council-of-high-intelligence-opencode

18 AI personas for structured deliberation in OpenCode

0
0
0
public
Forked

Council of High Intelligence for OpenCode

18 AI personas that deliberate on your hardest decisions — adapted for OpenCode, the open-source AI coding agent.

What is this?

A multi-persona deliberation system where historical thinkers (Aristotle, Feynman, Kahneman, Torvalds, and 14 more) analyze complex decisions through structured disagreement. Each persona is modeled as an OpenCode subagent with distinct analytical methods, blind spots, and polarities.

Why OpenCode?

Unlike Claude Code’s single-provider architecture, this adaptation distributes personas across multiple models for genuine cognitive diversity:

Model Personas Thinking Style
GLM-5 Aristotle, Socrates, Aurelius, Lao Tzu, Watts, Sutskever, Kahneman, Taleb Deep reasoning, philosophy, safety
Kimi K2.5 Feynman, Torvalds, Sun Tzu, Ada, Machiavelli, Musashi, Karpathy, Meadows, Munger, Rams Empirical, pragmatic, building

The 18 Council Members

Persona Domain Polarity
Aristotle Categorization & structure Classifies everything
Socrates Assumption destruction Questions everything
Sun Tzu Adversarial strategy Reads terrain & competition
Ada Lovelace Formal systems & abstraction What can/can’t be mechanized
Marcus Aurelius Resilience & moral clarity Control vs acceptance
Machiavelli Power dynamics & realpolitik How actors actually behave
Lao Tzu Non-action & emergence When less is more
Feynman First-principles debugging Refuses unexplained complexity
Linus Torvalds Pragmatic engineering Ship it or shut up
Musashi Strategic timing The decisive strike
Alan Watts Perspective & reframing Dissolves false problems
Karpathy Neural network intuition How models actually learn and fail
Sutskever Scaling frontier & AI safety When capability becomes risk
Kahneman Cognitive bias & decision science Your own thinking is the first error
Meadows Systems thinking & feedback loops Redesign the system, not the symptom
Munger Multi-model reasoning & economics Invert — what guarantees failure?
Taleb Antifragility & tail risk Design for the tail, not the average
Dieter Rams User-centered design Less, but better — the user decides

Installation

# Clone and install
git clone https://github.com/yourusername/council-of-high-intelligence-opencode.git
cd council-of-high-intelligence-opencode
./install.sh

This installs:

  • /council command → ~/.config/opencode/commands/
  • 18 persona agents → ~/.config/opencode/agents/
  • Coordinator agent → ~/.config/opencode/agents/
  • Skill documentation → ~/.config/opencode/skills/council/

Usage

# Full deliberation (3 rounds, auto-selected panel)
/council Should we migrate to microservices?

# Quick mode (2 rounds)
/council --quick Should we add Redis caching?

# Duo dialectic (2-member polarity pair)
/council --duo Should we build or buy?

# Specific domain triad
/council --triad strategy What's our competitive moat?

# Custom panel
/council --members socrates,feynman,ada Is our architecture sound?

# Profile panel
/council --profile execution-lean Should we ship this feature?

Deliberation Modes

Full Mode (default)

3-round structured deliberation:

  1. Independent Analysis — Each persona analyzes blind-first
  2. Cross-Examination — Members engage each other’s positions
  3. Final Crystallization — 100-word position statements

Quick Mode (--quick)

2-round rapid analysis for time-sensitive decisions.

Duo Mode (--duo)

2-member dialectic using polarity pairs (e.g., Socrates vs Feynman).

Available Triads

Pre-built 3-member panels for specific domains:

Triad Members Use For
architecture Aristotle + Ada + Feynman System design, technical decisions
strategy Sun Tzu + Machiavelli + Aurelius Competitive positioning, market entry
ethics Aurelius + Socrates + Lao Tzu Moral dilemmas, duty conflicts
debugging Feynman + Socrates + Ada Root cause analysis, bug hunting
ai-safety Sutskever + Aurelius + Socrates AI alignment, capability-risk tradeoffs
design Rams + Torvalds + Watts UX decisions, product simplification
decision Kahneman + Munger + Aurelius High-stakes choices, bias detection
shipping Torvalds + Musashi + Feynman Release timing, feature scope

(20 triads total — see skills/council/configs/triads.yaml)

Architecture

council/
├── commands/
│   └── council.md              # /council slash command entry point
├── agents/
│   ├── coordinator.md          # Orchestrates deliberation
│   ├── council-aristotle.md    # Persona subagents (18 files)
│   ├── council-socrates.md
│   └── ...
└── skills/
    └── council/
        ├── SKILL.md            # Protocol documentation
        └── configs/
            ├── triads.yaml     # Triad definitions
            └── profiles.yaml   # Panel profiles

How It Works

  1. User invokes /council — Command parses mode and panel selection
  2. Coordinator loads — Reads skill protocol and persona definitions
  3. Multi-provider routing — Assigns personas to GLM-5 or Kimi K2.5
  4. Parallel spawning — Task tool spawns persona subagents for each round
  5. Structured deliberation — Independent → Cross-exam → Final positions
  6. Verdict synthesis — Coordinator produces formatted verdict

Enforcement Mechanisms

  • Dissent quota — At least 2 members must articulate non-overlapping objections
  • Novelty gate — Each round must add new insight not present in previous rounds
  • Anti-recursion — Socrates gets 3-level questioning depth limit
  • Agreement check — If >70% agree, force 2 dissenters to steelman opposing view

Configuration

Model Assignments (IMPORTANT)

The default model assignments are configured for our local setup. You must adapt them to your available models.

Current assignments:

Model Personas
GLM-5 (bailian-coding-plan) Aristotle, Socrates, Aurelius, Lao Tzu, Watts, Sutskever, Kahneman, Taleb
Kimi K2.5 (kimi-for-coding) Feynman, Torvalds, Sun Tzu, Ada, Machiavelli, Musashi, Karpathy, Meadows, Munger, Rams

Principle: Split polarity pairs across different models for genuine cognitive diversity. Assign “deep reasoning” personas to your most capable model and “pragmatic/building” personas to your secondary model.

To find your configured models:

# List all available models in your OpenCode setup
opencode models

# Or check your config file
cat ~/.config/opencode/opencode.json | grep -A 5 '"models"'

To change model assignments:

Edit each persona agent file in agents/council-{name}.md:

---
model: your-provider/your-model   # Change this line
temperature: 0.3                   # Adjust if needed
---

Example: If you only have one model (e.g., gpt-4), assign all personas to it:

---
model: openai/gpt-4
---

After changing models, reinstall:

./install.sh

Adding New Personas

  1. Create agents/council-{name}.md with OpenCode agent frontmatter
  2. Add to skills/council/configs/triads.yaml for triad membership
  3. Add polarity pairs in agents/coordinator.md
  4. Reinstall with ./install.sh

Testing

# Run a test deliberation
/council --quick Which memory system should I use?

Credits

Adapted from Council of High Intelligence by 0xNyk for Claude Code.

License

MIT

v0.3.3[beta]