Community Tips

This page collects practical tips from the community. It is not a strict spec. When using third‑party skills/plugins, read their docs and SKILL.md, and assess security risk before enabling them.

If you want the underlying concepts first:

Skills: /en/docs/agent/skills/
Plugins: /en/docs/tools/plugin/
Security boundary: /en/docs/gateway/security/

1) Skills: install and iterate fast

Via CLI (if your version supports it)

openclaw skills search "code review"
openclaw skills install clawhub/code-review
openclaw skills list

Manual install (portable)

Unzip into ~/.openclaw/skills/ (global) or <workspace>/skills/ (project-level, higher priority):

mkdir -p ~/.openclaw/skills
unzip my-skill.zip -d ~/.openclaw/skills/my-skill
ls ~/.openclaw/skills/my-skill/SKILL.md

Then restart the Gateway:

openclaw gateway restart

Security note: Skills influence the agent’s behavior and capability surface. Only install skills you trust, and read SKILL.md before enabling.

2) MCP via skills: add search/vision capabilities

When your chosen model/provider lacks certain integrations (search, vision, etc.), a common community approach is to install an MCP-related skill that provides those tools.

Example:

openclaw skills search "mcp"
openclaw skills install clawhub/coding-plan-mcp
openclaw gateway restart

3) DingTalk: community plugins

DingTalk (钉钉) integrations are typically shipped as community plugins (not built into core). Prefer actively maintained plugins with clear docs, and keep a rollback plan.

Entry points:

Plugins: /en/docs/tools/plugin/
CLI plugins: /en/docs/cli/plugins/
Channel troubleshooting: /en/docs/channels/troubleshooting/

4) Multi-agent setups: isolate roles and risk

Instead of putting every workflow into one main session, split roles into multiple agents with separate workspaces/sessions:

Work agent: coding + command execution + work channels
Personal agent: personal chats + scheduling
Sandbox agent: test new skills/config safely first

See:

Debugging