Collection and provenance
Scheduled collection discovers configured public GitHub repositories, clones the current default branch, finds SKILL.md files, validates their schema, removes exact content duplicates, and generates static pages. Each record preserves the repository, path, commit SHA, raw source URL, license, and collection timestamp.
Automated quality score
The 100-point score is a screening signal, not a review verdict. It has four independently displayed components:
| Component | Maximum | Signals |
|---|---|---|
| Documentation | 30 | Description length, source length, headings, code blocks, examples, usage, and workflow sections. |
| Specificity | 25 | Concrete task verbs, triggers, numbered workflow, expected output, commands, paths, and source links. |
| Maintenance | 20 | Source repository push recency and frontmatter completeness. |
| Trust signals | 25 | License, frontmatter, limitations, safety language, prerequisites, links, and deductions for higher-risk static signals. |
Scores are capped within each component. They do not use repository stars as a quality input.
Ranking boards
- Weekly, monthly, and yearly boards use repository star snapshot changes.
- When snapshot history is incomplete, repository stars are used only as a cold-start ordering signal.
- The quality board uses the deterministic 100-point score above.
- The new-repository board uses the source repository creation date, not a claimed SKILL.md publication date.
Static risk signals
Rule-based scanning identifies source lines that appear to involve network access, script execution, file reads, file writes, or outbound data. The page shows the line, excerpt, reason, and severity. A detected signal is a prompt to inspect the source; no detection is proof of safety.
Classification
Industries, functions, scenarios, technology stacks, and Agent platforms are inferred from weighted name, description, frontmatter, headings, and opening prose. Classification is deterministic but can produce false positives. Corrections are applied to rules and then regenerated across the catalog.
Date labels
Repository creation and push dates are labeled as repository dates. The collection timestamp is labeled as the page data retrieval date. SkillSignal does not use repository creation as an article publication date. A page-level publication or modification date is added only when file history or explicit upstream metadata supports it.
Search indexability
A detail page must first pass automated thresholds for documentation depth, quality, language match, uniqueness, and placeholder detection. It must also have an explicit independent editorial or test review record. Automated checks alone never place a detail page in the sitemap. Pages that fail remain available to users with noindex, follow.
Refresh and reproducibility
Catalog data is intended to refresh daily. Current totals and generation time are available in stats.json, while the complete machine-readable summary is available in skills-index.json.