Research / Coding AI Agent Skills Survey

Coding AI Agent Skills Survey

Coding AI Agent Skills Survey: Existing Vendor and OSS Prompt Libraries for Software Engineering Domains

2026-03-22 · agentic-ai tools-infrastructure knowledge-management software-engineering · medium · source → · wiki →
key claims
  1. The public market is split between packaging mechanisms and domain catalogs, with GitHub / Microsoft leading the catalog side through `awesome-copilot` while Anthropic and OpenAI more clearly lead the published skill-packaging pattern through reusable `SKILL.md` bundles (https://github.com/github/awesome-copilot ; https://developer.microsoft.com/blog/awesome-github-copilot-just-got-a-website-and-a-learning-hub-and-plugins ; https://raw.githubusercontent.com/anthropics/skills/main/README.md ; https://developers.openai.com/api/docs/guides/tools-skills)
  2. Anthropic and OpenAI independently document `SKILL.md`-based bundles with progressive disclosure, bundled resources, and reusable workflows, which indicates that packaged skills are now a first-party product pattern rather than a purely community convention (https://raw.githubusercontent.com/anthropics/skills/main/README.md ; https://support.claude.com/en/articles/12512198-creating-custom-skills ; https://developers.openai.com/api/docs/guides/tools-skills ; https://cookbook.openai.com/examples/skills_in_api)
  3. Cursor, Windsurf, Cline, and JetBrains confirm that persistent customization is now mainstream across coding assistants, but their published first-party materials currently emphasize configuration surfaces more than a single installable cross-domain engineering library (https://cursor.com/docs/rules ; https://docs.windsurf.com/windsurf/cascade/memories ; https://docs.windsurf.com/windsurf/cascade/skills ; https://docs.cline.bot/ ; https://www.jetbrains.com/help/ai-assistant/prompt-library.html ; https://www.jetbrains.com/help/ai-assistant/configure-project-rules.html)
  4. Curated community catalogs are where the widest public experimentation is visible today, because they aggregate many narrow artifacts across frontend, backend, API, testing, database, and notebook use cases that no single non-GitHub vendor catalog exposed at comparable breadth in this survey (https://raw.githubusercontent.com/PatrickJS/awesome-cursorrules/main/README.md ; https://raw.githubusercontent.com/VoltAgent/awesome-agent-skills/main/README.md ; https://github.com/github/awesome-copilot)
  5. `AGENTS.md` and `SKILL.md` are the safest medium-term formats to standardize around, because multiple vendors now acknowledge them and because they fit naturally into version control, code review, and repository governance workflows (https://agents.md/ ; https://developers.openai.com/api/docs/guides/tools-skills ; https://support.claude.com/en/articles/12512198-creating-custom-skills ; https://docs.windsurf.com/windsurf/cascade/skills)
  6. Public material is strongest where engineering work is repetitive and framework-bound, which is why .NET, Python backend, API design, testing, security, and database work surfaced repeatedly across the most credible public catalogs inspected here (https://raw.githubusercontent.com/github/awesome-copilot/main/docs/README.instructions.md ; https://raw.githubusercontent.com/PatrickJS/awesome-cursorrules/main/README.md ; https://developer.microsoft.com/blog/awesome-github-copilot-just-got-a-website-and-a-learning-hub-and-plugins)
  7. Public material is weakest where design quality depends on organisation-specific semantics or deep architectural trade-offs, which is why DDD, CQRS, Kafka ECST, data architecture, data modelling, classification, and semantic extraction remained thin across the strongest public catalogs (https://martinfowler.com/bliki/CQRS.html ; https://martinfowler.com/articles/201701-event-driven.html ; https://github.com/github/awesome-copilot ; https://github.com/PatrickJS/awesome-cursorrules ; https://airflow.apache.org/docs/)
  8. A responsible adoption policy is therefore triage-based: adopt faster in low-disagreement domains, adapt with local review in medium-context domains, and build internally for architecture-heavy domains where public catalogs remain sparse or overly opinionated (https://owasp.org/www-project-application-security-verification-standard/ ; https://spec.openapis.org/oas/v3.1.0 ; https://www.asyncapi.com/ ; https://12factor.net/ ; https://martinfowler.com/bliki/CQRS.html)

Research Question

What actively maintained, publicly available agent skills, prompt libraries, instructions files, and system prompts exist — from vendors such as Microsoft and from the Open Source Software (OSS) community — that cover the following software engineering domains: UI/UX (User Interface/User Experience) development, Python backend development, data architecture, data modelling, software architecture, SOLID (Single Responsibility, Open-Closed, Liskov Substitution, Interface Segregation, Dependency Inversion) software design, clean code, clean architecture, API (Application Programming Interface) design, DDD (Domain-Driven Design), .NET API development, Kafka event design for ECST (Event-Carried State Transfer), .NET architecture, .NET CQRS (Command Query Responsibility Segregation), database design, data pipelines, design, data visualisation, classification, semantic and concept extraction, unit testing, E2E (End-to-End) testing, integration testing, security architecture, security engineering, Apache Airflow, and Jupyter Notebooks — and which of these are suitable for adoption as-is to standardise and remove opinion from software development practice?

Findings

Executive Summary

Key Findings

  1. [inference] [high confidence] The public market is split between packaging mechanisms and domain catalogs, with GitHub / Microsoft leading the catalog side through awesome-copilot while Anthropic and OpenAI more clearly lead the published skill-packaging pattern through reusable SKILL.md bundles (GitHub `awesome-copilot` repository — GitHub / Microsoft ; Microsoft blog on Awesome GitHub Copilot website, learning hub, and plugins ; Anthropic skills README ; OpenAI Skills guide
  2. [inference] [high confidence] Anthropic and OpenAI independently document SKILL.md-based bundles with progressive disclosure, bundled resources, and reusable workflows, which indicates that packaged skills are now a first-party product pattern rather than a purely community convention (Anthropic skills README ; support.claude.com ; OpenAI Skills guide ; OpenAI `Skills in API` cookbook page
  3. [inference] [high confidence] Cursor, Windsurf, Cline, and JetBrains confirm that persistent customization is now mainstream across coding assistants, but their published first-party materials currently emphasize configuration surfaces more than a single installable cross-domain engineering library (Cursor rules documentation ; Windsurf Memories / Rules documentation ; Windsurf Skills documentation ; Cline documentation ; JetBrains AI Assistant prompt library documentation ; JetBrains AI Assistant project rules documentation
  4. [inference] [high confidence] Curated community catalogs are where the widest public experimentation is visible today, because they aggregate many narrow artifacts across frontend, backend, API, testing, database, and notebook use cases that no single non-GitHub vendor catalog exposed at comparable breadth in this survey (awesome-cursorrules README ; awesome-agent-skills README ; GitHub `awesome-copilot` repository — GitHub / Microsoft
  5. [inference] [high confidence] AGENTS.md and SKILL.md are the safest medium-term formats to standardize around, because multiple vendors now acknowledge them and because they fit naturally into version control, code review, and repository governance workflows (AGENTS.md standard site ; OpenAI Skills guide ; support.claude.com ; Windsurf Skills documentation
  6. [inference] [high confidence] Public material is strongest where engineering work is repetitive and framework-bound, which is why .NET, Python backend, API design, testing, security, and database work surfaced repeatedly across the most credible public catalogs inspected here (`awesome-copilot` instructions index — GitHub / Microsoft ; awesome-cursorrules README ; Microsoft blog on Awesome GitHub Copilot website, learning hub, and plugins
  7. [inference] [medium confidence] Public material is weakest where design quality depends on organisation-specific semantics or deep architectural trade-offs, which is why DDD, CQRS, Kafka ECST, data architecture, data modelling, classification, and semantic extraction remained thin across the strongest public catalogs (Martin Fowler — CQRS ; Martin Fowler — Event-Driven article with Event-Carried State Transfer (ECST) ; GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-cursorrules repository — PatrickJS ; Apache Airflow documentation index
  8. [inference] [high confidence] A responsible adoption policy is therefore triage-based: adopt faster in low-disagreement domains, adapt with local review in medium-context domains, and build internally for architecture-heavy domains where public catalogs remain sparse or overly opinionated (OWASP Application Security Verification Standard (ASVS) ; OpenAPI Specification 3.1 ; AsyncAPI home page ; 12-Factor App methodology ; Martin Fowler — CQRS

Domain Coverage Matrix

Domain Label Best public source(s) Coverage Adoption view
UI/UX development [inference] awesome-cursorrules README ; Anthropic skills README high Strong public examples exist; adopt selectively with local design-system constraints.
Python backend development [inference] awesome-cursorrules README ; `awesome-copilot` instructions index — GitHub / Microsoft high One of the strongest public domains; adoption is viable after project-specific testing and security alignment.
Data architecture [inference] Apache Airflow documentation index ; awesome-agent-skills README low Standards and tools exist, but public prompt/skill packs remain thin; build internally.
Data modelling [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-cursorrules repository — PatrickJS low Some database-related rules exist, but robust data-modelling prompt libraries did not strongly surface.
Software architecture [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; `awesome-copilot` instructions index — GitHub / Microsoft medium Good reviewer and blueprint artifacts exist, but wholesale adoption still needs local architecture principles.
SOLID software design [inference] awesome-cursorrules README medium Community rules exist; adopt only after aligning terminology and examples to house style.
Clean code [inference] awesome-cursorrules README ; JetBrains AI Assistant project rules documentation medium Easy to standardize, but many artifacts are opinionated and need local review.
Clean architecture [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-cursorrules repository — PatrickJS low-medium Public examples exist, but not a dominant, standards-grounded catalog.
API design [inference] `awesome-copilot` instructions index — GitHub / Microsoft ; OpenAPI Specification 3.1 ; AsyncAPI home page high Strong public support exists; adopt with OpenAPI / AsyncAPI as the normative layer.
DDD [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-cursorrules repository — PatrickJS low Public DDD prompt libraries did not strongly surface; do not standardize from public prompts alone.
.NET API development [inference] `awesome-copilot` instructions index — GitHub / Microsoft ; GitHub `awesome-copilot` repository — GitHub / Microsoft high Strong public guidance exists in vendor-backed Copilot assets.
Kafka event design for ECST [inference] Martin Fowler — Event-Driven article with Event-Carried State Transfer (ECST) ; awesome-cursorrules repository — PatrickJS low Conceptual standards exist, but prompt catalogs are sparse and not adoption-ready.
.NET architecture [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; `awesome-copilot` instructions index — GitHub / Microsoft medium-high Good public material exists, but enterprise architecture still needs local adaptation.
.NET CQRS [inference] Martin Fowler — CQRS ; GitHub `awesome-copilot` repository — GitHub / Microsoft low Public CQRS prompts are far weaker than the underlying architectural literature.
Database design [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-cursorrules README medium-high Good practical artifacts exist, but schema and operational decisions remain context-dependent.
Data pipelines [inference] Apache Airflow documentation index ; awesome-agent-skills README medium Supporting public material exists, but domain-specific prompt packs are not yet dominant.
Design [inference] Anthropic skills README ; awesome-cursorrules README medium Plenty of creative/design artifacts exist, but engineering-standard alignment is uneven.
Data visualisation [inference] awesome-cursorrules README ; Jupyter home page medium Good notebook- and plotting-oriented examples exist; adopt with local charting conventions.
Classification [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-cursorrules repository — PatrickJS low Public catalogs did not surface strong, high-trust classification prompt packs.
Semantic and concept extraction [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-agent-skills README low A few adjacent semantics-oriented artifacts exist, but coverage is shallow and fragmented.
Unit testing [inference] GitHub `awesome-copilot` repository — GitHub / Microsoft ; awesome-cursorrules README high One of the most mature public prompt domains; adoption is viable after aligning to test framework choices.
E2E testing [inference] awesome-cursorrules README ; GitHub `awesome-copilot` repository — GitHub / Microsoft medium-high Strong Cypress / Playwright examples exist, though they still need environment-specific adaptation.
Integration testing [inference] awesome-cursorrules README ; GitHub `awesome-copilot` repository — GitHub / Microsoft medium-high Good public artifacts exist, especially around web stacks and migrations.
Security architecture [inference] OWASP Application Security Verification Standard (ASVS) ; `awesome-copilot` instructions index — GitHub / Microsoft medium Public prompts help, but the authoritative baseline is the security standard, not the prompt file.
Security engineering [inference] `awesome-copilot` instructions index — GitHub / Microsoft ; OWASP Application Security Verification Standard (ASVS) high Security-oriented prompts exist, and they can be anchored to ASVS for stronger standardization.
Apache Airflow [inference] Apache Airflow documentation index ; awesome-agent-skills README low Official domain docs are rich, but prompt libraries are not yet strong enough for broad as-is adoption.
Jupyter Notebooks [inference] Jupyter home page ; awesome-cursorrules README ; Cline documentation medium Public notebook-oriented rules exist, but they are mostly community examples rather than vendor-backed standards.

Assumptions

Analysis

Risks, Gaps, and Uncertainties

Open Questions

sources

← newerApplied context engineering
older →Technology Capability Models

Connected items

Loading…

View full knowledge graph →