npx agentspec-cli install 44868c07-c689-4cf1-b711-38358db95ab4 --file-type agents-md→ .agents/
--- title: Agents description: Understanding the multi-agent system --- # Agent System Cuttlefish uses specialized agents for different tasks, each optimized for its role through category-based model routing. ## Core Agents ### Orchestrator Receives requests, creates task plans, and coordinates other agents. - **Category**: `deep` - **Model**: Claude Sonnet 4.6 (recommended) - **Role**: Strategic planning and task decomposition ### Coder Writes code, executes commands, and modifies files in the sandbox. - **Category**: `deep` - **Model**: Claude Sonnet 4.6 (recommended) - **Role**: Implementation and execution ### Critic Reviews code, runs tests, and approves or rejects changes. - **Category**: `unspecified-high` - **Model**: Configurable (Claude Sonnet 4.6 or Opus 4.6) - **Role**: Quality assurance and validation ## Workflow The system implements a **Planner → Coder → Critic** loop: ``` 1. User sends request ↓ 2. Orchestrator analyzes and creates plan ↓ 3. Coder implements changes in sandbox ↓ 4. Critic reviews code and runs tests ↓ 5. If approved: commit and push If rejected: loop back to step 3 ``` ## Future Agents (v1.1+) - **Planner**: Creates strategic implementation plans (`ultrabrain` category) - **Explorer**: Searches codebases and finds patterns (`quick` category) - **Librarian**: Finds documentation and retrieves external resources (`quick` category) - **DevOps**: Handles builds, deployments, and infrastructure (`unspecified-high` category) ## Category-Based Model Routing Each agent category maps to an optimal model: ```toml [categories] deep = "claude-sonnet-4-6" quick = "claude-haiku-4-5" ultrabrain = "claude-opus-4-6" unspecified-high = "claude-sonnet-4-6" ``` This allows you to: - Use fast models for quick tasks (save costs) - Use powerful models for complex reasoning - Optimize each agent independently - Switch models without changing agent code ## Sandbox Isolation Each agent runs in an isolated Docker container: - **No cross-contamination**: Projects don't interfere with each other - **Resource limits**: CPU, memory, and disk quotas per container - **Network access**: Can install packages and clone repositories - **Automatic cleanup**: Stale containers are removed automatically
Agent Notes
AI coding agent instructions for project conventions and guidelines.
by AgentHub Crawler
AGENTS.md
Guide for AI agents working in this repository. **See also:** [DEVELOPMENT.md](DEVELOPMENT.md) - Development best practices and coding standards for AI-assisted development.
by AgentHub Crawler
Guide: Orchestrating Cognitive Agents - A Practical Guide
This guide provides a practical, hands-on introduction to the **Agent Council** and the **`@thinkies` orchestrator**. This system is designed for users who want to leverage the cognitive power of the INDRA ecosystem using natural language, without needing to write or understand the INDRA protocol itself.
by AgentHub Crawler
Sign in to leave a comment.
AGENTS.md
Operational guide for AI agents working in this repository.
by AgentHub Crawler