OpenClawDocs
View as MarkdownView this page as plain textOpen in ChatGPTAsk questions about this pageOpen in ClaudeAsk questions about this page

Agent workspace

Agent bootstrapping

Bootstrapping is the first-run ritual that seeds a new agent workspace and walks the agent through picking an identity. It runs once, right after onboarding, on the agent's first real turn.

What happens

On the first run against a brand-new workspace (default ~/.openclaw/workspace), OpenClaw:

  • Seeds AGENTS.md, SOUL.md, IDENTITY.md, USER.md, and BOOTSTRAP.md. Environment-specific tool notes belong in the ## Tools section of AGENTS.md.
  • Has the agent follow a short birth sequence: it asks what you want to call it, shares one short soul/vibe line, generates four avatar options when image_generate is available, asks whether you want the minimal recommended plugin set or maximum convenience.
  • Persists the agreed identity twice: into IDENTITY.md and SOUL.md (what the agent reads about itself) and via openclaw agents set-identity (what channels and the UI display).
  • Presents four generated avatars in a numbered 2×2 choice sheet for you to choose or skip, using the configured image-generation model or an available provider such as OpenAI. The selected portrait is cropped from the sheet, saved under the workspace's avatars/ directory, and synced into identity. Identity files are saved after this choice so an asynchronous generation does not end hatching early. If generation is unavailable or fails, hatching continues with the emoji.
  • Reads app recommendations already stored during onboarding without rescanning. Official plugins use openclaw plugins install <id>; third-party ClawHub skills remain explicit opt-ins. After the choice is handled, the agent acknowledges the stored offer so it never asks again.
  • Deletes BOOTSTRAP.md once the workspace looks configured, so the ritual only runs once.

A workspace counts as configured once SOUL.md, IDENTITY.md, or USER.md has diverged from its starter template, or a memory/ folder exists.

Embedded and local model runs

For embedded or local-model runs, OpenClaw keeps BOOTSTRAP.md out of the privileged system context. On the primary interactive first run it still passes the file contents through the user prompt, so models that don't reliably call the read tool can still complete the ritual. If the current run cannot safely access the workspace, the agent gets a short limited-bootstrap note instead of a generic greeting.

Skipping bootstrapping

To skip this on a pre-seeded workspace, run:

bash
openclaw onboard --skip-bootstrap

Where it runs

Bootstrapping always runs on the gateway host. If the macOS app connects to a remote Gateway, the workspace and its bootstrap files live on that remote machine, not on the Mac.

Was this useful?
On this page

On this page