Aethic Claw: Skills, Advanced Settings & FAQs

Detailed guide for Aethir Claw skills, advanced settings and FAQs

What's covered in this guide

🛠️ Section 1 — Skills	⚙️ Section 2 — Advanced Configurations	❓ Section 3 — FAQ
Installing, managing and understanding all 14 built-in skills	Multi-model routing, local models, and complete config JSON examples	Answers to the most common troubleshooting questions

---

1. Skills

---

1.1 Installing Additional Skills

If the built-in skills don't cover your needs, you can install more from ClawHub using either of these methods:

Method	How
Method 1 — Natural language	Use the social media bot or the OpenClaw Dashboard to ask the agent to install a skill in plain English.
Method 2 — Command line	Run: clawhub install <skill>

---

1.2 Log In to ClawHub

ClawHub is OpenClaw's official public skill registry and store. Logging in enables faster installs and upgrades — without it, you may hit permission errors.

ℹ NOTE — ClawHub login requires a Remote Desktop connection to the cloud server. Refer to Section 2.2 of the Aethir Claw Setup Guide for setup steps.

Login procedure:

1. Connect to your cloud server via Remote Desktop (see Section 2.2 of the Aethir Claw Setup Guide).

2. Open the command-line window and run:

   clawhub login

3. A browser window will open. Click "Sign in with GitHub" in the top-right corner.

4. Log in using your GitHub, Google, or Apple account credentials.

5. Once logged in, you will land on the ClawHub homepage.

6. Search for the skill you want to install.

Key ClawHub commands:

Action	Command
Install a skill	clawhub install <skill>
Update all skills	clawhub update --all
Uninstall a skill	clawhub uninstall <skill>

✔ TIP — Example: install the Notion skill at https://clawhub.ai/steipete/notion

clawhub install notion

---

1.3 Built-in Skills Reference

The following 14 skills come pre-installed on every Aethir Claw cloud server. Enable them all at once with:

sh /preclaw/pre_skills.sh

Quick overview:

#	Skill	What it does
1	skill-vetter	Audits ClawHub skills for safety and generates risk reports
2	Agent Browser	Browser automation CLI for AI agents — uses accessibility tree (93% less context)
3	Tavily Web Search	Real-time web search to keep agents up-to-date
4	find-skills	Lets the agent search and install skills on ClawHub autonomously
5	weather	Current weather and forecasts
6	self-improving-agent	Self-reflection, self-criticism, and continuous learning loop
7	summarize	Summarise URLs or files (web, PDF, image, audio, YouTube)
8	Proactive Agent	Turns the agent from reactive to proactive — anticipates needs
9	gog	Full Google suite integration (Gmail, Calendar, Drive, Docs)
10	Clawsec	MITM proxy to monitor and audit AI agent traffic in real time
11	Multi Search Engine	Multi-engine search with advanced operators, time filters & WolframAlpha
12	ontology	Typed knowledge graph for structured agent memory
13	GitHub	Interact with GitHub via gh CLI (issues, PRs, CI, API)
14	office-automation-test	Generate daily/weekly reports, meeting minutes, and office automation

---

1.3.1 skill-vetter

Purpose

Audits ClawHub skills for safety. Produces a risk rating and a detailed audit report.

Example Use Cases

1. Review the summarize skill on ClawHub for safety: give me a risk rating and an audit report.

2. Check the safety of this skill: https://clawhub.ai/ivangdavila/self-improving

---

1.3.2 Agent Browser

Purpose

A browser automation command-line tool built for AI agents. Uses the accessibility tree instead of the DOM to describe pages, saving 93% of context window when an agent controls the browser.

Prerequisites

npx playwright install  # Install the Playwright plugin

Example Use Cases

1. Open a web page:

   agent-browser open https://github.com/vercel-labs/agent-browser

2. Get a page snapshot (interactive elements):

   agent-browser snapshot -i

3. Click an element by ref:

   agent-browser click @e1

4. Fill an input field:

   agent-browser fill @e2 "text"

5. Close the browser:

   agent-browser close

---

1.3.3 Tavily Web Search

Purpose

OpenClaw's "real-time information brain." An online search skill that lets agents look up the latest news and data, avoiding hallucinated answers and solving the information-lag problem.

Setup — configure the API key:

ℹ NOTE — Apply for a Tavily API key at: https://app.tavily.com/home

# Open the .env file and add your key:
vim ~/.openclaw/.env
TAVILY_API_KEY=tvly-xxxxxxxxxxxxxxxxxxxxxxxx

# Apply the change and restart the gateway:
source ~/.openclaw/.env
openclaw gateway

Resulting entry in ~/.openclaw/OpenClaw.json:

"tools": {
  "profile": "full",
  "web": {
    "search": {
      "enabled": true
    }
  }
}

Example Use Cases

1. Specify result count: Search for OpenClaw installation steps — return up to 5 results.

2. Brief answer: Search for Tavily's free quota and include a short summary in the results.

3. Specify format: Search for solutions to the xdg-open command on Linux and return results in JSON format.

4. Combined: Search for AI agent trends in 2026, max 4 results, with a brief answer and official links.

---

1.3.4 find-skills

Purpose

Lets the AI Agent autonomously search ClawHub and install the skills it needs — no manual lookup required.

Example Use Cases

1. Find a skill that can organise my downloads folder, and install it directly once found.

---

1.3.5 weather

Purpose

Retrieves current weather conditions and forecasts for any location.

Example Use Cases

1. Use the weather skill to check tomorrow's weather in Beijing.

2. Query today's temperature and humidity in Beijing.

---

1.3.6 Self-Improving Agent

Purpose

Self-reflection + self-criticism + self-learning + self-organising memory. The agent evaluates its own outputs, identifies errors, and continuously improves over time.

Example Use Cases

1. I just finished a Python coding task — do a self-reflection and identify areas for optimisation.

2. Self-critique the copywriting result just generated, point out where it doesn't meet my requirements, and summarise improvement rules.

---

1.3.7 summarize

Purpose

Uses the summarize CLI to summarise URLs or files — web pages, PDFs, images, audio, and YouTube videos. Requires the OCR skill for image/scanned-document support.

Example Use Cases

1. Summarise the following content in one sentence, concise and clear: [paste text here]

2. Summarise our recent conversation and organise the key action items.

3. Use tavily-search to find the latest news about Python AI frameworks in 2026, then use summarize to extract the key points.

---

1.3.8 Proactive Agent

Purpose

Transforms the agent from a reactive task executor into a proactive partner that anticipates needs, sends automatic reminders, and continuously improves its own behaviour.

Example Use Cases

1. Enable proactive agent mode to activate automatic reminders and proactive optimisation features.

---

1.3.9 gog

Purpose

Full Google suite integration — Gmail, Google Calendar, Google Drive, and Google Docs.

ℹ NOTE — Requires OAuth authentication on the Google Cloud Console platform before use.

Reference Documentation

https://clawhub.ai/steipete/gog

---

1.3.10 Clawsec

Purpose

Manages ClawSec Monitor v3.0 — a MITM HTTP/HTTPS proxy that records AI agent traffic and detects data leakage and injection threats in real time.

Example Use Cases

1. Scan a ClawHub skill:

   clawsec audit github

2. Scan a local skill directory:

   clawsec audit /path/to/skill/dir

Reference Documentation

https://clawhub.ai/chrisochrisochriso-cmyk/clawsec

---

1.3.11 Multi-Search Engine

Purpose

A multi-engine search skill supporting advanced search operators, time filtering, site-specific search, privacy-preserving engines, and WolframAlpha knowledge base queries.

Example Use Cases

1. Use multi-search-engine to simultaneously search 'AI regulatory policy' on Google and Bing, then compile and compare the results.

Reference Documentation

https://clawhub.ai/gpyAngyoujun/multi-search-engine

---

1.3.12 ontology

Purpose

A typed knowledge graph for structured agent memory and composable skills. Create and query entities (people, projects, tasks, events, documents) and the relationships between them.

Example Use Cases

1. Organise the module dependency graph for a complex project.

Reference Documentation

https://clawhub.ai/oswalpalash/ontology

---

1.3.13 GitHub

Purpose

Interact with GitHub using the gh CLI — handle issues, submit PRs, run CI pipelines, and perform advanced API queries.

Authentication setup:

# Install gh:
sudo apt install gh   # Ubuntu/Debian
brew install gh       # macOS

# Login (interactive):
gh auth login

# Token authentication:
# Go to GitHub → Settings → Developer settings → Personal access tokens
# Create a Fine-grained token with required permissions
# (repo, issues, pull_requests, actions)

gh auth login --with-token < your-github-token-file

# Or paste directly:
echo "your-token" | gh auth login --with-token

Example use cases — natural language to gh command:

Category	Natural Language Command	Equivalent gh Command
PR Management	Check the CI status of PR 55 in owner/repo	gh pr checks 55 --repo owner/repo
Workflow Monitoring	List the 10 most recent GitHub Actions runs in owner/repo	gh run list --repo owner/repo --limit 10
Failure Analysis	View failed step logs for run ID xxx in owner/repo	gh run view xxx --repo owner/repo --log-failed
Issue Management	List all issue titles in owner/repo in JSON format	gh issue list --repo owner/repo --json title
Advanced API	Get the title, status, and submitter of PR 55	gh api repos/owner/repo/pulls/55 --jq '.title,.state,.user.login'

Reference Documentation

https://clawhub.ai/steipete/github

---

1.3.14 office-automation-test

Purpose

Quickly generates standardised daily reports, weekly reports, and meeting minutes. Supports custom formatting based on key inputs, and integrates office APIs for broader automation.

Capabilities:

Capability	Description
Calendar management	Create meetings, set reminders
Email handling	Categorise, organise, and auto-reply
Document editing	Generate reports, adjust formatting
Data processing	Excel analysis, chart generation

ℹ NOTE — Some capabilities require a corresponding API key for the integrated service.

Reference Documentation

https://clawhub.ai/earlyyuokubgw25-coder/office-automation-test

---

2. Multi-Model Routing & Advanced Configuration

This section covers how to configure multi-model routing in openclaw.json to optimise cost and performance. All configuration lives in ~/.openclaw/openclaw.json.

⚠ IMPORTANT — This is advanced usage. Familiarity with JSON and LLM model identifiers is assumed. Misconfigured files can prevent OpenClaw from starting — always keep a backup copy before editing.

---

2.1 openclaw.json — Configuration Reference

The table below explains the key fields in the agents.defaults block, which controls global model behaviour.

Field	Description
system	String. The system prompt injected into every session. Defines the agent's global behaviour, collaboration process, and output style.
model.primary	String. The main model to use, e.g. "anthropic/claude-opus-4-5". This model handles all requests by default.
model.fallbacks	Array. Ordered list of backup models used if the primary is unavailable or rate-limited.
models	Object. Maps model identifiers to short aliases usable with the /model command, e.g. { "anthropic/claude-opus-4-5": { alias: "opus" } }.
heartbeat.every	String. Interval for background keep-alive pings, e.g. "30m". Uses a cheap model to avoid token waste.
heartbeat.model	String. Model used for heartbeat pings. Use a low-cost model such as "google/gemini-2.5-flash-lite".
subagents.model	String. Model used by sub-agents for background tasks (search, extract, calculate). Use a cost-effective model.
subagents.maxConcurrent	Integer. Maximum number of sub-agents running simultaneously. Higher = faster but more token cost.
subagents.archiveAfterMinutes	Integer. Minutes of inactivity before a sub-agent session is archived.
subagents.system	String. System prompt for sub-agents — keep them focused on execution, not commentary.
imageModel.primary	String. Model used for vision/image tasks. Recommended: "google/gemini-3-flash".
contextTokens	Integer. Maximum context window size in tokens. Set based on your chosen model's limit, e.g. 200000.

Annotated configuration snippet:

// ~/.openclaw/openclaw.json
{
  "agents": {
    "defaults": {

      // Global system prompt — defines agent behaviour and output rules
      "system": "You are a top AI expert with multi-model collaboration capabilities...",

      // Primary model with ordered fallback chain
      "model": {
        "primary": "anthropic/claude-opus-4-5",
        "fallbacks": [
          "openai/gpt-5.2",
          "deepseek/deepseek-reasoner",
          "google/gemini-3-flash"
        ]
      },

      // Short aliases for /model command
      "models": {
        "anthropic/claude-opus-4-5":  { "alias": "opus"   },
        "anthropic/claude-sonnet-4-5":{ "alias": "sonnet" },
        "google/gemini-3-flash":      { "alias": "flash"  },
        "deepseek/deepseek-chat":     { "alias": "ds"     }
      },

      // Cheap model for background heartbeat pings every 30 minutes
      "heartbeat": {
        "every": "30m",
        "model": "google/gemini-2.5-flash-lite",
        "target": "last"
      },

      // Sub-agents use a cost-effective model for background tasks
      "subagents": {
        "model": "deepseek/deepseek-reasoner",
        "maxConcurrent": 1,          // One task at a time — balances cost & speed
        "archiveAfterMinutes": 60,
        "system": "You are an efficient execution unit. Focus on the task..."
      },

      // Vision tasks (optional)
      "imageModel": {
        "primary": "google/gemini-3-flash",
        "fallbacks": ["openai/gpt-5.2"]
      },

      "contextTokens": 200000        // Maximum context token count
    }
  }
}

---

2.2 Mixing Local and Online Models

You can route expensive tasks to capable cloud models while using a free local model (e.g. Ollama) for sub-agent background work. This significantly reduces API costs.

{
  "agents": {
    "defaults": {
      "model": {
        "primary":   "anthropic/claude-3-5-sonnet", // Smart cloud model for main tasks
        "fallbacks": ["openai/gpt-4o"]               // Cloud fallback
      },
      "subagents": {
        "model":               "local/llama-3:8b",  // Free local model for background tasks
        "maxConcurrent":       2,                   // More concurrency — no API cost
        "archiveAfterMinutes": 30
      },
      "contextTokens": 128000
    }
  }
}

✔ TIP — Running a local model via Ollama on your Aethir Claw server means sub-agent calls are completely free. Set maxConcurrent higher (e.g. 3–4) to parallelise background work without worrying about token costs.

---

2.3 Complete openclaw.json Example

A full, working configuration file showing all major sections. Replace placeholder values (marked xxxxxxxx) with your actual credentials.

{
  "meta": {
    "lastTouchedVersion": "2026.3.8",
    "lastTouchedAt": "2026-03-11T10:57:01.076Z"
  },
  "wizard": {
    "lastRunAt": "2026-03-11T09:58:01.176Z",
    "lastRunVersion": "2026.3.8",
    "lastRunCommand": "doctor",
    "lastRunMode": "local"
  },
  "browser": {
    "enabled": true,
    "executablePath": "/usr/bin/chromium",
    "headless": true,
    "noSandbox": true,
    "attachOnly": false,
    "defaultProfile": "openclaw"
  },
  "auth": {
    "profiles": {
      "openrouter:default": { "provider": "openrouter", "mode": "api_key" }
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "kimi-k2-5": {
        "baseUrl": "https://integrate.api.nvidia.com/v1",
        "apiKey":  "nvapi-xxxxxxxx",
        "api":     "openai-completions",
        "models": [{
          "id":            "moonshotai/kimi-k2.5",
          "name":          "moonshotai/kimi-k2.5 (Custom Provider)",
          "reasoning":     false,
          "input":         ["text"],
          "cost":          { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
          "contextWindow": 16000,
          "maxTokens":     4096
        }]
      },
      "openrouter": {
        "baseUrl": "https://openrouter.ai/api/v1",
        "apiKey":  "sk-or-v1-xxxxxxxxxxxxxxxx",
        "api":     "openai-completions",
        "models":  [{ "id": "openrouter/auto", "name": "OpenRouter Auto", "input": ["text","image"] }]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary":   "kimi-k2-5/moonshotai/kimi-k2.5",
        "fallbacks": ["openrouter/openrouter/auto"]
      },
      "models": {
        "openrouter/openrouter/auto":      { "alias": "OpenRouter" },
        "kimi-k2-5/moonshotai/kimi-k2.5": { "alias": "nv.k2.5"   }
      },
      "workspace":  "/config/.openclaw/workspace",
      "compaction": { "mode": "safeguard" },
      "heartbeat":  { "every": "30m", "model": "openrouter/openrouter/auto", "target": "last" },
      "maxConcurrent": 4,
      "subagents": {
        "maxConcurrent":       1,
        "archiveAfterMinutes": 60,
        "model":               "kimi-k2-5/moonshotai/kimi-k2.5"
      }
    }
  },
  "tools":    { "profile": "coding" },
  "messages": { "ackReactionScope": "group-mentions" },
  "commands": { "native": "auto", "nativeSkills": "auto", "restart": true, "ownerDisplay": "raw" },
  "session":  { "dmScope": "per-channel-peer" },
  "channels": {
    "whatsapp": {
      "enabled":      true,
      "dmPolicy":     "allowlist",
      "selfChatMode": true,
      "allowFrom":    ["xxxxxxx"],   // Your phone number
      "groupPolicy":  "allowlist",
      "debounceMs":   0,
      "mediaMaxMb":   50
    }
  },
  "gateway": {
    "port": 18789,
    "mode": "local",
    "bind": "loopback",
    "auth": { "mode": "token", "token": "xxxxxxxxxxxxxxxxxxxxxxxxx" },
    "tailscale": { "mode": "off", "resetOnExit": false },
    "nodes": {
      "denyCommands": [
        "camera.snap", "camera.clip", "screen.record",
        "contacts.add", "calendar.add", "reminders.add", "sms.send"
      ]
    }
  },
  "skills":  { "entries": { "tavily":   { "enabled": true } } },
  "plugins": { "entries": { "whatsapp": { "enabled": true } } }
}

---

3. Frequently Asked Questions

This section covers the most common issues users encounter. If your problem is not listed here, contact Aethir Cloud support via email.

---

Q: Why can't I SSH into the cloud server even after configuring the SSH key?

1. Verify the SSH command syntax: confirm that -p PORT abc@PUBLICIP is correct. Check the Public IP and Port under the Network tab in My Instances.

2. Confirm the public key has been added to the cloud server under the SSH Key tab in My Instances.

3. Check that the private key filename after -i in your SSH command matches the public key configured on the server.

4. Make sure you run the SSH command from the directory where the private key file is stored. Use cd to navigate there first — then you can specify just the filename without the full path.

5. If the error mentions known_hosts, Host key has changed, or Host key verification failed:

   Open the known_hosts file in your .ssh directory with a text editor, find and delete all lines containing the IP address you're connecting to, then try again.

⚠ IMPORTANT — Example error:

Add correct host key in C:\Users\Administrator/.ssh/known_hosts
Offending ECDSA key in C:\Users\Administrator/.ssh/known_hosts:10
Host key for [128.14.78.5]:10231 has changed and you have requested strict checking.
Host key verification failed.

Fix: delete the offending line from known_hosts and retry.

---

Q: Why does the Telegram bot not respond after I configure it?

1. Check whether the OpenClaw Gateway is running. Connect via SSH and restart it:

   # Find and stop any existing gateway process:
   ps -ef | grep openclaw-gateway
   kill -9 <PID>
   
   # Start gateway in background (nohup keeps it running after terminal closes):
   nohup openclaw gateway > nohup.out 2>&1 &
   
   # Press Control+C to exit — the gateway continues running.

2. If the bot still doesn't respond, the Telegram token may be incorrect. Reconfigure it following the steps in Section 1.5.1 of the Aethir Claw Setup Guide.

---

Q: When logging in via email, why can't I receive the verification code?

1. If your email is a Gmail address, log in directly using Google Authentication — both methods work identically as long as the address is valid.

2. If your email is not a Gmail address and Google Authentication is not available, contact AethirCloudClaw technical support for assistance.

---

Q: What is the network bandwidth for my cloud server?

1. Dedicated bandwidth limits: Elite — 15 Mbps · Standard — 8 Mbps · Lite — 4 Mbps.

2. During the MVP phase, bandwidth is enhanced: Standard — 15 Mbps shared across 2 connections · Lite — 15 Mbps shared across 3 connections.

---

Q: How can I increase the network bandwidth for my cloud server?

1. Bandwidth cannot be individually adjusted on request. If you have special requirements, please contact customer service by email.

---

Q: Why is my remote desktop (RDP) connection so laggy?

1. Test the bandwidth between your local network and the cloud server region (US or Japan). Minimum requirements for smooth RDP: stable bandwidth ≥ 5 Mbps and packet loss rate < 1%.

2. Consider upgrading to the Elite plan. The Lite plan provides only 5 Mbps, and cross-country RDP typically experiences lag at that bandwidth.

---

Q: Why is my Telegram bot returning an "LLM request timeout" error?

1. This usually means your LLM API key has exhausted its token quota. Top up your balance on the provider's platform.

2. Alternatively, run openclaw configure to reconfigure or switch the API key.

---

Q: How do I add more port mappings to my cloud server?

1. For security, each cloud server instance exposes only one SSH port by default. All other inbound ports are blocked.

2. To expose additional ports, contact our customer service team by email, describe your use case, and our technical staff will assist offline.

---

Q: How is the security of the cloud server environment ensured?

1. Hardware-level port control: only one SSH port is open per instance; all other inbound ports are strictly blocked.

2. Network-level bandwidth throttling prevents any single server from monopolising shared bandwidth.

3. Cross-instance communication is disabled at the data centre network level — servers are physically isolated from each other.

4. Virtualisation isolation enforces strict user permission controls. Even if a user bypassed the container, access would be limited to that instance's own resources.

5. CPU, memory, and disk resources are physically isolated per instance at startup — no resource contention between servers.

6. The platform monitoring system continuously watches the data centre network. Anomalies trigger targeted alerts; exceeding thresholds results in automatic IP blacklisting.

---

Q: Can I renew my subscription after the lease expires?

1. When your lease expires, the instance is forcibly shut down. During the 7-day grace period, renewing the lease reactivates the instance and restores all data.

2. Once the grace period has passed, the system automatically erases all data on the server. This erasure is permanent and irreversible.

---

Q: Does the cloud server support image snapshots?

1. Snapshots are not supported in the current MVP version. This feature is planned for the upcoming V1.0 release.

---

Q: My cloud server instance has changed to "Error" status — what should I do?

1. An "Error" status is extremely rare and is typically caused by a hardware network failure or system malfunction. Contact our customer service team immediately — they will resolve it as quickly as possible.

2. The AethirCloud monitoring system detects errors in real time and alerts the technical team within 15 minutes. The team commits to responding and resolving within 30 minutes.

3. Even in an "Error" state, user data is generally preserved. Once the issue is resolved, the instance will resume normal operation.

---

End of Manual · For setup instructions, see the Aethir Claw Setup Guide