ai-plugins-and-skills

AI-Plugins-And-Skills

Cross-platform AI coding assistant skills following the Agent Skills open specification — they work across Claude Code, Copilot CLI, and VS Code Copilot.

Supported Platforms

Skills

Skills are situational knowledge files an AI coding harness loads when their description matches what you're doing — repeatable processes, agent-callable tools, or domain knowledge. See skills/README.md for the longer intro and per-harness install guides.

Review Helpers

Local, adversarial review workflows — first the plan (or spec), then the code commits.

For planning: run plan-send-review in your authoring session, bounce to a second session (or a different agent) for plan-intake-review, then return to the original session and run plan-receive-review to read the feedback.

For code commits, pick a path: one large ad-hoc review with code-send-review, or slice-and-dice a branch into multiple slices with code-review-plan-create that will then get reviewed in parallel and rolled into a single pre-merge action list via code-synthesize-reviews.

See documentation/workflows/ for the full plan-review and code-review workflows — state machines, directory layouts, and the spec-driven coding workflow that ties them together.

Tools

When a workflow is known up front, a curated tool almost always beats ad-hoc agent queries on tokens. The skills below ship both Python and single-file C# variants; the Python ones just run, but the C# ones need a working .NET environment:

1. .NET 10 SDK installed
2. Azure CLI installed — or any other credential source DefaultAzureCredential will pick up
3. Authenticated via az login

Language Helpers

Plenty of general-purpose language helpers exist already, so no need to duplicate. These are about keeping and enforcing standards. If you already lean on EditorConfig and StyleCop for your C# work, the csharp skill slots in naturally. The Python skill bakes in the usual suspects — ruff, pyright, pre-commit hooks — for enforcing good script hygiene.

Workflow Composers

Where the review helpers above handle one review, these decide how much process a piece of work deserves in the first place — and then hold you to it. Start with spec-tier to route a task to a Low/Medium/High spec depth by stakes, let plan-workflow move the resulting document through the planning/ kanban while plan-writing-syntax governs what goes on the card — the frontmatter, the per-kind status vocabularies, and cross-references written as names so they survive the document moving column — and use test-filter-development instead of plain TDD when an agent is writing the code — it makes the test filter load-bearing by demanding captured RED output and a seeded mutant rather than a claim that the tests were written first.

Four more keep the board honest once work is underway: plan-sweep-blockers moves plans that have quietly stalled out of the in-progress column and records what each one is waiting on, plan-clear-blockers turns that blocked column into an agenda for the hour when the scarce device, environment or person is finally available, reminders-set files the tasks a person must actually do — the ones no test or gate will ever catch — and plan-record-review writes down what a review decided once you have evaluated it, so the rejections survive the conversation that produced them.

Domain Skills

Stack-specific expertise the agent loads when it recognizes the stack — framework idioms, current API surface, and the mistakes that stack in particular invites. Unlike the language helpers above, these are not about enforcing a house style; they are about knowing the material.

Curriculum — a separate, opt-in plugin

The curriculum-* skills are a seven-stage pipeline that turns source documentation into course definitions and course artifacts: discover → classify → extract → design → govern → generate, with the governance gate running twice (once on the definition, once on what was rendered from it).

They ship as their own plugin and are deliberately not part of the main one. Every installed skill's description sits in your context for the whole session whether you use it or not, and this pipeline is inert unless you have a curriculum workspace for it to read. Making you opt in keeps that cost off everyone else.

claude plugin install curriculum@ai-plugins-and-skills-ai-standards

Add --scope project to enable it only inside the repo that holds your curriculum workspace, rather than globally.

Start at curriculum-orchestrate — it documents the workspace layout the whole family expects, and curriculum-discover carries a sources.example.yaml to copy.

Contributing

See HOW-TO-DEV.md for development setup and contribution workflow.

Architecture

See ARCHITECTURE.md for cross-platform design and directory structure.