Awesome OpenClaw Skills

What It Does

VoltAgent's awesome-openclaw-skills is a curated markdown index of skills published on ClawHub, the public registry for the OpenClaw assistant (the artist formerly known as Clawdbot, hence the repo's rename trail). The repo holds a 1,255-line README with a sponsored ecosystem section and inline category previews, plus 30 per-category markdown files that carry the actual catalog: 5,151 bullet entries as of test day, each linking to a per-skill page on the companion site clawskills.sh, which in turn links to the underlying ClawHub listing. The audience is OpenClaw users hunting for skills and skill authors seeking distribution. The pitch is curation: the README claims 7,215 registry entries were filtered out (spam, duplicates, low quality, crypto, and 373 malicious entries identified by external security audits), leaving the 5,300-plus worth your time. There is no executable code in the repo; it is a reading product with an install instruction.

The Good

The install path works, and the artifacts it lands are supply-chain clean. Following the README's npx clawhub install instruction, 4 of 6 sampled catalog entries (steipete/slack from the CONTRIBUTING example, xizhuomengcontin/orca-replay, tuanpmt/1p-shortlink, rhyssullivan/answeroverflow) installed real SKILL.md files with valid frontmatter, a _meta.json with version and publish timestamp, and a .clawhub/origin.json provenance manifest recording the registry URL, owner handle, installed version, timestamp, and a SHA fingerprint. That provenance file is better hygiene than most skill distribution channels GearScope has tested.

clawskills.sh is a real product, not a link farm. The companion site (Astro-built, live, 5,167 skill pages) gives every skill a page with version, download count, a VirusTotal report link, the official OpenClaw security status (Benign or Suspicious), the original SKILL.md, an example workflow, and requirements. A sampled page (attio-crm) linked a real VirusTotal file-analysis URL. This is security context the raw GitHub list cannot carry, and it is the strongest reason to use this catalog over searching ClawHub directly.

Link freshness and intake cadence are excellent. All 35 sampled catalog URLs returned HTTP 200. The repo is alive: 499 commits, 100+ contributors, 5 open PRs at test time all following the "Add skill: owner/name" convention, and merge activity within five days of review. The one CI workflow enforces that PR descriptions contain a ClawHub link, which matches the repo's ClawHub-only contribution policy. The README's security notice is also refreshingly blunt: skills here are "curated, not audited."

The Bad

The repo cannot count itself, in six different ways. The actual entry count across the 30 category files is 5,151. The README hero says "Discover 5300+", the README badge says 5200, the GitHub repo description says "5,400+", the README table of contents sums to 5,319, and the per-category header counts sum to 5,376. Twenty-seven of thirty category headers overstate their own file's entry count (Gaming claims 36 and has 26, Moltbook claims 44 and has 35, Marketing and Sales claims 108 and has 98). For a product whose entire value proposition is filtering a registry down to trustworthy numbers, none of its numbers reconcile.

The headline filter claims are leaky. The README says 886 crypto, blockchain, and finance entries were excluded, and CONTRIBUTING says "No crypto, blockchain, DeFi, or finance-related skills for now." A strict scan of the shipped catalog finds 129 to 141 crypto-adjacent entries still listed, including a CoinGecko price fetcher, two Zapper DeFi portfolio skills, a Sui blockchain explainer, a "Secure crypto wallet for AI agents," a Bitcoin Lightning pay-per-request proxy, and a USDC payment skill. Duplicate filtering also leaks: 3 skill URLs appear in two categories each and 20 display names repeat.

Listed entries go stale with no sweep. Two of six sampled entries (kesslerio/attio-crm in Search and Research, matanle51/agentic-paper-digest) fail to install with "Skill not found," and ClawHub's own API confirms both are gone from the registry. The catalog still lists them, and attio-crm's clawskills.sh page still advertises 1.7k downloads. Liveness of the link is not liveness of the skill. Related small defects: the clawhub CLI exits 0 after printing an install error, its unauthenticated rate limiter (roughly one install per minute) reports itself as "Skill not found or unavailable to this account," and it writes into /skills/ when the working directory is a git repo, which dropped skill folders into an unrelated project during testing.

Editorial quality below its own bar, and private tooling left in the open. CONTRIBUTING requires descriptions of 10 words or fewer; the catalog's own entries routinely run longer, many machine-truncated mid-sentence ("...comments in a specific."), one description is "| Key | Value |.", another is a personal fragment ("- Neill explicitly requests Billy system repair."), and single-letter skill names ("1", "a") survived the junk filter. The committed .claude/settings.local.json exposes the maintainer's home directory path and reveals that categorization runs through a private /tmp/categorize.py that is not in the repo, so the category files and their inflated counts cannot be regenerated or audited by anyone else. Five sponsor placements (Crawlbase, SerpApi, trentclaw, EveryFeed, LaunchKit) plus a "Become a Sponsor" CTA monetize a list that presents itself as a neutral filter, and the README badge and CONTRIBUTING issue links still point at the pre-rename awesome-clawdbot-skills name.

Smoke Test Results

Tested on the host (macOS 26.5.2, aarch64) against a fresh shallow clone, since the repo is pure markdown with no install step. Scripts and logs live under sandbox/skills/awesome-openclaw-skills/.

Structural validation (13 checks)

$ git clone --depth 1 https://github.com/VoltAgent/awesome-openclaw-skills /tmp/aoc-structure
✅ 1.4MB clone, 30 category files, LICENSE, CONTRIBUTING, 1 CI workflow

$ count entries across categories/*.md
✅ 5,151 bullet entries, 5,148 unique URLs (3 cross-category duplicates)

$ grep -c '^\*\*[0-9]* skills\*\*' per category file vs actual
❌ 27 of 30 headers overstated (header sum 5,376 vs actual 5,151)

$ strict crypto-pattern scan of shipped entries
❌ 129 crypto-adjacent entries listed despite the 886-excluded claim

$ junk-description scan (| Key | Value |., <!--, leading dash, personal fragments)
❌ 8 in sample classes, plus 34 machine-truncated descriptions

$ README TOC counts vs actual
❌ TOC sums to 5,319 vs actual 5,151

$ badge "skills-5200" vs actual
❌ badge 5200 vs actual 5,151 (hero says 5300+, GitHub desc says 5,400+)

$ HTTP HEAD on 10 sampled clawskills.sh skill pages
✅ 10 of 10 return 200

$ check .claude/settings.local.json for committed personal paths
❌ exposes /Users/necatiozmen/... and references private /tmp/categorize.py

Pass rate: 7 of 13. Every failure is a bookkeeping or filter-integrity defect, not a broken link or missing file.

Full structure log →

Install-path functional checks (6 sampled entries)

$ npx clawhub install xizhuomengcontin/orca-replay
✅ Installed v1.0.1, SKILL.md + _meta.json + origin.json with SHA fingerprint

$ npx clawhub install kesslerio/attio-crm
❌ "Skill not found or unavailable to this account" (ClawHub API: Skill not found)

$ npx clawhub install tuanpmt/1p-shortlink
✅ Installed v1.0.0 with valid frontmatter and provenance manifest

$ npx clawhub install rhyssullivan/answeroverflow
✅ Installed v1.0.2 (earlier probe)

$ npx clawhub install matanle51/agentic-paper-digest
❌ "Skill not found" (ClawHub API confirms gone; entry still listed)

$ npx clawhub install steipete/slack (CONTRIBUTING's own example)
✅ Installed v1.0.0 to ~/.openclaw/workspace/skills/@steipete/slack

Functional pass rate: 4 of 6. Both failures are stale catalog entries, confirmed against ClawHub's API, not rate-limit noise.

Full functional log →

What the runs tell you

The reading product works: every link sampled is live and the install instruction delivers real, provenance-stamped skills more often than not. The catalog's metadata layer is what fails: counts inflated at every level, filter claims contradicted by the shipped list, and stale entries that no sweep catches. Treat it as a well-organized pointer board, not a verified database.

Setup Walkthrough

1. Browse to clawskills.sh or open the repo README and pick a category (30 exist, from Coding Agents and IDEs at roughly 1,162 entries down to iOS and macOS Development at 29).
2. Click a skill to open its clawskills.sh page, which shows the VirusTotal report, OpenClaw security status, version, and download count.
3. Install with openclaw skills install / if you run OpenClaw, or npx clawhub install / for a registry-managed folder outside a full OpenClaw workspace.
4. Verify what landed: a SKILL.md with frontmatter, _meta.json, and .clawhub/origin.json containing the registry URL and fingerprint.

Gotchas from testing: the clawhub CLI rate-limits anonymous installs to roughly one per minute and reports that limit as "Skill not found," so retry after 60 seconds before concluding an entry is stale. Inside a git working tree it installs to /skills/@owner/ rather than the home directory, so check where you are standing when you run it. If an entry fails install, search the slug on clawhub.ai to see whether the registry still has it.

Alternatives

1. ClawHub directly - the upstream registry at clawhub.ai with search, security status, and VirusTotal reports; prefer it when you know the skill name and want ground truth, since this catalog can lag deletions.
2. clawskills.sh standalone - same data as the repo with better presentation and install copy commands; skip GitHub entirely if you only browse.
3. awesome-claude-code (hesreallyhim) - GearScope-reviewed alternative if your harness is Claude Code rather than OpenClaw; stronger link-governance CI, smaller scope.
4. wshobson/agents - a multi-harness plugin marketplace (Claude Code, Codex, Cursor, OpenCode, Copilot) if you want installable agents rather than a link catalog.