microsoft fluidframework
OFFICIALLABSCO SUMMARY
Ten of the twenty-eight are specific to shipping code in this exact repository: fluid-pr and fluid-pr-guide compose the pull request, fluid-release drives version bumps and changelogs, ci-readiness-check and api-changes catch CI failures and release-tag requirements before a push, policy-check validates package compliance, updating-noridocs and trigger-pipelines-for-copilot-pr cover documentation and Copilot-PR pipeline triggers, and ff-oce-dashboard and ff-oce-kusto query Fluid's own Office telemetry — Office_Fluid_FluidRuntime Kusto tables, session lookups by docId, Loop and Whiteboard error monitoring.
The other eighteen are general software-engineering practice, not particular to Fluid at all: brainstorming, test-driven-development, systematic-debugging, root-cause-tracing, receiving-code-review, and writing-plans read the same whether or not the codebase underneath is Fluid Framework. All of it runs under a named agent workflow the skills call 'Nori' — starting-nori is the skill every session is supposed to load first, and using-skills tells the agent to read it before any other work.
READ THE FULL ANALYSIS
Two of the twenty-eight are not usable outside Microsoft at all, regardless of what account key a reader holds. Eleven need nothing beyond installing them and ten want a local tool already on the machine, but six want an account key on file and one talks to an MCP server — and ff-oce-dashboard and ff-oce-kusto specifically require live access to Microsoft's internal Office Fluid telemetry, which no account key a reader can obtain on their own will get them into.
The README never mentions skills at all. It is Fluid Framework's own build-and-contribute guide for the client, server, and tooling packages that make up the monorepo — it never names Nori, ff-oce, or any of the 28 skills — so everything above about how they are actually used comes from reading the skills' own descriptions, not from Microsoft's documentation of them.
ALSO IN THIS PACKAGE
ff-oce
Fluid Framework On-Call Engineer agent and tools
nori
Senior SWE profile — skills, subagents, and commands
ff-oce
Assists engineers on the Fluid Framework Client OCE rotation.
nori-change-documenter
Creates documentation about changes to a codebase. Call the nori-change-documenter whenever you change the codebase.
nori-code-reviewer
Use after making code changes but before submitting a PR as a final check. Explain what the purpose of the change is when using this agent.
nori-codebase-analyzer
Analyzes codebase implementation details. Call the nori-codebase-analyzer agent when you need to find detailed information about specific components.
nori-codebase-locator
Locates files, directories, and components relevant to a feature or task. Call `nori-codebase-locator` with human language prompt describing what you're looking for.
nori-codebase-pattern-finder
nori-codebase-pattern-finder is a useful subagent_type for finding similar implementations, usage examples, or existing patterns that can be modeled after. It will give you concrete code examples based on what you're looking for!
nori-initial-documenter
Creates documentation about a codebase. This agent creates the initial documentation for the codebase; use this agent when you want to create documentation and no existing documentation is present.
nori-web-search-researcher
Do you find yourself desiring information that you don't quite feel well-trained (confident) on? Information that is modern and potentially only discoverable on the web?
nori-knowledge-researcher
Research specialist for querying the Nori knowledge base to find relevant context and information. Executes focused, iterative research with clear stopping criteria.
WHAT'S INSIDE
15 showing · 15 totalNothing else to set up — install it and go.
brainstorming
A way of working out what to build before any code is written: you get asked one question at a time rather than handed an answer, and nothing moves on until you agree.
building-ui-ux
Designs a screen with you rather than for you — you see the options, you say what to change, and the winner gets hooked up to the real data at the end.
creating-debug-tests-and-iterating
When a bug will not reproduce, this drops the existing tests entirely and goes after the real running program instead, adding logging on every attempt until the cause is visible.
finishing-a-development-branch
The end-of-job routine for a piece of work — tests run, code tidied, a second pair of eyes on it, and the request to merge opened and watched until the checks pass.
handle-large-tasks
A job too big to hold in one go gets cut into small pieces, each handed to a helper with a test that proves when it is done, then checked that the pieces fit together.
receiving-code-review
Every review comment gets verified against the codebase before it's touched, then fixed and tested one at a time and pushed as a batch. Suggestions that don't hold up get factual technical pushback instead of automatic agreement.
root-cause-tracing
You trace a deep bug backward through the call chain, from where it surfaces to where the invalid data or bad state actually originated, and fix it there instead of at the symptom.
systematic-debugging
The discipline of not guessing at a bug — nothing in the code gets changed until there is evidence for what is actually causing it.
test-driven-development
Tests first, code second — because a test you never watched fail might not be testing anything at all.
testing-anti-patterns
The mistakes that leave a test looking fine while proving nothing, or quietly push test-only code into the real product — with the question to ask before keeping one.
using-git-worktrees
A second, separate copy of the project to work in, so an experiment never disturbs the one you already had open — set up, installed and tested before anything starts.
using-screenshots
The assistant can in fact see your screen — this has it take the picture itself and then look at what is on it, rather than replying that it has no way to.
using-skills
A standing rule against the assistant going from memory: what it thinks it remembers about how a job is done does not count, and the written instructions are opened again every time.
webapp-testing
Chases a bug in a website by driving a real browser against the real running site, adding logging each time round, and ends by showing you the fix happen on screen.
writing-plans
Once it is settled what to build, this turns that into a written plan detailed enough for somebody who has never seen the codebase to work from, testing included.
HOW TO GET IT
npx skills add microsoft/fluidframeworknpx skills add microsoft/fluidframework --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.