Suede Skills Directory · Method

How skills are listed and graded

The directory lists public Agent Skills that pass a published, mechanical check. This page is the whole method: where skills come from, what is checked, what gets a page of its own, and how to change a listing.

Where the skills come from

The directory reads the repositories cited by at least two of these 8 curated lists: BehiSecc/awesome-claude-skills, ComposioHQ/awesome-claude-skills, VoltAgent/awesome-agent-skills, heilcheng/awesome-agent-skills, hesreallyhim/awesome-claude-code, karanb192/awesome-claude-skills, skillmatic-ai/awesome-agent-skills, travisvn/awesome-claude-skills. It adds the official collections those lists name and repositories accepted through listing requests. Every SKILL.md in those repositories is read from the default branch, except in hidden folders and dependency or build folders (node_modules, vendor, dist, build), and nothing in them is executed. Accepted requests may name exact canonical skill folders, including a publisher’s hidden distribution folder; only those reviewed roots and skill names are included. System skills, unreviewed mirrors and fixtures remain excluded. This snapshot was taken on October 19, 2026.

The Suede Skill Check

Reviewed listings may also supply supplementary citations from independent curated collections. Each document is pinned to a commit, checked for a direct repository link and reverified on refresh. These citations use the same scoring rubric and do not expand the discovery lists above.

Every SKILL.md runs through a required frontmatter check and seven graded checks: five on the SKILL.md format, following the Agent Skills format, plus a declared license and a commit to the repository in the last year. These checks decide the listing. The quality score then measures how much each listed skill gives an agent to work with.

CheckWhat passesListed skills passing
frontmatterFrontmatter parses and declares a name and a description3,798 of 3,798
name_formatName uses lowercase letters, digits and hyphens, 64 characters or fewer3,798 of 3,798
name_matches_folderName matches the skill's folder name3,682 of 3,798
description_lengthDescription is between 50 and 1,024 characters3,789 of 3,798
description_triggerDescription says when an agent should use the skill2,734 of 3,798
concise_bodySKILL.md body is 500 lines or fewer3,696 of 3,798
licenseA license is declared in the frontmatter or the repository3,798 of 3,798
maintainedRepository has a commit in the last 12 months3,798 of 3,798

How the quality score is built

The score adds whole points from eleven components, all read from published evidence: the SKILL.md frontmatter, description and body, the files in the skill folder, the repository’s last commit and the curated lists that cite it. The rubric totals 100 points.

ComponentPointsHow the points are earned
Structural checks568 points for each of the seven graded checks
Description length55 for 150–600 characters; 3 for 100–900; 1 for 50–1,024
Boundary clause44 when the description says when not to use the skill (“Not for …”, “Do not use …”, “Use … instead”)
Example request33 when the description quotes an example request (3–60 characters in quotation marks)
Instruction depth77 for a SKILL.md body of 500–3,000 words; 4 for 250–4,500; 1 for 100 or more
Reference files33 when the skill folder ships a references/, reference/ or docs/ folder
Bundled scripts22 when the skill folder ships a scripts/ folder
Recent maintenance99 for a repository commit within 30 days of the snapshot; 6 within 90; 3 within 180; 1 within 365
Curated-list citations77 for a repository cited by four or more curated lists; 4 for three; 3 for two; 1 for one
Tool permissions declared22 when the frontmatter declares allowed-tools
Compatibility declared22 when the frontmatter declares compatibility

The published score is the total, capped at 99. 100 is reserved: structural evidence can show that a skill is well formed, well described, documented and maintained, but it cannot prove a skill perfect, so the scale tops out at 99. All arithmetic uses whole numbers; nothing is rounded. Missing evidence earns zero points for its component. A source fetch failure is unscored and stops a refresh from replacing the published snapshot.

Grades in this snapshot

GradeScoreListed skills
A90–99125
B80–891,576
C70–791,373
D0–69724

Across 3,798 listed skills, the highest score is 95 and the median is 78. No listed skill scores 96–99, and 1,701 score 80–99 and carry the quality shields. Failed required frontmatter receives grade F and is not listed.

The score measures the recorded SKILL.md structure and description, instruction depth, bundled files, license declaration, maintenance and citation evidence. It does not measure runtime performance. Frontmatter is a separate required gate; a declared license and all listing-policy rules still apply.

Security scan status is separate

External directory skills are marked not scanned. The HOL/Cisco CI scanners run on Suede’s own repository; their results do not certify the external repositories listed here. The shields’ “Suede Certified” label refers to the published structural quality check. A security result needs its own per-skill scan evidence.

Two README badges and a limited-time feature offer

Skills scoring 80–99 receive separate numeric and A/B shields, each linking to their actual Suede listing. Scores from 90–99 and grade A use gold; scores from 80–89 and grade B use cool metallic styling. The Suede S remains the main mark on both shields. Every listed skill, at any score, keeps its listing badge, and its existing badge.svg and score.svg links keep working.

Display either the numeric or letter badge in your repository or skill README, linked to your Suede listing, then request verification. Displaying both also qualifies. Suede checks the rendered README at a recorded commit before a feature placement is activated. An unverified request, no valid linked badge or failed fetch cannot activate promotion. Verification is checked again after a dataset refresh. Existing editorial full pages keep their separate rules.

Limited-time feature offer: display either the numeric or letter badge in your repository or skill README (both also qualify), then request badge verification. Eligible skills scoring 80–99 can receive a featured placement after Suede verifies at least one valid badge linked to the skill’s listing. Suede has not announced an end date.

Scored below 80? We welcome authors who want help improving their skills. Each full skill page lists its points by component. Send your repository and the points you want to raise, then request a new review after updating the skill.

Listing rules

Which skills get a full page

Every listed skill appears in its category directory and on its repository's page. Skills that pass all seven graded checks get a page of their own once the directory's editors have written their one-sentence summary, up to twelve per repository, chosen by how much ships with the skill: reference files, scripts, assets and a structured SKILL.md. A skill keeps its page while it stays listed, so its README badge and link keep working. Each page carries the summary, a pre-install list of the scripts, commands, environment variables and network use found in the skill's files, install steps (or, where the license restricts copying, the publisher's terms), and a README badge. 733 skills have a full page in this snapshot. A repository whose single listed skill has a page is listed through that page.

Categories

Categories group skills by their main task, so you can compare alternatives side by side. Each skill sits in one of 16 categories, assigned from its name and description.

Refreshes and changes

Each refresh re-reads the source repositories, so quoted descriptions, file facts and check results follow the published SKILL.md. To request a listing, a correction or a removal, use the listing request page; requests are reviewed by hand.