Guide · rating reference
Resource library rating methodology
Understand the five rating criteria, the evidence behind each resource's score, and how to reassess it.
- Source
- Editorial assessment of catalog and existing guides
- Last verified
- October 10, 2026
- Reading mode
- Read or maintain local ratings
- Linked workflow effects
- Maintenance commands update local CSV and Markdown files
- Outcome
- Understand and reassess a resource's usefulness score
Assessed 2026-10-10 (America/Denver). Rubric version 1. Covers all 71 catalog rows, including hidden drafts and excluded resources.
What the number means
The rating is an editorial assessment of how useful the recorded resource is for learning, deciding, or evaluating an agent-skills workflow. It measures the information available in this library's catalog and the existing guides identified below. It is not an installation recommendation, a security score, a popularity measure, or a tested performance guarantee. A high score can accompany unresolved risks when the resource helps a reader understand those risks.
These assessments use the catalog snapshot at commit 02bd1ffe5a06024f89789a4886132f945efb7011 and existing repository guides. External catalog targets were not freshly audited for this rating pass. No catalog package was installed or executed, no benchmark was rerun, and no current availability, maintenance, or safety claim is implied. A catalog-only assessment is an inference from its recorded title, URL, summary, and scope. A catalog-and-guide assessment also uses the linked guide content; any prior scan or benchmark retains that guide's original date, artifact, and limitations. The assessment date is separate from verified_on and from the date the row was added.
Five equally weighted criteria
Each criterion receives 0, 0.5, or 1 point. The rating is their sum, from 0.0 to 5.0, displayed to one decimal. Half-point steps avoid suggesting fine precision that this editorial review cannot support. Missing evidence earns no points for that criterion; it does not prove that the underlying resource is poor.
| Criterion | 0 points | 0.5 points | 1 point |
|---|---|---|---|
| R: relevance | No established relation to the library's purpose | A plausible but unreviewed lead or adjacent topic | A direct skill, agent-workflow, evaluation, or governance use case |
| A: actionability | No reviewed guidance or usable description | Orientation, a feature list, discovery aid, or a bounded discussion | Concrete procedures, technical constraints, examples, or a reusable decision method |
| P: source traceability | No identifiable source | Secondary discovery index, shared conversation, or sources limited to related reading | Identifiable originating document/repository, or a guide tracing substantive claims to named sources |
| V: verification support | No recorded way to assess claims or outcomes | Diagnostic concepts or a limited check, with no complete evaluation method in the assessed material | Explicit tests, validation rules, reproducible evaluation steps, or scoped findings with their limits |
| B: boundaries and limitations | Scope or limits are absent/unreviewed | Intended scope is clear but limitations are lightly developed | Explicit compatibility, uncertainty, failure, data/action, or adoption boundaries |
Publisher identity, stars, installs, price, publication age, featured, and existing scores do not add points. A third-party index can be useful for discovery while earning fewer traceability or verification points than a source-linked procedure. Internal guides receive the same criteria as external resources; being authored by this library adds no points.
How to interpret and change a rating
Scores of 4.5–5.0 indicate strong documented usefulness across most criteria. Scores of 3.0–4.0 indicate a useful resource with specific evidence or scope gaps. Scores below 3.0 indicate a limited lead or sparse assessed information. These are descriptions of the information assessed, not quality certifications. An unreviewed draft can score 1.0 from its plausible topic and identifiable link only and remain hidden. A rating does not publish a draft or imply that its contents were reviewed.
The new rubric replaces the previous unexplained values, including existing nonblank scores. The original October 10, 2026 snapshot linked above contained 15 blank ratings. Current distribution: 5.0: 25; 4.5: 22; 4.0: 10; 3.5: 1; 3.0: 7; 2.5: 2; 1.0: 4. All 71 rows now have a documented score.
To change a score, edit the five criterion values, rationale, evidence references, and catalog fingerprint in resource-ratings.json, then run npm run ratings:sync. Use npm run ratings:check to reject missing or duplicate assessments, invalid points, stale catalog metadata, a CSV score that differs from its assessment, or Markdown that differs from the ledger. Reassess a title, URL, or summary change rather than copying the old rating forward. New IDs require their own assessment, including drafts. Review the underlying guide and update its dated evidence before treating a changed external claim as verified. The script reproduces the arithmetic and publication files; editorial judgment remains explicit in the ledger.
Per-resource assessment ledger
R, A, P, V, and B use the criteria above. Every row uses its recorded catalog description. Additional evidence links point to the exact existing repository snapshot used for this pass. Catalog targets are listed in the CSV and are not fresh source-review receipts.
| Resource ID | Rating | R | A | P | V | B | Basis | Rationale and additional evidence |
|---|---|---|---|---|---|---|---|---|
| avoid-ai-writing | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-and-guide | Concrete detect/rewrite modes and style-versus-authorship limits; the existing comparison adds scoped review context, not demonstrated writing effectiveness. Evidence: guide 1. |
| aura-skills | 3.0 | 1 | 0.5 | 1 | 0 | 0.5 | catalog-only | Identifiable design-skills catalog with a clear workflow niche; the description supplies no outcome checks or assessed package evidence. |
| ming-design-layer-and-editable-design | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | Source-linked output choices, editability checks, exact-artifact findings, and explicit execution/data limits form a complete decision guide. Evidence: guide 1. |
| ling-ui-design | 4.5 | 1 | 0.5 | 1 | 1 | 1 | catalog-and-guide | The reconstruction use case is clear and the existing guide supplies scoped findings and checks; unresolved endpoint and write risks limit adoption readiness despite the useful evidence. Evidence: guide 1. |
| image-to-editable-ppt | 4.5 | 1 | 0.5 | 1 | 1 | 1 | catalog-and-guide | The editable-slide use case has exact-artifact findings and output checks; documented upload and export risks remain unresolved and this score does not authorize execution. Evidence: guide 1. |
| agent-skills-overview | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | A direct format introduction with concrete package concepts; the catalog records limited evaluation detail and few explicit caveats. |
| agent-skills-specification | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | Required structure, frontmatter constraints, file references, and validation provide a traceable technical contract and observable checks. |
| blader-humanizer | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-and-guide | A reusable rewriting method with claim-preservation and voice boundaries; the comparison provides scoped review context rather than a measured quality benchmark. Evidence: guide 1. |
| agent-plugins | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | A named packaging standard with manifest and interoperability concepts; recorded validation details and limits are less developed. |
| openai-build-skills | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | Focused workflow boundaries, supporting resources, dependencies, and representative tests make the recorded authoring guidance directly usable and checkable. |
| openai-record-replay | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | A concrete record-and-draft workflow tied to a stable Mac demonstration; the catalog does not specify a full acceptance test or wider compatibility limits. |
| anthropic-what-are-skills | 3.0 | 1 | 0.5 | 1 | 0 | 0.5 | catalog-only | A traceable plain-language introduction; the recorded scope is conceptual rather than procedural or evaluative. |
| anthropic-complete-guide-building-skills | 4.5 | 1 | 1 | 1 | 1 | 0.5 | catalog-only | The catalog describes an end-to-end authoring, testing, and troubleshooting method; product and operational limits are less explicit in the assessed description. |
| oreilly-skills-context-window | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | An attributed chapter explains package design and progressive disclosure; the recorded description provides fewer concrete tests and caveats. |
| openai-skills-concepts | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-only | A clear division between procedural guidance and MCP-controlled live data/actions; the catalog supplies conceptual verification rather than a complete test method. |
| troubleshoot-validate-structure | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | A specific validator-first diagnostic with named failure classes and a clear distinction between structural validity and runtime behavior. |
| troubleshoot-triggering | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | Positive and negative trigger cases create observable checks, and specificity/breadth warnings bound the recommended edit. |
| anthropic-use-skills | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | An identifiable troubleshooting checklist for enablement and selection; outcome checks and environmental limits are only partly described. |
| troubleshoot-loading | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | Concrete enablement and package checks, valid-frontmatter requirements, and product-dependent loading paths support bounded diagnosis. |
| claude-code-skills | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-only | A concrete scope/precedence reference for Claude Code; the catalog gives clear compatibility boundaries but limited evaluation detail. |
| troubleshoot-precedence | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-only | A specific conflict-resolution procedure with explicit product and namespace scope; a full acceptance test is not recorded. |
| anthropic-skill-authoring | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-only | Concise authoring procedures, deterministic scripts, dependencies, and runtime compatibility are recorded; explicit outcome tests are less developed. |
| troubleshoot-runtime | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | A usable environment/dependency/file/tool checklist with permission checks and explicit OS/host compatibility requirements. |
| agent-skills-course | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | A structured learning path includes creation and diagnosis; assessment methods and platform limitations are only partly specified in the catalog. |
| claude-cowork-record-skill | 3.0 | 1 | 0.5 | 1 | 0 | 0.5 | catalog-only | An attributed workflow demonstration explains the idea; the recorded description lacks reusable acceptance checks or demonstrated reliability evidence. |
| replit-agent-skills | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | Installation, creation, and recurring project preferences are actionable; the catalog has limited verification and boundary detail. |
| anthropic-how-we-use-skills | 4.5 | 1 | 1 | 1 | 1 | 0.5 | catalog-only | Concrete internal workflow examples emphasize observable checks; broader transfer limits are lightly specified in the assessed description. |
| anthropic-verification-loops | 4.5 | 1 | 1 | 1 | 1 | 0.5 | catalog-only | An explicit observe/fail/fix loop makes checks repeatable; compatibility and failure limits are less developed in the catalog record. |
| anthropic-self-service-analytics | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | A measured case study combines governed foundations, evaluation, ambiguity reduction, and adversarial review, supplying both method and constraints. |
| skillopt-project | 4.5 | 1 | 0.5 | 1 | 1 | 1 | catalog-only | A defined research method with scored rollouts, bounded edits, and held-out validation; the overview is less directly operational than the reproduction repository. |
| skillopt-github | 4.5 | 1 | 1 | 1 | 1 | 0.5 | catalog-only | Named source, benchmark configurations, and reproduction guidance support implementation and checking; operational limits are lightly recorded here. |
| skillopt-third-party-snapshot | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | The existing source-linked snapshot separates release/source maturity, evaluation, costs, and authority boundaries and supplies a bounded pilot method. Evidence: guide 1. |
| jetbrains-caveman-token-benchmark | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | A source-linked paired benchmark, accounting caveats, forced-activation limits, and representative pilot checks support a bounded interpretation. Evidence: guide 1. |
| jetbrains-rtk-token-benchmark | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | The guide distinguishes compression counters from billed completed work and supplies trial, compatibility, and failure checks. Evidence: guide 1. |
| headroom-context-compression | 4.5 | 1 | 0.5 | 1 | 1 | 1 | catalog-and-guide | A clear toolkit use case with mixed completed-task evidence and provider/cache limits; the linked evidence guide supports evaluation without implying installation readiness. Evidence: guide 1. |
| headroom-context-compression-evidence | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | Dated source comparisons, denominator/sponsorship limits, scoped security findings, and a paired workload pilot make the evidence inspectable. Evidence: guide 1. |
| jetbrains-ponytail-code-benchmark | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | Paired-task methods, activation and quality caveats, and explicit local pilot gates make the measured efficiency claim useful and bounded. Evidence: guide 1. |
| langchain-eval-engineering | 4.5 | 1 | 1 | 1 | 1 | 0.5 | catalog-only | Trace inspection, measurable abilities, and reproducible Harbor tasks provide a concrete evaluation method; transfer limits are lightly recorded. |
| hamel-evals-skills | 4.5 | 1 | 1 | 1 | 1 | 0.5 | catalog-only | Named evaluation workflows include failure analysis, judge validation, RAG checks, and human review; operational boundaries are less explicit in this article's catalog record. |
| aws-skill-equipped-agent-evaluation | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | A concrete trajectory/production-trace evaluation method distinguishes selection from instruction following and names its runtime/tool context. |
| skill2env-nvidia | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | Public tasks, outcome tests, a skill-derived rubric, and a bounded model/training result provide checkable research claims with a defined experimental scope. |
| obra-superpowers | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-and-guide | Composable development procedures have a specific workflow scope; the existing guide provides review limits, while effectiveness is not established by the security review. Evidence: guide 1. |
| mattpocock-skills | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-and-guide | Modular engineering procedures and selective-adoption boundaries are useful; scoped security review is distinct from measured productivity or correctness. Evidence: guide 1. |
| openspace | 3.0 | 1 | 0.5 | 1 | 0 | 0.5 | catalog-only | An identifiable platform and local/cloud scope; the recorded feature list supplies neither an outcome evaluation nor detailed limitations. |
| vercel-skills-cli | 4.0 | 1 | 1 | 1 | 0.5 | 0.5 | catalog-only | Concrete skill lifecycle commands and named supported agents are useful; the catalog provides limited outcome and operational-boundary detail. |
| skills-rest | 2.5 | 1 | 0.5 | 0.5 | 0 | 0.5 | catalog-only | Domain/origin/setup filters support discovery; a third-party index alone supplies limited primary-source traceability or verification support. |
| aisa-skills | 3.0 | 1 | 0.5 | 0.5 | 0 | 1 | catalog-only | An API-backed discovery catalog with explicit authentication/endpoint/permission caveats; the assessed description contains no concrete outcome checks. |
| elite-ai-tools | 2.5 | 1 | 0.5 | 0.5 | 0 | 0.5 | catalog-only | Keyword discovery and repository links are useful leads; indexing public repositories does not itself evaluate their contents or outcomes. |
| agent-framework-python-skills | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-only | An attributed release description records progressive disclosure and script/approval controls; the production-ready claim is a publisher statement, not a reproduced result. |
| nvidia-scan-agent-skills | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-only | The recorded scanning method covers static/semantic checks, named risk classes, and release-gate triage; this score describes guidance, not scanner accuracy. |
| skillspector-enterprise-training | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | An explicit static/semantic review sequence, coverage interpretation, evidence outputs, and human-governed release decision are documented. Evidence: guide 1. |
| skillspector-skill-demo | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | The annotated skill exposes triggers, ordered steps, verification/update requirements, and explicit authorization boundaries with named references. Evidence: guide 1. |
| skillspector-skill-demo-v2 | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | Distinct registry/static/semantic evidence layers and precise completeness/authorization rules provide a traceable review procedure. Evidence: guide 1. |
| vercel-skills-sh-security-check | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | The guide specifies canonical identity resolution, three-provider evidence collection, coverage limits, and inspection without target installation. Evidence: guide 1. |
| cisco-skill-scanner | 4.5 | 1 | 1 | 1 | 0.5 | 1 | catalog-only | Named scanner layers and optional review support a concrete risk workflow; the catalog records human-judgment limits but no reproduced detection benchmark. |
| sage-agent-security | 4.0 | 1 | 0.5 | 1 | 0.5 | 1 | catalog-only | Tool-call allow/ask/deny behavior and local/optional-check boundaries are clear; the recorded architecture supplies fewer operational steps or measured checks. |
| clawhub | 4.0 | 1 | 1 | 0.5 | 0.5 | 1 | catalog-only | Lifecycle operations and review-before-install boundaries aid discovery; registry-level verification signals do not establish every listed bundle's provenance or runtime safety. |
| so-you-found-a-skill-checklist | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | The source-linked checklist spans full-artifact inspection, dependencies, permissions, reliability, updates, and rollback decisions. Evidence: guide 1. |
| skills-in-the-agentic-performance-system | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | A source-linked system framework connects failure diagnosis, controlled evaluation, configured tools/data, and governance to explicit outcomes. Evidence: guide 1. |
| agentic-performance-source-library | 4.5 | 1 | 0.5 | 1 | 1 | 1 | catalog-and-guide | An annotated source map distinguishes evidence types and limits; its main function is reading and investigation rather than a complete operational procedure. Evidence: guide 1. |
| when-not-to-use-a-skill | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | Named sources, a concrete decision test, observable-verification questions, and trust-boundary constraints make adoption choices explicit. Evidence: guide 1. |
| enterprise-agent-skills-presenter-readiness | 4.5 | 1 | 1 | 0.5 | 1 | 1 | catalog-and-guide | Detailed role-based questions, rehearsal checks, and uncertainty/governance limits aid preparation; cited further reading is mainly other library guides. Evidence: guide 1. |
| academic-research-skills | 3.0 | 1 | 0.5 | 1 | 0 | 0.5 | catalog-only | An identifiable collection and review-first lead; the sparse recorded description lacks concrete research procedures or evaluation evidence, and exclusion remains unchanged. |
| superpowers-evaluation-dialogue | 3.0 | 1 | 0.5 | 0.5 | 0 | 1 | catalog-only | An attributed discussion frames evaluation questions and explicitly disclaims primary evidence; it does not provide a recorded validated outcome. |
| theo-skills-code-video | 1.0 | 0.5 | 0 | 0.5 | 0 | 0 | catalog-only | A plausible skills/code lead with an identifiable video URL; title, transcript, and summary remain unreviewed. The draft stays hidden. |
| skills-video-long | 1.0 | 0.5 | 0 | 0.5 | 0 | 0 | catalog-only | A plausible skills lead and video URL only; content review and source summary are pending. The draft stays hidden. |
| skills-video-short | 1.0 | 0.5 | 0 | 0.5 | 0 | 0 | catalog-only | A plausible skills lead and video URL only; content review and source summary are pending. The draft stays hidden. |
| gemini-skills-share | 1.0 | 0.5 | 0 | 0.5 | 0 | 0 | catalog-only | A plausible shared-conversation lead; content and primary-source summary remain unreviewed. The draft stays hidden. |
| ai-evals-short-guide | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | Concrete human-trace-to-evaluation steps, pinned nine-skill evidence, and explicit coverage/runtime limits support a bounded starting workflow. Evidence: guide 1. |
| skillselion | 3.5 | 1 | 0.5 | 0.5 | 0.5 | 1 | catalog-and-guide | Discovery rankings have documented denominators and third-party-signal limits in the existing guide; popularity and passed-through verdicts do not establish version-level safety. Evidence: guide 1. |
| skill-security-over-time | 5.0 | 1 | 1 | 1 | 1 | 1 | catalog-and-guide | A dated source-linked lifecycle method includes exact-version receipts, update triggers, scoped review coverage, quarantine, and rollback limits. Evidence: guide 1. |