Claude Code
How Claude Code works inside coSPEC sandboxes
What is Claude Code?
Claude Code is Anthropic's CLI coding agent. coSPEC runs Claude Code inside sandboxes as the execution engine for every run.
Connecting Your Anthropic Key
Go to Integrations → Claude Code in the dashboard. Enter your Anthropic API key and click Connect.
coSPEC stores your key securely and injects it into the sandbox at runtime. The agent uses your key to call Claude models directly.
How It Works
When you create a run, coSPEC:
- Provisions a sandbox with your template configuration
- Clones the repository
- Launches Claude Code with your prompt
- Records events (messages, tool calls, file edits) as the agent works
Models
Claude Code uses Anthropic's Claude models. Set the model via the model field when creating a run.
| Model | Description |
|---|---|
sonnet | Default. Fast and capable — good for most tasks. |
opus | Most capable. Best for complex, multi-step work. |
haiku | Fastest and cheapest. Good for simple, focused tasks. |
MCP Support
Claude Code discovers MCP servers from the .mcp.json file in your repository root. Any servers defined there will be available to the agent during execution.
Skills
Claude Code picks up skills from the .claude/ directory in your repository. Define custom slash commands and workflows that the agent can invoke during a run.
Project Instructions
Claude Code automatically reads CLAUDE.md files in your repository. Use this to provide project-specific context:
- Coding conventions and style guides
- Architecture decisions
- Build and test commands
- Anything the agent should know before working on your code