18 AI personas for structured deliberation in OpenCode
18 AI personas that deliberate on your hardest decisions — adapted for OpenCode, the open-source AI coding agent.
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.
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 |
| 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 |
# 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/~/.config/opencode/agents/~/.config/opencode/agents/~/.config/opencode/skills/council/# 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?
3-round structured deliberation:
--quick)2-round rapid analysis for time-sensitive decisions.
--duo)2-member dialectic using polarity pairs (e.g., Socrates vs Feynman).
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)
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
/council — Command parses mode and panel selectionThe 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
agents/council-{name}.md with OpenCode agent frontmatterskills/council/configs/triads.yaml for triad membershipagents/coordinator.md./install.sh# Run a test deliberation
/council --quick Which memory system should I use?
Adapted from Council of High Intelligence by 0xNyk for Claude Code.
MIT