OpenClaw Skills

OpenClaw Skills combine AgentSkills-compatible folders with workspace precedence, managed installs, ClawHub workflows, load-time gates, and explicit security guidance for untrusted skills.

OpenClaw Skills use AgentSkills-compatible folders to teach the agent how to use tools and follow workflows. Each skill is a directory with SKILL.md frontmatter and instructions.

OpenClaw adds platform-specific behavior around precedence, plugin participation, registry workflows, metadata gating, and environment injection.

OpenClaw loads bundled skills, managed or local skills, and workspace skills. Workspace skills have the highest precedence, followed by managed/local skills, then bundled skills.

That precedence is useful for team overrides, but it means two machines or workspaces can resolve the same skill name differently unless the active source is logged.

OpenClaw plugins can ship their own skills, and ClawHub provides a public registry for discovering, installing, updating, and syncing skills.

Registry-based installation changes the supply-chain conversation. Teams should track source, version, maintainer, and update cadence for every installed skill.

OpenClaw follows the AgentSkills layout intent but documents additional metadata options. Skills can be filtered by operating system, required binaries, environment variables, config values, and installer metadata.

The practical effect is that a skill can appear or disappear depending on host capabilities and configuration. Sandboxed runs may need the same binary installed inside the sandbox, not only on the host.

OpenClaw's own documentation says third-party skills should be treated as untrusted code. That is the right default for any skill that can influence tools, inject environment values, or point agents at scripts.

The platform also documents env and apiKey injection into the host process for an agent turn. Those values must not be copied into prompts, examples, transcripts, or skill output.

OpenClaw's flexible loading model should be paired with an inventory of skill locations, enabled config entries, injected environment variables, and registry sources.

For enterprise workflows, centralize audit evidence around which skill source won precedence, which env values were available, and what files or tools the skill caused the agent to touch.