OWonlyWeb

Knowledge

OpenClaw Configuration and Permissions

skills/openclaw-configuration-and-permissions.md

title: OpenClaw Configuration and Permissions category: skills tags: [openclaw, config, permissions, gateway] aliases: [OpenClaw Config and Permissions] relationships:

type: implements

type: related_to sources: [_raw/openclaw/p234-docs-infra-archive-2026-05-25/] summary: A practical map of the main OpenClaw configuration surfaces that control tools, command execution, gateway exposure, browser behavior, and approvals. provenance: extracted: 0.82 inferred: 0.16 ambiguous: 0.02 base_confidence: 0.84 lifecycle: draft lifecycle_changed: 2026-05-25 tier: supporting created: 2026-05-25T08:50:00Z updated: 2026-05-25T08:50:00Z

OpenClaw Configuration and Permissions

OpenClaw command and tool behavior is governed by several layers at once: openclaw.json, the gateway surface, command-execution policy, approvals, sandbox behavior, and host operating-system permissions.

Key Ideas

  • Tool access and command execution are separate concerns; a channel may be allowed to invoke a skill while the underlying exec path is still constrained.
  • tools.profile, tools.allow, tools.deny, and tools.exec.* shape the tool baseline.
  • commands.native and commands.nativeSkills decide whether commands run via native execution or remain constrained.
  • gateway.nodes.denyCommands can still block high-risk operations even if other settings are permissive.
  • Gateway, WebUI, browser, and terminal are different capability paths and should not be conflated.
  • .env affects network behavior, but not all operational permissions live there.

Most Important Surfaces

  • ~/.openclaw/openclaw.json for runtime structure and capability boundaries.
  • ~/.openclaw/.env for proxy and environment-variable injection.
  • LaunchAgent/runtime environment for whether those variables actually reach the gateway.
  • Exec approvals and host permissions for final command success on macOS.

Practical Guidance

  • When troubleshooting “can it run commands?”, inspect tools, commands, deny lists, approvals, sandbox, and host permissions together.
  • Treat skills: ["*"] on a chat channel as widening the callable surface, not as unconditional host access.
  • Prefer explicit, auditable settings over assuming a permissive default.

Related

Sources