Security

SECURITY.md

Security Decisions — OpenClaw Dev Workspace

Gateway

Bind: loopback only — gateway is not exposed to the network.
Port: 18789 (default) — accessible only from localhost.
Mode: local — no remote connections accepted.

Filesystem Scope

Workspace: Configured via WORKSPACE_ROOT environment variable — all agent operations are restricted to this directory.
Permissions: All workspace directories are 700 (owner-only read/write/execute).
Config file: ~/.openclaw/openclaw.json is 600 (owner-only read/write).

Skills

Local skills only: Skills are loaded from $WORKSPACE_ROOT/skills via skills.load.extraDirs.
No remote registry: ClawHub and remote skill registries are not configured. No clawhub install commands should be run.
Bundled skills: Allowed (allowBundled: true) as these ship with the verified npm package.

Channels

No channels connected by default. The onboarding wizard was configured with minimal channel access.
DM policy: When channels are added, use "dmPolicy": "pairing" (require pairing code approval).
No browser control: Not enabled.

Credentials

No production credentials are stored in this workspace or config.
No long-lived tokens beyond what OpenClaw's daemon requires for local gateway operation.
API keys should be set via environment variables, not stored in openclaw.json.

Ongoing

Run openclaw security audit --deep periodically.
Run openclaw doctor after any config changes.
Review this file when adding channels or tools.

There aren't any published security advisories