Skip to content

Repository files navigation

🧠 AI Agent Skills Catalog

A curated collection of master-level agent skills, guidelines, and reference materials for Google Antigravity, Claude Code, OpenAI Codex, Aider, and Cline.

🌐 Browse the interactive catalog page: hiadamhere.github.io/skills β€” explore the MSAF Architect skill visually, right in your browser.


πŸ“‚ Available Skills

πŸ›οΈ msaf-architect (Microsoft Agent Framework C# Architect)

Master architecture guidelines, execution models, and API mappings for building multi-agent systems using the Microsoft Agent Framework.

Important

API Ground-Truth Alignment Tutorials and LLM training data describe Microsoft Agent Framework APIs that were never shipped (e.g. a WorkflowSuspendedException suspend pattern) and cause compiler failures.

Every claim in this skill is verified against the actual v1.10.0 through v1.17.0 assemblies from NuGet β€” reflection surface extraction plus compile tests against the pinned packages. Reference documents carry per-version verification stamps, and an automated gate rejects any unverified API identifier before release. The latest guidance targets v1.17, whose public API surface is byte-identical to v1.16 (Magentic manager prompt overrides and response language, both behind the MAAI001 experimental diagnostic), on top of v1.15 (latest-checkpoint resolution and the blockOnPendingRequest streaming overload).

Both v1.15 and v1.16 are purely additive by mechanical surface diff, and v1.17 changes nothing at all β€” v1.14 code compiles unchanged throughout β€” so the traps are in adoption, not migration: MagenticPromptOverrides properties are init-only, GetLatestCheckpointAsync returns a nullable CheckpointInfo?, and the v1.16 prompt API is still a compile error until MAAI001 is suppressed.

The interactive visualizer includes a dedicated v1.14 Migration Map showing the unchanged Workflows layer and the compile-verified agent-mode, message-injection, tool-approval, and approval-middleware replacements, plus a v1.15 & v1.16 Additions view covering latest-checkpoint resolution, the pending-request streaming choice, the seven Magentic prompt slots, and the MAAI001 gate.

πŸ“± maui-engineer (.NET MAUI Architect)

Version-aware architecture and planning guidance for .NET MAUI apps: the decisions to get right before writing feature code β€” target-platform and SDK/workload/package strategy, project layout, MVVM/DI, navigation, state and offline data, platform-abstraction boundaries, performance budgets, accessibility, and publish/signing constraints. It resolves the real SDK, target frameworks, workloads, and packages first, and requires platform/runtime evidence instead of treating one successful desktop build as cross-platform proof.

The navigation guidance covers choosing one primary model, deep links that survive a cold start, and the DI lifetimes behind the classic page/view-model leak. The performance guidance deliberately ships no threshold numbers β€” it covers what to budget and how to measure and hold it, because a budget that did not come from your own baseline on your own target hardware is decoration.

The bundled environment inspector produces a reviewable JSON snapshot of the project's toolchain. This is a methodology skill: API specifics and platform behavior are resolved from the target project and official Microsoft docs, not asserted from a pinned DLL surface.

🎨 spectre-console (Spectre.Console Terminal-UI Expert)

Verified, version-matched guidance for building rich .NET terminal UIs with Spectre.Console: tables, panels, trees, markup & color, live displays (Status/Progress/Live), and interactive prompts β€” plus the terminal-gating discipline that keeps interactive features from hanging in CI or hosted contexts.

Important

API Ground-Truth Alignment Every type, method, and property is verified against the actual Spectre.Console v0.57.2 assemblies (Spectre.Console, Spectre.Console.Ansi, Spectre.Console.Testing) β€” reflection surface extraction plus compile tests against the pinned package. The same automated gate rejects any unverified API identifier before release.

🧩 microsoft-extensions-ai (Microsoft.Extensions.AI β€” .NET LLM abstractions)

Verified, version-matched guidance for Microsoft.Extensions.AI, .NET's unified LLM layer: IChatClient calls and streaming, ChatOptions, tool/function calling, structured output, embeddings, and the middleware/DI pipeline β€” so generated code targets the current GA API instead of renamed preview members.

It covers the two mistakes that fail silently rather than loudly: declaring ChatOptions.Tools without the UseFunctionInvocation() middleware that actually executes them, and middleware ordering β€” where the cache sits relative to function invocation decides whether a cache hit skips the entire tool loop.

Important

API Ground-Truth Alignment Every type, method, and property is verified against the actual Microsoft.Extensions.AI v10.8.1 assemblies (Microsoft.Extensions.AI, Microsoft.Extensions.AI.Abstractions) β€” reflection surface extraction plus compile tests against the pinned package. Training data mixes the old preview API with GA (e.g. the renamed GetResponseAsync); the same automated gate rejects any unverified API identifier before release.

πŸ”Œ mcp-sdk (Model Context Protocol SDKs)

Verified guidance for building MCP servers with the official SDKs in C#, TypeScript, and Python: defining tools, wiring the stdio transport, and standing up a minimal server with the correct, version-matched API. MCP postdates most training data, so models invent its shapes β€” this anchors them to the real SDK.

Each language carries a different trap, and each is compile- or type-checked here: in Python, from mcp.server.fastmcp import FastMCP β€” the import in nearly every MCP tutorial β€” does not exist in mcp 2.0.0 (the class is now MCPServer); in TypeScript, the widely-shown server.tool() is @deprecated in favour of registerTool, and import paths need .js suffixes; in C#, the WithHttpTransport() / MapMcp() wiring most samples show lives in a separate package and does not compile against the pinned one.

Important

API Ground-Truth Alignment All three language references are verified against their own pinned SDK: C# against the real ModelContextProtocol v2.0.0-preview.3 assemblies (reflection surface extraction + compile tests), TypeScript against @modelcontextprotocol/sdk 1.30.0 shipped type definitions (type-checked with tsc --noEmit), and Python against mcp 2.0.0 shipped type information (type-checked with pyright). The same automated gate rejects any unverified API identifier before release.


πŸš€ Installing the Skills

Three ways to install β€” pick whichever fits your setup.

1. With spm (recommended) β€” installs into every coding agent

spm (Skill & Plugin Manager) is a package manager for AI-agent skills. It installs this catalog into all your coding agents β€” Claude Code, Codex, Cursor, Cline, Aider, GitHub Copilot, and more β€” from the command line.

Install spm (choose one β€” no repo clone required):

npm install -g @hiadamhere/spm     # Node β€” self-contained, no .NET required
dotnet tool install -g spm          # …or the .NET global tool

Install every skill from this catalog:

spm install hiadamhere/skills

Don't want a global install? Run it once with npx: npx @hiadamhere/spm install hiadamhere/skills

Or subscribe to the catalog to keep it updated and install skills by name:

spm catalog add hiadamhere https://github.com/hiadamhere/skills.git
spm catalog sync
spm list                      # browse available skills
spm search table              # find one
spm install spectre-console   # install a single skill
spm install maui-engineer     # install the MAUI architect
spm uninstall spectre-console # remove it

2. One-line remote install β€” no package manager needed

No Node or .NET? A self-contained script deploys all catalog skills to your user-level agent config folders (~/.agents/skills for Codex/Gemini, ~/.claude/skills for Claude Code):

On Windows (PowerShell):

irm https://raw.githubusercontent.com/hiadamhere/skills/main/install.ps1 | iex

On macOS/Linux (Bash):

curl -fsSL https://raw.githubusercontent.com/hiadamhere/skills/main/install.sh | bash

3. Local / workspace install (clone) β€” copy or symlink, per-skill, custom scope

Clone the repo and run the installer with flags for finer control:

  • Mode: Copy files or Symlink (recommended for automatic updates via git pull).
  • Scope: Global (user-profile) or Folder (workspace-level β€” copies into <folder>/.agents/skills/ and <folder>/.claude/skills/).
  • Skills: Multi-select specific skills; interactive runs present a menu with ALL as the first choice.

Windows (PowerShell):

# Interactive (prompts for mode, scope, target folder, and skills)
.\install.ps1
# Non-interactive copy of specific skills to a workspace
.\install.ps1 -Mode Copy -Scope Folder -Path C:\MyProject -Skills msaf-architect
# Uninstall
.\uninstall.ps1 -Scope Folder -Path C:\MyProject -Skills msaf-architect

macOS/Linux (Bash):

# Interactive
./install.sh
# Non-interactive symlink of specific skills to a workspace
./install.sh --mode link --scope folder --path /path/to/my-project --skills msaf-architect
# Uninstall
./uninstall.sh --scope folder --path /path/to/my-project --skills msaf-architect

Distributed under the MIT License.

About

No description, website, or topics provided.

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages