Sigil reference
The reference the agent reads to answer questions about itself. Every file here is
plain markdown — browsable on GitHub, and loaded at runtime by Sigil's read_docs
tool, so what you read is exactly what the agent reads.
Ask Sigil in chat ("how does the workspace work?", "what models can I use?", "how do I schedule a job?") and it answers from these pages. Or read them directly:
| Topic | What it covers |
|---|---|
| overview | What Sigil is and how a request flows through it |
| skill-compilation | The compiler: SKILL.md → AG-IR → runnable agent, the gates, eject |
| configuration | Every configure key and what it does |
| chat-and-tools | Chat mode and the full tool-belt |
| workspace-and-tools | The workspace, the sandbox switch, the exec gate, and never-timed-out shell jobs |
| automation-and-cron | Scheduling work the agent runs later |
| ejecting-agents | Taking a compiled skill out: one-shot program, or a whole agent |
| daemon | Running headless: the background scheduler, one graph, start-at-login |
| channels | Connecting Discord / Telegram / WhatsApp / Slack |
| memory-and-skills | Durable memory and the compiled-skill library |
| models | Model tiers, aliases, fallbacks, and providers |
| claude-code | Running on your Claude Code CLI, and serving compiled skills to it — setup, costs, limits |
From the CLI: sigil docs lists topics, sigil docs <topic> prints one. In the chat
REPL: /docs and /docs <topic>.