mmpm-cognitive-os
Constitutional multi-agent decision architecture for major programme governance. 16 agents across 4 layers (Programme Cognitive Chair, 7 Str
Open source Repository Open in the app JSON README (API)
About
Constitutional multi-agent decision architecture for major programme governance. 16 agents across 4 layers (Programme Cognitive Chair, 7 Strategic Domain agents, 5 Cross-Domain Coupling agents, 3 Adversarial Governance agents) grounded in Oxford MMPM doctrine. Supports 6 operating modes: Tutor (doctrine explanation), Analyst (focused domain analysis), Council (full multi-agent deliberation), Exploration (structured scenario analysis), Research (academic support), and Brief (executive summaries). Features constitutional principles with operational checks, evidence quality gates, dissent preservation, confidence aggregation, and mandatory adversarial review for high-stakes decisions.
Details
- Kind
- Plugins
- Topic
- AI, RAG & memory
- Publisher
- chikeka
- Origin
- marketplace
- Category
- ferramentas
- Last push
- 2026-04-21T22:37:12Z
- Repository state
- ativo
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
chikeka/mmpm-cos/mmpm-cognitive-os
README
# MMPM Cognitive Operating System (MMPM-COS)
A constitutional multi-agent decision architecture for major programme governance, grounded in the Oxford Major Programme Management (MMPM) lectures.
MMPM-COS is a **governed deliberation engine** — not a project management tool — designed to improve decision quality, systemic robustness, and governance integrity in major programmes under uncertainty.
---
## Architecture
### 4-Layer Agent Model (16 Agents)
```
LAYER 1 — PROGRAMME COGNITIVE CHAIR
Orchestrates deliberation. Classifies questions, infers parameters,
routes to agents, runs the evidence quality gate, synthesises findings,
and preserves dissent.
LAYER 2 — STRATEGIC DOMAIN INTELLIGENCE (7 Agents)
Organizational Design and Architecture
Governance, Stakeholder and Institutional Leadership
Commercial and Contracting
Risk, Uncertainty and Fragility
Strategic Leadership and Performance
Socio-Technical Impact and Legitimacy
Futures and Foresight
LAYER 3 — CROSS-DOMAIN COUPLING INTELLIGENCE (5 Agents)
Interface Integrity
Cascading Effects
Temporal Dynamics
Constraint and Tradeoff
Emergence and Fragility
LAYER 4 — ADVERSARIAL GOVERNANCE (3 Agents)
Contrarian
Failure Mode / Red Team
Public Value and Ethics
```
### Orchestration Flow
```
1. Intake 6. Coupling Review 11. Synthesis
2. Classification 7. Challenge Round 12. Memory Write
3. Parameter Inference 8. Rebuttal Round
4. Agent Selection 9. Adversarial Review
5. Domain Assessment 10. Evidence Quality Gate
```
Unresolved disagreement is **preserved as tension**, not iterated away.
---
## Theoretical Foundations
The system draws on the Oxford MMPM lecture curriculum and its core academic literature, including:
- **The S3 Framework** (Armanios, Ventresca, Itani & McCulloch, 2025) — Major Program Value Creation and Capture: mitigating risk propagation to maximize opportunities across distributed value networks
- **Armanios** on infrastructure governance, resilience, and institutional disruption — foundational to the programme's treatment of how major programmes create and capture value in complex institutional environments
- **Flyvbjerg** on reference class forecasting, optimism bias, and the iron law of megaproject management
- **Kahneman** on the planning fallacy, inside view vs. outside view, and decision quality under uncertainty
- **Meadows** on systems thinking, leverage points, and feedback dynamics
- **Drummond** on escalation of commitment in megaprojects
- **Taleb** on black swan risk and fat-tailed distributions
- **Perrow** on normal accidents and tight coupling in complex systems
- **Galbraith** on organizational design and information-processing theory
- **Tetlock** on epistemic calibration and expert judgment
Each agent definition includes specific references to MMPM lecture modules, named authors, and the concepts that ground its reasoning.
---
## Commands
| Command | Mode | Description |
|---|---|---|
| `/context` | Setup | Set programme context (sector, politics, constraints) |
| `/tutor` | Light | Explain MMPM concepts from the lecture curriculum |
| `/analyst` | Medium | Focused domain analysis (1-3 agents) |
| `/council` | Full | Multi-agent governance deliberation |
| `/scenario` | Full | Structured "what if" consequence mapping |
| `/research` | Medium | Academic research framing |
| `/brief` | Compressed | Board-ready executive summary |
| `/memory` | Utility | Review past decisions |
---
## Constitutional Principles
12 principles bind all agents and outputs:
1. Public Value Over Cosmetic Success
2. Uncertainty Must Be Surfaced
3. Cross-Domain Interactions Over Isolated Optimization
4. Optimism Bias Must Be Challenged
5. Governance Quality Is Performance Quality
6. Evidence Strength Must Match Claim Strength
7. Separate Facts, Assumptions, Interpretations, and Speculation
8. Stakeholder Legitimacy Is an Operating Condition
9. Temporal Mismatches Must Be Made Explicit
10. Socio-Material Effects Must Not Be Ignored
11. Traceability of Reasoning
12. Robust Judgment Over Fluent Synthesis
---
## Mandatory Routing Rules
| Condition | Required Agent |
|---|---|
| >1 domain agent active | Interface Integrity |
| Consequential decision | Contrarian |
| Irreversible downside | Red Team |
| Public/societal effects | Public Value and Ethics |
| Timing is central | Temporal Dynamics |
| Competing options | Constraint and Tradeoff |
| Tight coupling detected | Emergence and Fragility |
---
## Parameter System
The Chair infers parameters from the question and programme context:
- **Strategic**: criticality, transformation ambition, public visibility
- **Governance**: stakeholder fragmentation, decision latency, sponsor sensitivity
- **Risk**: tolerance, tail-risk sensitivity, schedule rigidity
- **Impact**: equity weighting, cultural sensitivity, environmental sensitivity
- **Futures**: time horizon, disruption sensitivity
Parameters rated **high** trigger mandatory agent review. Conflicting parameters invoke the Constraint and Tradeoff Agent.
---
## Project Structure
```
MMPM Agent/
├── CLAUDE.md # Constitutional core
├── .claude/
│ ├── agents/ # 16 agent definitions
│ └── commands/ # 8 slash commands
├── configs/
│ ├── constitution/ # Interaction protocol, memory system
│ ├── routing/ # Agent selection logic
│ ├── parameters/ # Parameter schemas and defaults
│ └── output-contracts/ # Structured output templates
├── memory/
│ ├── decisions/ # Decision log
│ └── practitioner-context/ # User annotations
└── docs/
└── source-design/ # Original design documents
```
---
## Getting Started
1. Install [Claude Code](https://docs.anthropic.com/en/docs/claude-code) CLI
2. Open this project in Claude Code
3. Run `/context` to set your programme context
4. Run `/council`, `/analyst`, `/scenario`, or `/tutor` with your question
Opus 4.6 recommended for full Council deliberations.