platform

name	platform
description	Manage skills, browse curated recommendations, and use platform services. Use when: user asks to install/search/manage skills, view platform curated skills, or use platform-specific features.
metadata	{"openclaw":{"requires":{"bins":["curl"]}}}

Platform Services

This skill provides access to the multi-user platform's shared services. All API calls go through the local bridge at http://localhost:18080.

When to Use

• User asks to install a skill / search for skills
• User asks about recommended/curated skills
• User wants to submit a skill for platform review
• User asks about platform capabilities

1. Curated Skills (Platform Recommended)

The platform maintains a curated list of tested, recommended skills.

List curated skills

curl -s http://localhost:18080/api/curated-skills | python3 -m json.tool

Response is a JSON array. Each skill has:

• id — unique ID (needed for install)
• name — skill name
• description — what it does
• author — who made it
• category — category tag
• is_featured — whether it's a staff pick
• install_count — how popular it is
• installed — whether it's already installed in this container

Install a curated skill

curl -s -X POST http://localhost:18080/api/curated-skills/<SKILL_ID>/install | python3 -m json.tool

Replace <SKILL_ID> with the id from the list above.

Submit a skill for platform review

If the user has created or found a useful skill, submit it for admin review:

curl -s -X POST http://localhost:18080/api/curated-skills/submit \
  -H "Content-Type: application/json" \
  -d '{"skill_name": "my-skill", "description": "What it does", "source_url": "https://..."}' \
  | python3 -m json.tool

Check my submissions

curl -s http://localhost:18080/api/curated-skills/submissions/mine | python3 -m json.tool

2. Marketplace Skills (skills.sh)

Search and install community skills from the public marketplace.

Search

curl -s -X POST http://localhost:18080/api/marketplaces/skills/search \
  -H "Content-Type: application/json" \
  -d '{"query": "SEARCH_TERM", "limit": 10}' \
  | python3 -m json.tool

Results include slug, url, and installs count.

Install from marketplace

curl -s -X POST http://localhost:18080/api/marketplaces/skills/install \
  -H "Content-Type: application/json" \
  -d '{"slug": "owner/repo@skill-name"}' \
  | python3 -m json.tool

Or use the CLI directly:

npx --yes skills add owner/repo@skill-name -g -y --copy

3. Installed Skills

List installed skills

curl -s http://localhost:18080/api/skills | python3 -m json.tool

Check what's in the skills directory

ls ~/.openclaw/skills/

Enable/disable a skill

# Enable
curl -s -X PUT http://localhost:18080/api/skills/SKILL_NAME/toggle \
  -H "Content-Type: application/json" \
  -d '{"enabled": true}' | python3 -m json.tool

# Disable
curl -s -X PUT http://localhost:18080/api/skills/SKILL_NAME/toggle \
  -H "Content-Type: application/json" \
  -d '{"enabled": false}' | python3 -m json.tool

Agent-Specific Skills

Each agent can have its own skills. By default, skills are created for the "main" agent. Use the X-Agent-Id header to specify which agent's skills to manage.

Skill storage locations:

• Main agent: ~/.openclaw/workspace/skills/
• Other agents: ~/.openclaw/workspace-<agentId>/skills/

For example, skills for "coder" agent are stored in ~/.openclaw/workspace-coder/skills/.

List skills for a specific agent

# List skills for "coder" agent (use X-Agent-Id header)
curl -s "http://localhost:18080/api/skills" \
  -H "X-Agent-Id: coder" | python3 -m json.tool

# List skills for "main" agent (default)
curl -s "http://localhost:18080/api/skills" | python3 -m json.tool

Upload skill to a specific agent

# Upload skill to "coder" agent
curl -s -X POST "http://localhost:18080/api/skills/upload" \
  -H "X-Agent-Id: coder" \
  -F "file=@my-skill.zip" | python3 -m json.tool

Delete skill from a specific agent

# Delete skill from "coder" agent
curl -s -X DELETE "http://localhost:18080/api/skills/my-skill" \
  -H "X-Agent-Id: coder" | python3 -m json.tool

Download skill from a specific agent

# Download skill from "coder" agent
curl -s "http://localhost:18080/api/skills/my-skill/download" \
  -H "X-Agent-Id: coder" -o my-skill.zip

Important: Always use the X-Agent-Id header to specify which agent's skills to operate on. This ensures each agent can only manage its own skills. If not specified, defaults to "main".

4. ClawHub Skills

ClawHub (clawhub.com) is another community skill registry. The clawhub skill (built-in from OpenClaw) teaches you how to use its CLI. If clawhub binary is not installed yet:

npm i -g clawhub

Then search and install:

clawhub search "postgres backups"
clawhub install my-skill

See the clawhub skill for full CLI usage.

Workflow

When a user asks to install or find skills, search all sources in this order:

1. Platform curated skills — check first, these are platform-tested and recommended
2. ClawHub — community registry, search via clawhub search
3. Marketplace (skills.sh) — broad community skills, search via API or npx skills find
4. After install — tell the user the skill is ready to use (it takes effect immediately, no restart needed)

If the user has a specific skill name, check curated list first. If they describe a need ("I need something for web scraping"), search all three sources and present the best options together.

When presenting results, show them in a clean table format with name, description, source, and status.

5. Agent Management

You can create, view, update, and delete AI Agents. Each agent has its own workspace and can be configured with different models.

List all agents

curl -s http://localhost:18080/api/agents | python3 -m json.tool

Response includes agent details: id, name, identity, workspace, etc.

Create a new agent

curl -s -X POST http://localhost:18080/api/agents \
  -H "Content-Type: application/json" \
  -d '{"name": "my-agent", "workspace": "~/.openclaw/workspace-myagent", "model": "claude-sonnet-4-5"}' \
  | python3 -m json.tool

Parameters:

• name (required): Agent ID, lowercase letters, numbers, underscores and hyphens only
• workspace (optional): Custom workspace path, defaults to ~/.openclaw/workspace-<name>
• model (optional): Model to use, e.g., claude-sonnet-4-5, gpt-4o, qwen-max

Update an agent

curl -s -X PUT http://localhost:18080/api/agents/AGENT_ID \
  -H "Content-Type: application/json" \
  -d '{"name": "new-name", "model": "gpt-4o"}' \
  | python3 -m json.tool

⚠️ Delete an Agent (Requires User Confirmation)

Deleting an agent is IRREVERSIBLE! Before calling the delete API, you MUST:

1. Ask the user for confirmation first in their language
2. Explain what will be deleted (the agent, its workspace, all conversation history)
3. Wait for explicit confirmation ("yes", "confirm", "是", "确认", etc.)
4. Never delete without user consent

Example workflow (Chinese user):

你：确定要删除 "insurance-agent" 吗？这将同时删除其工作区文件和所有对话历史，此操作无法撤销。

用户：是的，删除吧

你：（然后调用删除 API）

Example workflow (English user):

You: Are you sure you want to delete "insurance-agent"? This will also delete all its workspace files and conversation history. This action cannot be undone.

User: yes, delete it

You: (then call the delete API)

Delete command:

# Delete agent (keep workspace files)
curl -s -X DELETE "http://localhost:18080/api/agents/AGENT_ID" | python3 -m json.tool

# Delete agent AND workspace files
curl -s -X DELETE "http://localhost:18080/api/agents/AGENT_ID?delete_files=true" | python3 -m json.tool

Get available models

curl -s http://localhost:18080/api/models | python3 -m json.tool

This returns available models from the platform with their providers.