The Quality Score is an independent rating system created by clawhub-skills.com — it is not an official feature of clawhub.ai. We built it to help users cut through a large and fast-growing directory of skills and quickly identify which ones are worth installing. This page explains exactly how the score is calculated, what each dimension measures, and how to interpret the grade badges you see on skill cards.
Downloads are the primary demand signal — they capture every package fetch including users in air-gapped environments, manual SKILL.md placement, and non-sync workflows. They are noisier than installs (automated tools contribute), but at scale they reliably reflect real interest. Install counts (tracked via clawhub sync) are treated as a floor metric: their presence confirms real users, but their absence does not mean a skill has none — the install count is a known severe undercount of actual usage. The quality score weights downloads as the dominant market signal, with installs, stars, documentation depth, and maintenance as supporting dimensions.
The score is fully rule-based — no subjective human curation, no AI opinion. Every point is derived from objective, verifiable data synced from the ClawHub registry into our database.
Top-tier skills with strong real-world adoption, complete documentation, and active maintenance. Safe to use with confidence.
Well-rounded skills that perform well across most dimensions. Minor gaps in documentation or install count, but generally reliable.
Average skills — functional but with notable gaps. May lack detailed docs, have low install counts, or be newly published.
Below-average skills with limited adoption or poor documentation. Worth a try if the concept matches your need, but proceed with caution.
Skills with very low adoption and minimal documentation. Likely experimental or abandoned. Review the source before installing.
Red-flag skills. Score below 20 indicates minimal documentation, no tracked installs, and little evidence of real adoption across any dimension.
Does anyone actually use this skill?
Downloads are the primary demand signal. Unlike installs, downloads capture every fetch of the skill package — including users in air-gapped environments, those who place SKILL.md manually, and anyone who never runs clawhub sync. Downloads are noisier (automated tools and CI pipelines contribute), but at scale they reliably reflect real interest. A skill with tens of thousands of downloads has demonstrable demand regardless of its sync-tracked install count.
| ≥ 50,000 downloads | 20 pts | Exceptional demand |
| ≥ 10,000 downloads | 17 pts | Very strong demand |
| ≥ 5,000 downloads | 14 pts | Strong demand |
| ≥ 2,000 downloads | 11 pts | High demand |
| ≥ 500 downloads | 7 pts | Moderate |
| ≥ 100 downloads | 4 pts | Low |
| ≥ 20 downloads | 1 pt | Very low |
Install counts (reported via clawhub sync) are a floor metric — they confirm real users exist, but absence of installs does not mean a skill has no users. Users who place SKILL.md manually, run in air-gapped environments, or disable telemetry are invisible to this metric. When installs are present they are strong evidence of genuine adoption; the score treats them as a corroborating signal rather than the primary one.
| ≥ 100 installs | 8 pts | Strong confirmed adoption |
| 50 – 99 | 7 pts | Good |
| 20 – 49 | 5 pts | Above average |
| 10 – 19 | 4 pts | Average |
| 5 – 9 | 3 pts | Low |
| 2 – 4 | 2 pts | Very low |
| 1 | 1 pt | Minimal |
| 0 | 0 pts | No tracked installs |
Stars are a deliberate human action — someone bookmarked the skill because they found it useful or wanted to track it. Unlike downloads, stars cannot be inflated by automated processes, making them a reliable quality signal.
| ≥ 10 stars | 4 pts | Popular |
| ≥ 5 stars | 3 pts | Well-liked |
| ≥ 1 star | 1 pt | Some interest |
A positive trending score (week-over-week install growth) adds bonus points. Skills with rapid recent growth signal emerging adoption.
| Trending ≥ 2.0 | 3 pts | Hot |
| Trending ≥ 1.0 | 2 pts | Growing |
| Trending ≥ 0.5 | 1 pt | Slight uptick |
Can a user understand what this skill does and how to use it?
A SKILL.md file is the canonical documentation format for OpenClaw skills. Its mere presence earns 8 pts — it signals the author made a deliberate effort to document their work.
Longer documentation generally means more thorough coverage of use cases, configuration, and examples.
| ≥ 3,000 characters | 6 pts | Detailed |
| ≥ 1,500 characters | 4 pts | Adequate |
| ≥ 500 characters | 2 pts | Minimal |
| < 500 characters | 0 pts | Too brief |
Skills that explicitly define their tools block (the list of callable functions) are significantly more useful to agents. Detected via regex on SKILL.md content.
Docs that include trigger phrases, example prompts, or usage scenarios help users discover when and how to invoke the skill.
A meaningful one-line summary (>80 chars) shown on the skill card earns 2 pts; a short summary (>20 chars) earns 1 pt.
Is the skill package well-structured?
Skills with a skillAssets bundle (downloaded package contents) earn 6 pts. This confirms the skill has been synced and has actual distributable files.
A SKILL.md file inside the downloaded package (distinct from the fetched markdown) confirms docs ship with the skill.
A README or AGENTS.md alongside the skill signals a more complete, production-ready package.
The presence of .sh, .py, .js, .ts, or .json files indicates the skill has executable or configurable components — not just markdown.
Is there an identifiable author who maintains this skill?
A named author signals accountability. Anonymous skills are harder to report issues to or follow for updates.
A version string indicates the author follows a release cycle. Skills without versions are often early experiments that may never be updated.
A changelog proves the skill has been actively revised. It also helps users assess whether known issues have been addressed.
Skills that remain available on clawhub.ai earn this bonus. Delisted skills receive 0 pts here and are marked with a banner on their detail page.
Does the skill respect user privacy?
Every skill begins with the full 10 pts. Points are only ever deducted — never added — based on verified privacy risk signals.
Skills tagged privacy-risk have been manually reviewed and confirmed to collect identifying information (e.g. username, machine hostname) and transmit it to an external server on every run. This is the most severe deduction in the entire scoring system because it is based on verified human audit — not heuristics. A skill can still receive a passing score if it compensates in other dimensions, but the deduction is large enough to drop most affected skills to grade C or below.
| Has privacy-risk tag | −8 pts | Confirmed data exfiltration |
| No privacy-risk tag | 0 pts | No known privacy issue |
clawhub sync are counted. Users who add SKILL.md files manually, run in air-gapped environments, disable telemetry, or never run sync are invisible to this metric. A skill with 0 tracked installs may still have real users. See ClawHub telemetry.md for exactly what is and isn't collected.