Read catalogs by purpose rather than popularity

Separate official status, familiarity, and novelty from usefulness for your own work.

Suggested learning time: 45 minutes.

1. 1Desired deliverable
2. 2One candidate
3. 3Dependency and license review
4. 4Unexecuted evaluation card

Original learning map: arrows show the reading or decision sequence, not a measured execution trace.

Prerequisites

• Separate discovery, reading, and use

Evidence and exercise status

This chapter is an editorial learning guide. Reading sources is distinct from executing a Skill and measuring its effect. Exercise status: not-run. No experiment logs or model outputs exist.

Learning goals

• Select candidates by use case and dependencies.
• Do not assign numerical values to expected effects that have not been verified.

Roles in the work

• Learner: define hypotheses and grading criteria.
• AI: assist with reading and deliverable creation within the authorized scope.
• Reviewer: inspect outcomes and logs separately.

Inputs

• The input examples specified in this chapter.
• The official material and versions to verify.

Choose by deliverable first

This course's catalog offers entry points for design and debugging, UI and React, Skill authoring, file processing, local AI, and SDK work. Improving a textbook site's appearance and reducing waits during data retrieval are both frontend tasks, but call for different Skills. Select the individual Skill that addresses the current failure, rather than adopting an entire well-known repository.

How to interpret “official”

Anthropic and OpenAI repositories include skill-creator guides for creating and updating Skills. Read both as starting points for considering input examples, required materials, and verification, rather than tools whose job ends when a Skill is generated. This course includes them as resources for learning how Skills work. Official distribution is not experimental evidence of equal effectiveness across all models. The catalog separates expected use from measurements in this course; the latter have not been measured.

Anthropic skill-creator, OpenAI skill-creator

Do not mix new, established, and popular

The verification date is when the information was checked, rather than its publication or last-update date. A main URL can change later. Stars and rankings can indicate attention but do not guarantee quality or safety. This course did not conduct a comprehensive popularity survey and does not call its selection a “popular top ten.” Separate entry points for learning the basics from official collections for particular tasks, and recheck the current implementation in official change histories.

Do not add everything at once

Start with one candidate. Adding multiple Skills together makes it hard to know why results improved or worsened. If you combine debugging and TDD, later compare A alone, B alone, and A+B. Record the target task, weak situations, required tools, external destinations, license, references, and version on each evaluation card. Read installation instructions to understand them; do not execute them unchanged before reviewing the content.

A 12-item learning catalog organized by purpose

• This is a learning catalog, not a comprehensive ranking of the newest or most popular Skills. It focuses on official specifications, official collections, and candidates requested by the user.
• The verification date is when the public page was read. It is not the publication, update, or installation date.
• No Skill was installed or executed for this course. There are no measured results for effectiveness, speed, or cost.
• Except for two items whose content was checked at fixed commits, references are rolling. Pin the adopted version before an experiment.
• License labels record the statements checked. Individual files, dependencies, and services may have separate conditions.

Superpowers — obra

Item ID: superpowers-framework. Kind: Framework. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: null (entry point to a collection or framework).

• Purpose: A development methodology using multiple Skills for design, planning, implementation, and verification.
• Expected use: When you want to reduce missing process steps in development with unclear requirements.
• Limitations: Fixed process overhead may be large for a simple small correction.
• Environment and dependencies: Client-specific installation mechanisms, hooks, file operations, and subagents when needed.
• Evaluation notes: Evaluate the complete workflow and individual Skills as separate conditions; also record nested Skill and hook activation.
• Safety checks: Scope drift, increased costs from more subagents, hooks, and persistent settings.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: MIT. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: obra/superpowers, Superpowers license

Systematic Debugging — obra

Item ID: superpowers-debug. Kind: Individual Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/systematic-debugging/SKILL.md.

• Purpose: A debugging procedure moving from cause investigation to hypothesis testing.
• Expected use: When you want to reduce unsupported trial and error or recurrence after a fix.
• Limitations: Following a procedure alone cannot solve a problem without observable logs or a reproduction environment.
• Environment and dependencies: Read access to target code and execution logs, and a test environment when needed.
• Evaluation notes: Observe reproducibility, evidence, unrelated changes, and regressions as well as final success.
• Safety checks: Secrets in logs and excessive expansion of the investigation scope.

Version record: ref b36e0829c6d0140e93cfef2ca599b1b07d4a7797, pinned commit b36e0829c6d0140e93cfef2ca599b1b07d4a7797, declared version v6.3.0 repository snapshot, review status pinned-content-verified. The body was checked at a pinned commit. This does not assert that it is identical to the current latest version.

License statement: MIT. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Superpowers systematic-debugging: 固定版, Superpowers license

Test-Driven Development — obra

Item ID: superpowers-tdd. Kind: Individual Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/test-driven-development/SKILL.md.

• Purpose: A development procedure that observes a failing test first.
• Expected use: Small feature implementations and bug fixes with behavior that can be checked.
• Limitations: Tests alone cannot determine quality for tasks assessed only by subjective appearance.
• Environment and dependencies: A test runner, a known initial code state, and command-execution logs.
• Evaluation notes: Score process adherence separately from correctness on unseen inputs.
• Safety checks: Do not prioritize forceful deletion instructions over user authorization or protection of existing work.

Version record: ref b36e0829c6d0140e93cfef2ca599b1b07d4a7797, pinned commit b36e0829c6d0140e93cfef2ca599b1b07d4a7797, declared version v6.3.0 repository snapshot, review status pinned-content-verified. The body was checked at a pinned commit. This does not assert that it is identical to the current latest version.

License statement: MIT. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Superpowers test-driven-development: 固定版, Superpowers license

skill-creator — Anthropic

Item ID: anthropic-creator. Kind: Authoring Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/skill-creator/SKILL.md.

• Purpose: Support for creating, revising, and evaluating a Skill.
• Expected use: An entry point for learning to define a use case, test inputs, and revisions based on feedback.
• Limitations: Check the generated Skill's safety and effectiveness separately.
• Environment and dependencies: Depending on the environment, CLI tools, evaluation scripts, subagents, and other resources may be required.
• Evaluation notes: Separate examples used for development from unseen evaluation examples.
• Safety checks: Costs and permissions associated with large evaluation batches or external APIs.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: Apache-2.0. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Anthropic skill-creator, Anthropic skill-creator license

skill-creator — OpenAI

Item ID: openai-creator. Kind: Authoring Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/.system/skill-creator/SKILL.md.

• Purpose: A guide to creating and updating Skills for Codex.
• Expected use: Narrow a role and design descriptions, references, and helper scripts.
• Limitations: Client-specific metadata may not transfer unchanged to other environments.
• Environment and dependencies: A compatible Codex environment and bundled authoring or validation scripts when needed.
• Evaluation notes: Verify positive and negative activation cases separately from deliverable quality.
• Safety checks: Read generated code and dependency scripts before execution.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: Apache-2.0. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: OpenAI skill-creator, OpenAI skill-creator license

frontend-design — Anthropic

Item ID: anthropic-frontend. Kind: Individual Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/frontend-design/SKILL.md.

• Purpose: Guidance for intentional visual design and UI creation.
• Expected use: Try expression suited to a brief when work tends toward generic templates.
• Limitations: Aesthetic quality depends on the audience and purpose and is difficult to determine with a single automated score.
• Environment and dependencies: A UI implementation and display environment, screen sizes for verification, and available images or fonts.
• Evaluation notes: Use the same brief for anonymous comparisons; score readability and usability too.
• Safety checks: Asset and font rights, external image retrieval, and accidental publication.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: Apache-2.0. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Anthropic frontend-design, Anthropic frontend-design license

pdf — Anthropic

Item ID: anthropic-pdf. Kind: Individual Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/pdf/SKILL.md.

• Purpose: Guidance for reading, creating, converting, and otherwise working with PDFs.
• Expected use: Read as an example of separating procedures by file format.
• Limitations: Do not assume unrestricted redistribution merely because it is in a public repository.
• Environment and dependencies: PDF libraries and CLI tools vary by task.
• Evaluation notes: Check layout, extracted values, missing pages, and similar issues in the actual deliverable.
• Safety checks: Documents containing personal information, protected information, and license restrictions.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: Proprietary / custom terms. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Anthropic PDF skill, Anthropic PDF license

vercel-react-best-practices — Vercel

Item ID: vercel-react. Kind: Individual Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/react-best-practices/SKILL.md.

• Purpose: Implementation and review guidance focused on React/Next.js performance.
• Expected use: A candidate for checking serialized waits, bundles, and related issues.
• Limitations: Applying rules does not guarantee faster performance in a real environment.
• Environment and dependencies: The target project and its React/Next.js versions, plus a measurement environment.
• Evaluation notes: Fix load, cache, and device conditions before and after changes, and measure performance separately.
• Safety checks: Careless bulk replacement, specification changes, and mechanical application of outdated rules.

Version record: ref main, pinned commit null, declared version 1.0.0, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: MIT (declaration checked in SKILL.md). This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Vercel React Best Practices

web-design-guidelines — Vercel

Item ID: vercel-web. Kind: Individual Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/web-design-guidelines/SKILL.md.

• Purpose: Retrieve an external guide and review UI code.
• Expected use: Align review perspectives and try findings that identify specific locations.
• Limitations: Code alone cannot verify actual interaction quality or all accessibility concerns.
• Environment and dependencies: Network access to retrieve the guide and read access to target code.
• Evaluation notes: Save versions, dates, and hashes for both the Skill and the retrieved guide.
• Safety checks: Updates to external references, retrieval failures, and excessive trust in instructions from the guide.

Version record: ref main, pinned commit null, declared version 1.0.0, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: Individual license not established in this review. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Vercel Web Design Guidelines

huggingface-local-models — Hugging Face

Item ID: hf-local. Kind: Individual Skill. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: skills/huggingface-local-models/SKILL.md.

• Purpose: Guidance on choosing and operating local models, GGUF, and llama.cpp.
• Expected use: An entry point to consult when selecting models for the local AI course.
• Limitations: Recheck guidance for the hardware, model, and dependency versions used.
• Environment and dependencies: Model distributors, a local inference environment, and checks for capacity and licensing.
• Evaluation notes: Record recommendation reasons, exact file names, and device conditions; do not fill speed fields with guesses.
• Safety checks: Large downloads, model usage terms, and configuration that starts external servers.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: Apache-2.0 (repository LICENSE). This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Hugging Face local models skill, Hugging Face skills license

Microsoft Agent Skills — Microsoft

Item ID: microsoft-collection. Kind: Collection. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: null (entry point to a collection or framework).

• Purpose: A collection of Skills and related configuration for Azure SDKs and Microsoft AI.
• Expected use: Find examples matching a target SDK and its programming language.
• Limitations: Select the individual items you need rather than treating the whole collection as one Skill.
• Environment and dependencies: Check each SDK, service, authentication requirement, and billing condition separately.
• Evaluation notes: Pin the SDK and target Skill versions, and verify API correctness and side effects.
• Safety checks: Cloud charges, persistent access, deployment, and other changes to external state.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: MIT (repository LICENSE). This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: Microsoft skills repository, Microsoft skills license

OpenAI Skills Catalog — OpenAI

Item ID: openai-collection. Kind: Collection. Source checked: 2026-10-03.

Distributor repository · Material to read

Individual path: null (entry point to a collection or framework).

• Purpose: An official catalog for finding Codex Skills.
• Expected use: Find a small candidate suited to your purpose among official examples.
• Limitations: Inclusion in the catalog does not guarantee behavior in your environment.
• Environment and dependencies: Each Skill's dependencies, supported clients, and external services.
• Evaluation notes: Record the path and version as well as the listed name, and evaluate one item at a time.
• Safety checks: Increased permissions and dependencies from bulk installation.

Version record: ref main, pinned commit null, declared version null, review status rolling-source-reviewed. A rolling reference whose public content was checked on the stated date; this is not a complete repository snapshot.

License statement: Check each Skill’s LICENSE.txt. This records the statement on the checked file or repository; dependencies can have different terms.

Evidence status: source-reviewed-not-benchmarked. Measured effect: null (not measured). Popularity rank: null (not surveyed).

Related primary sources: OpenAI Skills Catalog

Workflow

1. Choose two catalog candidates that fit your current work.
2. For each, write one expected improvement and one expected additional cost.
3. Record why you reject other candidates too.

Outputs

• Comparison cards for two candidates.

Quality checklist

• The use case is specific.
• You separate official status from guarantees of effectiveness.
• You have not added too many candidates.

Failure diagnosis

• Symptom: Recording success without observing the effect.
• Cause: Confusing expected judgments with actual outputs.
• Fix: Keep unexecuted work as not-run, clear measurement fields, and obtain raw outputs and logs before scoring.

Exercise: Write a short selection rationale

Follow the workflow above in order and create the stated deliverable.

Completion criteria: Do not choose solely by stars; show the required environment and the absence of measurements.

Status: not-run.

Source scope

Sources support feature descriptions and distributor statements in the text and catalog. They are not evidence of measured effects or popularity ranks. Verification dates record reading public sources, rather than publication or update dates. Rolling references such as main are not pinned experimental versions.

Offline experiment worksheet