory |

Conductor compiles AGENTS.md + SKILL.md into MCP tools and routes high-stakes decisions through CHP — the same lock model Metabocommand uses for finance approvals.

License

MIT — see LICENSE. Vendored components retain their original MIT licenses.