Imported from raiph-ai/fireclaw (client/SKILL.md). Install upstream with npx skills add raiph-ai/fireclaw --skill client. Copyright stays with the author.

FireClaw Client Skill

The FireClaw client skill connects your main OpenClaw agent to a FireClaw proxy instance. It provides safe web browsing by routing all web fetches through the FireClaw sanitization pipeline.

What It Does

• Replaces web_fetch and web_search with fireclaw_fetch and fireclaw_search
• Routes requests to your FireClaw proxy (local sub-agent or remote instance)
• Shows FireClaw status and stats on demand
• Reminds the user at the start of each new thread if FireClaw is disabled
• Allows conversational control: "enable FireClaw", "disable FireClaw", "FireClaw status"

Installation

1. Copy the client/ folder into your OpenClaw skills directory
2. Configure the proxy connection in your OpenClaw config or client/config.yaml
3. Add to your agent's AGENTS.md: "Use fireclaw_fetch instead of web_fetch for all web browsing"

Configuration

fireclaw_client:
  enabled: true
  mode: "sub-agent"          # "sub-agent" or "remote"
  
  # Sub-agent mode
  sub_agent:
    model: "google/gemini-2.0-flash"
    tools_allow: ["web_fetch", "web_search"]
    tools_deny: ["exec", "message", "sessions_spawn", "gateway", "Write", "nodes", "canvas"]
  
  # Remote mode  
  remote:
    url: "http://192.168.1.x:8420"  # FireClaw proxy URL
    api_key: ""                       # Set via keychain
    timeout_ms: 30000
  
  # Behavior
  remind_when_disabled: true    # Remind at start of new threads
  reminder_message: "⚠️ FireClaw is currently disabled. Web browsing is unprotected against prompt injection. Say 'enable fireclaw' to re-enable."
  
  # Alerts
  alerts:
    enabled: true
    channel: null               # Use current channel
    severity_threshold: "medium"

Tools Provided

fireclaw_fetch(url, intent?)

Safely fetch a URL through the FireClaw proxy.

• url — The URL to fetch
• intent (optional) — What information you need (improves summarization accuracy)
• Returns: { content, metadata: { severity, detections, duration, cached, trustTier } }

fireclaw_search(query, count?)

Safely search the web through the FireClaw proxy.

• query — Search query
• count (optional) — Number of results (1-10)
• Returns: { content, metadata }

fireclaw_status()

Check FireClaw proxy status, connection health, and recent stats.

fireclaw_enable()

Enable the FireClaw proxy.

fireclaw_disable()

Disable the FireClaw proxy (will trigger reminders on new threads).

Thread Reminder

When FireClaw is disabled, the skill injects a reminder at the start of each new conversation thread:

⚠️ FireClaw is currently disabled. Web browsing is unprotected against prompt injection. Say "enable fireclaw" to re-enable.

This ensures the user is always aware when protection is off. The reminder is conversational — the user can re-enable with a simple command.