Skills: Teach Your Agent to Use Tools with Markdown Instructions

Skills are markdown instruction files that tell the agent when and how to use tools. Each skill lives in a directory containing a SKILL.md file with YAML frontmatter and a markdown body. OpenClaw loads bundled skills plus any local overrides, and filters them at load time based on environment, config, and binary presence.

• Creating skills, Build and test a custom skill from scratch.

• Skill Workshop, Review and approve agent-drafted skill proposals.

• Skills config, Full skills.* config schema and agent allowlists.

• ClawHub, Browse and install community skills.

Loading order

OpenClaw loads from these sources, highest precedence first. When the same skill name appears in multiple places, the highest source wins.

Skill roots support grouped layouts. OpenClaw discovers a skill whenever SKILL.md appears anywhere under a configured root (up to 6 levels deep):

<workspace>/skills/research/SKILL.md          ✓ found as "research"
<workspace>/skills/personal/research/SKILL.md ✓ also found as "research"

The folder path is for organization only. The skill's name and slash command come from the name frontmatter field (or the directory name when name is missing). Agent allowlists (below) also match on this name.

Note

Codex CLI's native $CODEX_HOME/skills directory is not an OpenClaw skill root. Use openclaw migrate plan codex to inventory those skills, then openclaw migrate codex to copy them into your OpenClaw workspace.

Node-hosted skills

A connected headless node can publish skills installed in its active OpenClaw skills directory (~/.openclaw/skills by default; profile environment overrides apply). They appear in the normal agent skill list while the node is connected and disappear when it disconnects. A local or Gateway skill keeps its name on collision; the node skill receives a deterministic node-prefixed name. Node-hosted v1 requires the directory name to match the skill's name frontmatter field.

The skill entry includes the node locator. Its files, relative references, and binaries live on the node, so load and execute it with exec host=node node=<node-id>. Restart the node host after changing its skill files. See Nodes for pairing and off-switches.

Per-agent vs shared skills

In multi-agent setups, each agent has its own workspace. Use the path that matches your desired visibility:

Agent allowlists

Skill location (precedence) and skill visibility (which agent can use it) are separate controls. Use allowlists to restrict which skills an agent sees, regardless of where they are loaded from.

{
  agents: {
    defaults: {
      skills: ["github", "weather"], // shared baseline
    },
    list: [
      { id: "writer" }, // inherits github, weather
      { id: "docs", skills: ["docs-search"] }, // replaces defaults entirely
      { id: "locked-down", skills: [] }, // no skills
    ],
  },
}

Allowlist rules

• Omit agents.defaults.skills to leave all skills unrestricted by default.
• Omit agents.entries.*.skills to inherit agents.defaults.skills.
• Set agents.entries.*.skills: [] to expose no skills for that agent.
• A non-empty agents.entries.*.skills list is the final set, it does not merge with defaults.
• The effective allowlist applies across prompt building, slash-command discovery, sandbox sync, and skill snapshots.
• This is not a host shell authorization boundary. If the same agent can use exec, constrain that shell separately with sandboxing, OS-user isolation, exec deny/allowlists, and per-resource credentials.

Plugins and skills

Plugins can ship their own skills by listing skills directories in openclaw.plugin.json (paths relative to the plugin root). Plugin skills load when the plugin is enabled, for example, the browser plugin ships a browser-automation skill for multi-step browser control.

Plugin skill directories merge at the same low-precedence level as skills.load.extraDirs, so a same-named bundled, managed, agent, or workspace skill overrides them. Gate a plugin skill's own eligibility via metadata.openclaw.requires in its frontmatter, same as any other skill.

See Plugins and Tools for the full plugin system.

Skill Workshop

Skill Workshop is a proposal queue between the agent and your active skill files. When the agent spots reusable work, it drafts a proposal instead of writing directly to SKILL.md. You review and approve before anything changes.

openclaw skills workshop list
openclaw skills workshop inspect <proposal-id>
openclaw skills workshop apply <proposal-id>

See Skill Workshop for the full lifecycle, CLI reference, and configuration.

Installing from ClawHub

ClawHub is the public skills registry. Use openclaw skills commands for install and update, or the clawhub CLI for publish and sync.

Install details

openclaw skills install installs into the active workspace skills/ directory by default. Add --global to install into the shared ~/.openclaw/skills directory, visible to all local agents unless agent allowlists narrow it.

Git and local installs expect SKILL.md at the source root. The slug comes from SKILL.md frontmatter name when valid, then falls back to the directory or repository name. Use --as <slug> to override. openclaw skills update tracks ClawHub installs only, reinstall Git or local sources to refresh them.

Verification and security scanning

openclaw skills verify @owner/<slug> asks ClawHub for the skill's clawhub.skill.verify.v1 trust envelope. Installed ClawHub skills verify against the version and registry recorded in .clawhub/origin.json. Bare slugs remain accepted for existing installed or unambiguous skills, but owner-qualified refs avoid publisher ambiguity.

ClawHub skill pages expose the latest security scan state before install, with detail pages for VirusTotal, ClawScan, and static analysis. The command exits non-zero when ClawHub marks verification as failed. Publishers recover false positives through the ClawHub dashboard or clawhub skill rescan @owner/<slug>.

Private archive installs

Gateway clients that need non-ClawHub delivery can stage a zip skill archive with skills.upload.begin, skills.upload.chunk, and skills.upload.commit, then install with skills.install({ source: "upload", ... }). This path is off by default and requires skills.install.allowUploadedArchives: true in openclaw.json. Normal ClawHub installs never need that setting.

Security

Warning

Treat third-party skills as untrusted code. Read them before enabling. Prefer sandboxed runs for untrusted inputs and risky tools. See Sandboxing for agent-side controls.

Path containment

Workspace, project-agent, and extra-dir skill discovery only accepts skill roots whose resolved realpath stays inside the configured root, unless skills.load.allowSymlinkTargets explicitly trusts a target root. Skill Workshop writes through those trusted targets only when skills.workshop.allowSymlinkTargetWrites is enabled. Managed ~/.openclaw/skills and personal ~/.agents/skills may contain symlinked skill folders, but every SKILL.md realpath must still stay inside its resolved skill directory.

Operator install policy

Configure security.installPolicy to run a trusted local policy command before skill installs continue. The policy receives metadata and the staged source path, applies to ClawHub, uploaded, Git, local, update, and dependency-installer paths, and fails closed when the command cannot return a valid decision.

Secret injection scope

skills.entries.*.env and skills.entries.*.apiKey inject secrets into the host process for that agent turn only, not into the sandbox. Keep secrets out of prompts and logs.

For the broader threat model and security checklists, see Security.

SKILL.md format

Every skill needs at minimum a name and description in the frontmatter:

---
name: image-lab
description: Generate or edit images via a provider-backed image workflow
---

When the user asks to generate an image, use the `image_generate` tool...

Note

OpenClaw follows the AgentSkills spec. Frontmatter is parsed as YAML first; if that fails, it falls back to a single-line-only parser. Nested metadata blocks (including multi-line YAML mappings) are flattened to a JSON string and re-parsed as JSON5, so the block form shown under Gating works. Use {baseDir} in the body to reference the skill folder path.

Optional frontmatter keys

• homepage (string), URL shown as "Website" in the macOS Skills UI. Also supported via metadata.openclaw.homepage.

• user-invocable (boolean, default: true), When true, the skill is exposed as a user-invocable slash command.

• disable-model-invocation (boolean, default: false), When true, OpenClaw keeps the skill's instructions out of the agent's normal prompt. The skill is still available as a slash command when user-invocable is also true.

• command-dispatch (tool), When set to tool, the slash command bypasses the model and dispatches directly to a registered tool.

• command-tool (string), Tool name to invoke when command-dispatch: tool is set.

• command-arg-mode (raw, default: raw), For tool dispatch, forwards the raw args string to the tool with no core parsing. The tool receives { command: "<raw args>", commandName: "<slash command>", skillName: "<skill name>" }.

Gating

OpenClaw filters skills at load time using metadata.openclaw (JSON5 object embedded in the frontmatter, see the parsing note above). A skill with no metadata.openclaw block is always eligible unless explicitly disabled.

---
name: image-lab
description: Generate or edit images via a provider-backed image workflow
metadata:
  {
    "openclaw":
      {
        "requires": { "bins": ["uv"], "env": ["GEMINI_API_KEY"], "config": ["browser.enabled"] },
        "primaryEnv": "GEMINI_API_KEY",
      },
  }
---

• always (boolean), When true, always include the skill and skip all other gates.

• emoji (string), Optional emoji shown in the macOS Skills UI.

• homepage (string), Optional URL shown as "Website" in the macOS Skills UI.

• os (("darwin" | "linux" | "win32")[]), Platform filter. When set, the skill is only eligible on a listed OS.

• requires.bins (string[]), Each binary must exist on PATH.

• requires.anyBins (string[]), At least one binary must exist on PATH.

• requires.env (string[]), Each env var must exist in the process or be provided via config.

• requires.config (string[]), Each openclaw.json path must be truthy.

• primaryEnv (string), Env var name associated with skills.entries.<name>.apiKey.

• install (object[]), Optional installer specs used by the macOS Skills UI (brew / node / go / uv / download).

Note

Legacy metadata.clawdbot blocks are still accepted when metadata.openclaw is absent, so older installed skills keep their dependency gates and installer hints. New skills should use metadata.openclaw.

Installer specs

Installer specs tell the macOS Skills UI how to install a dependency:

---
name: gemini
description: Use Gemini CLI for coding assistance and Google search lookups.
metadata:
  {
    "openclaw":
      {
        "emoji": "♊️",
        "requires": { "bins": ["gemini"] },
        "install":
          [
            {
              "id": "brew",
              "kind": "brew",
              "formula": "gemini-cli",
              "bins": ["gemini"],
              "label": "Install Gemini CLI (brew)",
            },
          ],
      },
  }
---

Installer selection rules

• When multiple installers are listed, the gateway picks one preferred option (brew when available, otherwise node).
• If all installers are download, OpenClaw lists each entry so you can see all available artifacts.
• Specs can include os: ["darwin"|"linux"|"win32"] to filter by platform.
• Node installs honor skills.install.nodeManager in openclaw.json (default: npm; options: npm / pnpm / yarn / bun). This only affects skill installs; the Gateway runtime should still be Node.
• Gateway installer preference: Homebrew → uv → configured node manager → go → download.

Per-installer details

• Homebrew: OpenClaw does not auto-install Homebrew or translate brew formulas into system package commands. In Linux containers without brew, brew-only installers are hidden; use a custom image or install the dependency manually.
• Go: OpenClaw requires Go 1.21 or newer for automatic skill installs. If go is missing and Homebrew is available, OpenClaw installs Go via Homebrew first; on Linux without Homebrew it can instead use apt-get as root or through passwordless sudo when the refreshed golang-go candidate meets the minimum version. The actual go install for the dependency always targets a dedicated OpenClaw-managed bin directory (Homebrew's bin on a fresh install, else ~/.local/bin) rather than your configured GOBIN, your own GOBIN, GOPATH, and GOTOOLCHAIN env vars are read but never overwritten.
• Download: url (required), archive (tar.gz | tar.bz2 | zip), extract (default: auto when archive detected), stripComponents, targetDir (default: ~/.openclaw/tools/<skillKey>).

Sandboxing notes

requires.bins is checked on the host at skill load time. If an agent runs in a sandbox, the binary must also exist inside the container. Install it via agents.defaults.sandbox.docker.setupCommand or a custom image. setupCommand runs once after container creation and requires network egress, a writable root FS, and a root user in the sandbox.

Config overrides

Toggle and configure bundled or managed skills under skills.entries in ~/.openclaw/openclaw.json:

{
  skills: {
    entries: {
      "image-lab": {
        enabled: true,
        apiKey: { source: "env", provider: "default", id: "GEMINI_API_KEY" },
        env: { GEMINI_API_KEY: "GEMINI_KEY_HERE" },
        config: {
          endpoint: "https://example.invalid",
          model: "nano-pro",
        },
      },
      peekaboo: { enabled: true },
      sag: { enabled: false },
    },
  },
}

• enabled (boolean), false disables the skill even when bundled or installed. The coding-agent bundled skill is opt-in, set skills.entries.coding-agent.enabled: true and ensure one of claude, codex, opencode, or another supported CLI is installed and authenticated.

• apiKey (string | { source, provider, id }), Convenience field for skills that declare metadata.openclaw.primaryEnv. Supports a plaintext string or a SecretRef object.

• env (true), "> Environment variables injected for the agent run. Only injected when the variable is not already set in the process.

• config (object), Optional bag for custom per-skill configuration fields.

• allowBundled (string[]), Optional allowlist for bundled skills only. When set, only bundled skills in the list are eligible. Managed and workspace skills are unaffected.

Note

Config keys match the skill name by default. If a skill defines metadata.openclaw.skillKey, use that key under skills.entries instead. Quote hyphenated names: JSON5 allows quoted keys.

Environment injection

When an agent run starts, OpenClaw:

Reads skill metadata

OpenClaw resolves the effective skill list for the agent, applying gating rules, allowlists, and config overrides.

Injects env and API keys

skills.entries.<key>.env and skills.entries.<key>.apiKey are applied to process.env for the duration of the run.

Builds the system prompt

Eligible skills are compiled into a compact XML block and injected into the system prompt.

Restores the environment

After the run ends, the original environment is restored.

Warning

Env injection is scoped to the host agent run, not the sandbox. Inside a sandbox, env and apiKey have no effect. See Skills config for how to pass secrets into sandboxed runs.

For the bundled claude-cli backend, OpenClaw also materializes the same eligible skill snapshot as a temporary Claude Code plugin and passes it via --plugin-dir. Other CLI backends use the prompt catalog only.

Snapshots and refresh

OpenClaw snapshots eligible skills when a session starts and reuses that list for all subsequent turns in the session. Changes to skills or config take effect on the next new session.

Skills refresh mid-session in two cases:

• The skills watcher detects a SKILL.md change.
• A new eligible remote node connects.

The refreshed list is picked up on the next agent turn. If the effective agent allowlist changes, OpenClaw refreshes the snapshot to keep visible skills aligned.

Skills watcher

By default, OpenClaw watches skill folders and bumps the snapshot when SKILL.md files change. Configure under skills.load:

{
  skills: {
    load: {
      extraDirs: ["~/Projects/agent-scripts/skills"],
      allowSymlinkTargets: ["~/Projects/manager/skills"],
      watch: true, // default
    },
  },
}

Watcher events use a built-in 250 ms debounce. Use allowSymlinkTargets for intentional symlinked layouts where a skill root symlink points outside the configured root, for example <workspace>/skills/manager -> ~/Projects/manager/skills. Enable skills.workshop.allowSymlinkTargetWrites only when Skill Workshop should also apply proposals through those trusted symlinked paths.

Remote macOS nodes (Linux gateway)

If the Gateway runs on Linux but a macOS node is connected with system.run allowed, OpenClaw can treat macOS-only skills as eligible when the required binaries are present on that node. The agent should run those skills via the exec tool with host=node.

Offline nodes do not expose remote-only skills. When a node stops responding to bin probes, OpenClaw removes its cached bin matches.

Token impact

When skills become eligible, OpenClaw inserts a compact XML block into the system prompt. The cost is deterministic and grows linearly with each skill:

• Base overhead (applies only when at least one skill is eligible): a fixed block of introductory text plus the <available_skills> wrapper.
• Per skill: roughly 97 characters plus the lengths of your name, description, and location fields.
• XML escaping converts & < > " ' into entities, adding a small number of characters per instance.
• At about 4 characters per token, 97 characters equals roughly 24 tokens per skill before field lengths.

If the resulting block would exceed the configured prompt budget (skills.limits.maxSkillsPromptChars), OpenClaw first keeps as many skill identities (name, location, and version) as the description-free compact format allows. It then uses any leftover budget for shortened descriptions. If no budget remains for descriptions, they are omitted. The prompt includes a note pointing at openclaw skills check whenever compact formatting or list truncation is required.

Keep descriptions short and descriptive to reduce prompt overhead.

Related

• Creating skills, Step-by-step instructions for writing a custom skill.

• Skill Workshop, Proposal queue for agent-generated skills.

• Skills config, Complete skills.* config schema and agent allowlists.

• Slash commands, How skill slash commands are registered and routed.

• ClawHub, Browse and publish skills on the public registry.

• Plugins, Plugins can bundle skills alongside the tools they document.