Configuring Darce: ~/.darcerc, skills and memory
Everything is optional. darce sets up ~/.darcerc on first run.
~/.darcerc
{
"mode": "auto",
"router": { "default": "anthropic/claude-haiku-5.5" },
"gears": ["qwen/qwen3-coder-next", "qwen/qwen3-coder", "deepseek/deepseek-v4-pro", "anthropic/claude-sonnet-5.5"],
"derbyModels": ["anthropic/claude-sonnet-5.5", "openai/gpt-5.6-sol", "google/gemini-3.1-pro-preview"],
"critic": false,
"why": true,
"passEnv": ["GH_TOKEN"]
}| Key | What it does |
|---|---|
mode | Default approval mode: auto, ask, plan or full. See Safety. |
router.default | The model Darce starts with. |
gears | The order of models for Shift+↑ / Shift+↓. See Models. |
derbyModels | The default lineup for /derby. See Swarm and derby. |
critic | Second-opinion review of every edit (same as /critic on). |
why | WHY notes at the end of meaningful changes. |
passEnv | Credential-like environment variables that commands may see. Everything matching *_KEY, *_TOKEN, *_SECRET, DATABASE_URL and similar is hidden otherwise. |
riskCheck | Set to false to turn off the classifier's second look at scripts. |
visionModel | Which model reads images when your main model can't. |
A repository's own .darcerc can't change your approval mode, endpoints or keys.
Project instructions
Darce reads your project's AGENTS.md, CLAUDE.md or .cursorrules, so existing instructions for other agents work as-is.
Skills
A skill is a folder with a SKILL.md. Darce loads it when a task matches its description:
---
name: deploy
description: How this team deploys the API. Use when asked to ship or release.
---
1. Run `make test`.
2. ...Darce looks in ~/.darce/skills/ and ~/.claude/skills/ first, then the project's .darce/skills/ and .claude/skills/, and installed Claude Code plugins. Your existing Claude Code skills work without changes, and your own skills take precedence over a repository's.
Built-in skills include ui-craft, security-review, teach and web-research. List them with /skills.
Memory
Darce remembers your preferences and project facts across sessions. /memory shows what it knows; /memory forget <text> removes something.
Suggestions
After each task, a small fast model suggests your likely next prompt as ghost text; Tab or → accepts it. On by default for paid plans; /suggest on|off to change.