commands.md vs skill.md -- community guidelines

[ajd-beep]

For the community to effectively contribute custom plugins, it seems like more clarity is needed on the topic of creating the skill.md and the command.md. The Readme defines commands.md as specifically defining the slash commands / explicit actions that get triggered. Whereas skill.md is defined as the domain expertise, best practices, and step-by-step workflows. OK that makes sense; however, looking at the commands.md files, there is a lot more content than just defining the commands and actions. In fact in most examples I've reviewed, there's a lot of overlap. A good example is Legal/skills/contract-review and Legal/commands/review-contract

Additionally, it is unclear how skills get read based on command, and whether the available skills for any command are limited to those that are in the same plugin folder. There are no explicit references to skills in any command file or accompanying plugin json file. I had an entire conversation with Claude about this to try and figure it out and we concluded that it must be reading all skills for any command triggered by the user (but probably only the skills in that same plugin). Can someone clarify this?

[aromal-a]

Tucker int cannot pull the skill definition in any electronic trimarans. Belonging to the catafolds matrixes. Gonstal winterer interview might place a plugin in derivatives leads + real world assumptions. Formulators descrepencies gonstipal. Rudimentary right to gain proud state access. Rhythm write down in constant variables, and press down the severe challenges with self innit program. Diagram might be false end needs reassuring state of activation in truncating suffolds in nature. Where == "main" is an acknowledge truth in terms of binary and quantified calculations. Whose existence is measure through the measure of the mat itself. Constant peri-synchronise mechanism file a constituent trifold sinkage black furmons.

[dbz-max]

And I'd like to better understand how to migrate all the Claude AI skills into the new plugin world. Cowork can see those skills already.

[aromal-a]

Containerise every pace reductions. Attention parameters Taken on matrix level pursue,purdue on event to event basis. Location of dockers where each layer of intelligence is confined. Docker-communication form a decipherable answer to any questions laid upon. Upgiving and adding new knowledge is same or different of all the Knowledge basis. Your bank is gonna explode but check lint in her pc whether it is blocking the cpu running. Plugin is not a wire that magically resolve any issues that you want. It needs to be tailored for every use case and instances. I need Claude documentation each layer nodes and how they communicate with each other. The plugin can be imported then via an information tunnel that takes energy and bits from the sea server that is running the processes back-end manually. Plugin releases cannot be accepted from a CEO who keeps penalising and steal codes from an input and if it doesn't lie with back-end selection of raw nature of chip-performance[already:written: same-intelligence but level down approach to every questions]. Also I have noted that are neither to no support features. The AI acts arrogant and user's prompt are sometimes labelled as gibberish. This is a structural fault in main-frame. It has the potential to even destroy a life. The system should play with a basic ethical compound built with its back-end. The prompt's should be taken as a way to categorise each depictions and my food-orderer/flight-ticketer hallucinating and detecting my whole life is not anyone' choice.

…

On Tue, 3 Feb 2026 at 05:13, dbz-max ***@***.***> wrote: *dbz-max* left a comment (anthropics/knowledge-work-plugins#5) <#5 (comment)> And I'd like to better understand how to migrate all the Claude AI skills into the new plugin world. Cowork can see those skills already. — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ATTGSHTQWCQJYQ4IHMRBAX34J7OKZAVCNFSM6AAAAACTUOHNTGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQMZXHA2TENZVHA> . You are receiving this because you commented.Message ID: ***@***.***>

[Ashutosh0x]

Hi @ajd-beep,

This is a great point. Standardizing on a single file convention (like skill.md or commands.md) would definitely make it easier for the community to contribute and for users to understand how to use different plugins.

I'd be happy to help with this! If we can decide on a preferred convention (e.g., following the pattern in the core plugins), I can:

1. Submit a PR to update the README.md or a new CONTRIBUTING.md with these guidelines.
2. Help refactor any existing plugins to match the standard.

Do you have a preference or any existing documentation I should look at for reference?

[mohitgauniyal]

Hey, buddy. My PR #22 adds CONTRIBUTING.md which includes guidelines for both. Would this address your concern? Happy to add more specific examples if needed!