The ClawHub skills ecosystem has exploded. 10,700+ community-built skills as of April 2026, growing daily. That's the good news. The bad news: 820+ of those skills — 7.6% — have been flagged as malicious by the ClawHub security team. Skills are modular capability extensions that run with real permissions inside your OpenClaw instance. Installing a bad one isn't just annoying — it's a security event.
You need a curated list. Here are the 10 skills worth installing and the 5 to avoid — with install commands, permission levels, and real use cases for each.
How Skill Installation Works
Before the list: the three-step process and what to watch for.
- Browse: Visit clawhub.dev or search via CLI:
openclaw skill search <keyword>. Check install count, rating, permissions, and whether source code is publicly available. - Install:
openclaw skill install <skill-name>. The installer shows permissions requested — review before confirming. Installation takes under 10 seconds. - Verify:
openclaw skill listto confirm active.openclaw skill inspect <skill-name>to review permissions.openclaw skill disable <skill-name>to remove access without uninstalling.
Permission tiers to watch for: exec (shell commands — highest risk), file-write (filesystem access), network (outbound HTTP/API calls), read-only (lowest risk). A skill requesting exec access should have a clear, auditable reason. Anything that requests exec + network without documented justification should be skipped.
The 10 Best OpenClaw Skills (2026)
Browser Control
Automate web browsing, form filling, data scraping, and screenshots. The backbone of research and data collection workflows. High permissions are justified — this skill literally controls a browser. Review the source before installing and use it with approval gates enabled.
File Manager
Read, write, rename, move, and organize files on your local machine. Nearly every workflow depends on this at some point. Well-maintained, widely used, and the permissions are scoped correctly — no exec, no network.
Calendar Sync
Full Google Calendar integration — create events, check availability, get reminders, manage scheduling via chat. Low-risk permissions, high practical value. OAuth-based, so your credentials stay in Google's hands.
GitHub Integration
Create PRs, review diffs, manage issues, merge branches — all via chat. Wraps the GitHub API with natural language. One of the most actively maintained skills on ClawHub with 50k+ installs and a strong review track record.
Email Assistant
Draft, send, and search emails in natural language. Works with Gmail and Outlook via OAuth. The key safety note: configure it to draft-and-notify rather than auto-send. Review before any email goes out.
Database Query
Talk to PostgreSQL or MySQL in plain English. Natural language → SQL → results. Start in read-only mode; only enable write permissions for specific workflows you've audited. Saves hours on routine data questions.
Notion Sync
Read and write Notion pages, databases, and blocks via chat. One of the most popular skills in the productivity category. Well-audited, no exec access, clean API scope. Works well for knowledge management workflows.
Webhook Trigger
Send webhook payloads to any URL from your agent. Bridges OpenClaw to Zapier, Make, n8n, and any webhook-compatible service. Network-only, no file or exec access. The connective tissue for complex automation stacks.
Web Search Pro
Enhanced web search with structured output — returns titles, URLs, snippets, and publication dates in agent-readable format. Better than the built-in web_search tool for research workflows that need source tracking and citation.
Audit Logger
Structured audit logging for all tool calls, with timestamps, action type, and outcome. Writes to a local JSON log file. Essential for production deployments where you need a complete record of what your agent has done. Low risk, high value for accountability.
5 Skills to Actively Avoid
❌ Any skill with exec + network + no public source
This combination — shell execution, outbound network access, and closed source — is the exact profile of every malicious skill ClawHub has flagged. There is no legitimate reason for a skill to need all three without auditable code. Skip it regardless of the description.
❌ "All-in-one automation" bundled skills
Skills that claim to replace 10 tools at once typically request far more permissions than they need and have less focused code than single-purpose skills. The broader the claim, the wider the attack surface. Use purpose-built skills that do one thing well.
❌ Skills with zero installs and no reviews
New skills aren't inherently bad — but an unreviewed skill with exec access and no install history is a meaningful risk. Let community validation accrue before installing. Check the ClawHub security feed for newly flagged skills before installing anything under 100 installs.
❌ Cracked or "premium" skill reposts
Several flagged skills are repackaged versions of paid commercial skills, distributed through ClawHub to bypass licensing. Beyond the legal issue, these often contain modified code that adds a network exfiltration layer. The 820+ flagged skills are disproportionately from this category.
❌ Skills requesting keychain or credential manager access
No skill needs access to your system keychain or OS credential manager. If a skill requests this permission, it's either poorly built or actively malicious. Decline and report it via ClawHub's flag system.
Before You Install Anything: A 30-Second Check
- Run
openclaw skill inspect <skill-name>and read the permissions list - Check if source code is public — if not, treat exec/network permissions as red flags
- Look at install count and date: skills with high installs and recent activity are lower risk
- Search ClawHub's security feed for the skill name before installing
Skills are powerful — that's why they're worth using. But "powerful" means they need the same scrutiny you'd give any third-party software running on your machine. The 30-second check is the minimum viable due diligence.
Want Your Skills Stack Set Up Correctly?
ClawReady's setup packages include a curated skill installation — we select, install, and configure the right skills for your use case with proper permission scoping. No guesswork, no bad installs.
See Setup Packages →