Curated AI Skills
High-frequency skill patterns, packaging models, and refresh sources that still matter in April 2026
How a skill works
Intent match
The agent compares the user request against skill descriptions, triggers, or explicit slash invocations to decide whether a reusable skill should activate.
Context loading
Only the skill entrypoint and the small set of supporting files needed for the current task are loaded, which keeps long reference material cheap until it is relevant.
Workflow execution
The skill can call scripts, tools, prompts, or graph nodes to apply a repeatable playbook such as code review, triage, research, or release preparation.
Shaped output
A good skill constrains the final output with clear quality bars, formatting rules, and completion criteria so the result is predictable across repeated runs.
How we keep the list fresh
The source of truth stays versioned in Data\SeedData\skills.json. We refresh the page by re-reviewing official docs, reference repositories, and the skill patterns that continue to appear across public agent ecosystems.
What triggers a refresh
- A major ecosystem ships a new skills feature, manifest format, or packaging convention.
- A workflow pattern starts appearing across several ecosystems instead of a single vendor stack.
- Official docs or starter repositories deprecate an older skill shape or recommend a new one.
Source of truth
Refresh checklist
- Review official skills, agents, or orchestration docs before editing any curated entry.
- Validate every documentation and repository link with Scripts\refresh-skills-links.ps1.
- Update the JSON review label, architecture notes, and any affected skill patterns.
- Rebuild the site and visually verify the Skills page before publishing.
Typical file architecture
There is no single universal skill package, but the shape repeats across ecosystems: a short manifest or instruction entrypoint, optional scripts or tools, supporting references, and a small amount of test or example material.
Claude Code
Folder-discovered skills built around a required SKILL.md entrypoint and optional supporting files.
.claude/skills/code-review-gate/ |- SKILL.md |- scripts/ | |- validate-checklist.ps1 |- references/ | |- review-rules.md |- assets/ | |- comment-template.md `- examples/ `- sample-output.md
Best for local or repository-scoped skills where the agent auto-discovers reusable playbooks from known skill folders.
OpenAI Agents SDK
Reusable Python packages where tool schemas, guardrails, and prompts define the skill boundary.
agents/release-notes-builder/ |- agent.py |- tools.py |- schemas.py |- prompts/ | `- system.md `- tests/ `- test_release_notes.py
A good fit when the skill is code-backed and needs explicit tools, structured outputs, tracing, or handoffs.
Claude Code skill patterns
High-frequency repo-local skills built around SKILL.md playbooks and optional helper scripts.
Code Review Gate
A reusable review playbook that applies repo rules, surfaces blocking issues first, and standardizes review output.
Frequently shared as a project skill because every team wants the same review bar without pasting a full checklist into each session.
Repo Audit Playbook
A structured skill for auditing architecture, dependencies, security notes, and operational drift inside an existing codebase.
Common in platform teams and consulting workflows where the same repository audit needs to be rerun across many projects.
Test Generation Workflow
A skill that turns source files and conventions into targeted unit or integration test plans with a fixed quality checklist.
Popular whenever teams need consistent test style, fixture usage, and coverage expectations across repositories.
OpenAI Agents SDK patterns
Code-backed agent skills that package tools, prompts, schemas, and guardrails for production workflows.
Browser Research Agent
A web-aware research skill that collects sources, extracts signals, and returns a structured brief with citations.
One of the most repeated agent patterns because it combines tool use, guardrails, and structured output in a single reusable package.
Retrieval QA Skill
A retrieval-augmented skill that blends vector search, citations, and answer validation for internal knowledge bases.
Persists because most real agent deployments still need repeatable grounding and auditable answers.
PR and Issue Triage Agent
A triage skill that clusters repository events, assigns urgency, and drafts next actions for maintainers.
Widely reused where teams want tool-backed automation but still need explicit schemas and approval checkpoints.
LangGraph skill patterns
Stateful orchestration patterns for skills that branch, retry, or require approvals across several steps.
Incident Commander Graph
A graph-driven response skill that detects signals, routes tasks, and produces a live incident summary with state tracking.
High value whenever the skill must survive retries, branching, or partial progress rather than one linear prompt.
Docs-to-Code Planner
A planning skill that reads specs or documentation, builds intermediate state, and emits an implementation sequence.
Useful when teams want planning, validation, and execution to stay connected instead of living in separate prompts.
Stateful Support Router
A routing skill that classifies requests, preserves conversation state, and hands work to the right downstream agent or human.
Common because routing skills usually need durable state and explicit branch logic once the happy path breaks.
Microsoft Skills
Real SKILL.md-based entries curated from the public microsoft/skills repository.
applicationinsights-web-ts
Instrument browser and web apps with the Application Insights JavaScript SDK for RUM, dependencies, exceptions, custom events, and browser-side GenAI traces correlated with backend observability.
Useful when teams need production-grade browser instrumentation and want agent guidance tied directly to the Azure SDK skill catalog.
continual-learning
Guide and hook-based skill for implementing continual learning in coding agents through memory scoping, reflection patterns, and cross-session persistence.
Notable because it adds practical long-term learning to coding agents with a lightweight hook and a clear project-versus-global memory split.
frontend-design-review
Review and create production-grade frontend interfaces with strong design quality, accessibility, and design-system compliance checks.
High-value for teams that want memorable UI reviews backed by an explicit rubric instead of generic frontend feedback.