Skills learning hub · Independent guidance

OpenClaw skills: install, trust, and operate them safely

An OpenClaw skill is instruction and resource content loaded into an agent context. Treat it as executable guidance: inspect the source, declared dependencies, triggers, external services, privileges, and update path before installation. Use official ClawHub discovery rather than a mirrored directory.

Current snapshotUpdated 2026-08-24Official docs main snapshot · OpenClaw 2026.8.1
01

A practical trust sequence

  1. 1

    Search without installing

    Find the owner-qualified package and compare alternatives.

    openclaw skills search "calendar"
  2. 2

    Verify provenance

    Review the registry decision and pinned source when available.

    openclaw skills verify @owner/skill --json
  3. 3

    Install deliberately

    Prefer a version pin for a reproducible workflow.

    openclaw skills install @owner/skill --version <version>
  4. 4

    Check readiness

    Confirm binaries, environment variables, config, and visibility.

    openclaw skills check --json
02

Do not confuse scanning with trust

A registry scan is one signal, not proof of safety. Owner identity, exact version, source commit, local changes, install policy, dependency behavior, and the privileges available to the agent all matter. Avoid broad trigger descriptions and undeclared write operations.

Common questions

Before you act

Is this site a skills directory?

No. Official ClawHub owns discovery. This site focuses on understanding, creating, verifying, and operating skills safely.

Can `--force` bypass a block?

No. Current OpenClaw policy distinguishes warnings from blocks; force flags do not override a blocking install policy result.

Primary sources

Verify against the owner.

Content snapshot follows official docs main at the recorded commit; the current package metadata reported 2026.8.1 when verified. Your installed release and live CLI schema remain authoritative.