# agents-consilium
Multi-agent orchestration skill for AI coding agents. Query [Codex CLI](https://github.com/openai/codex) and [Gemini CLI](https://github.com/google-gemini/gemini-cli) for independent expert opinions on architecture decisions, code reviews, bug investigations, and more.
## Install
```bash
npx skills add CodeAlive-AI/agents-consilium -g -y
```
## What it does
Consilium gives your AI coding agent the ability to consult other AI agents for a second opinion. Each agent has a distinct thinking role:
- **Codex** = Rigorous Analyst — precision, code correctness, edge cases, security
- **Gemini** = Lateral Thinker — cross-domain patterns, creative alternatives, big picture
Both respond using a structured template (Assessment, Key Findings, Blind Spots, Alternatives, Recommendation) for easy comparison.
## Prerequisites
- [Codex CLI](https://github.com/openai/codex) installed and authenticated
- [Gemini CLI](https://github.com/google-gemini/gemini-cli) installed and authenticated
## Usage
The skill is invoked automatically by your AI agent when it needs a second opinion. You can also trigger it explicitly:
```
/agents-consilium — review this service for security issues
```
### Scripts (direct usage)
```bash
# Single agent queries
scripts/codex-query.sh "your question" [context_file]
scripts/gemini-query.sh "your question" [context_file]
# Consensus query (both agents in parallel)
scripts/consensus-query.sh "architecture question"
# Pipe code for review
cat src/service.py | scripts/codex-query.sh "review this"
```
## When to use which
| Situation | Script | Why |
|-----------|--------|-----|
| Code review, security audit | `codex-query.sh` | Deep implementation analysis |
| Architecture decision | `consensus-query.sh` | Both depth and breadth |
| "Are we solving the right problem?" | `gemini-query.sh` | Challenges premises |
| Bug investigation | `codex-query.sh` | Implementation-level detail |
| Exploring alternatives | `gemini-query.sh` | Cross-domain analogies |
| High-stakes decision | `consensus-query.sh` | Two perspectives reduce blind spots |
## Environment variables
| Variable | Default | Description |
|----------|---------|-------------|
| `CODEX_MODEL` | `gpt-5.4` | Override Codex model |
| `GEMINI_MODEL` | `gemini-3.1-pro-preview` | Override Gemini model |
| `AGENT_TIMEOUT` | `1200` | Timeout in seconds |
## Design principles
- **Read-only mode**: External agents analyze but never modify files
- **Anti-bias protocol**: Questions are framed neutrally to get independent opinions
- **Intellectual independence**: Agents think from first principles and challenge assumptions
- **Clean output**: Only final answers are captured — no intermediate logs or session noise
## License
MIT