Feature request: official Claude Skill for Effect-ts

[PaulRBerg]

Note

Ended up rolling my own skill until they offer an official one. See the comment below.

What is the problem this feature would solve?

Effect-ts has a steep learning curve.

You guys have done an amazing job at making the official docs easily accessible via context7 MCP and llms.txt, but accessing these resources can be quite expensive in terms of AI token consumption. This is in part due to the large size of the Effect framework/ docs.

What is the feature you are proposing to solve the problem?

Provide an official Claude Skill (or Agent Skill) for Effect-ts, which would:

1. Be installable via a plugin marketplace (e.g. analogous to this Playwright Skill)
2. Enable lean AI-assisted development.
3. Serve as a 'compressed' documentation. The skill should not be a 1-to-1 match with the docs. It should be a pared-back version with essential tips and tricks and high-level guidance.

Example

You can see an example of our custom-built Tailwind CSS skill here.

Tip

You can use Anthropic's official plugin-development skill to create the first version of your skill.

[mikecfisher]

Yes would absolutely love to see this.

[williamlmao]

I would love this!

[PaulRBerg]

Until the team publishes an official skill, I will share the Claude Skill we wrote for Effect-ts. It's pretty robust and we've used to great success - all of us in the team (the humans) are Effect-ts novices (and generally not proficient in functional programming), but our AI agents have mastered Effect-ts using the skill.

https://github.com/PaulRBerg/agent-skills/blob/bb313ecf7b58bdb9e0c5c43933a64e078b5befe8/skills/effect-ts/SKILL.md

You can install the plugin like this:

npx skills add PaulRBerg/agent-skills
# and then follow the CLI prompt to install the `effect-ts` skill 

And then, activate the skill by telling Claude to use effect-ts. Here's a prompt I use often:

Toggle to see prompt

<your_role>

You are a senior TypeScript/React engineer with deep Next.js experience and expert-level knowledge of functional programming and Effect-ts.

</your_role>

<task>

Your task is to {{ TODO }}

</task>

<skill_activation>

To complete this task, activate the `effect-ts` skill.

</skill_activation>

If the installation command above doesn't work, just copy-paste the skill content from here.

Besides the raw Effect-ts guidance, the skill also provides reference files for @prb/effect-next (my Next.js library for Effect), and effect-atom, as well as other things like testing.

Warning

The skill is comprehensive, but obviously experimental. Use at your own risk.

Note

You will have to clone the official Effect-ts/effect repository to your ~/.effect directory. This is because Effect is a huge library and checking out local code is faster than using context7 MCP/web search. Also, you should add ~/.effect to your permissions.additionalDirectories setting in .claude/settings.json.

[williamlmao]

Thanks @PaulRBerg, this is excellent!

[LeonMueller-OneAndOnly]

The skill of @PaulRBerg is very nice - all tough also very project opinionated due to requiring a local effect repository copy at a specific path and including nextJS specific know how.

a more general effect ts i found here: https://smithery.ai/skills/ojowwalker77/effect-ts-expert

[PaulRBerg]

Thanks @LeonMueller-OneAndOnly.

The local Effect-ts copy makes agents smarter and keeps the API in continuous sync (assuming you run git pull often; I have launchd task to do that every day).

Regarding NextJS - yeah, that's an opinionated design. I used to have it as a separate skill but 100% of my use is in Next projects and I noticed less churn/better UX with a single, unified skill.

Feel free to fork my skill and adapt it to your needs if this is not relevant to you.

[LeonMueller-OneAndOnly]

@PaulRBerg Providing an agent with the funcitonality to search a codebase is indeed the best return on invest you can do!

Have you tried: https://btca.dev/? I love to use that, in order to avoid context bloat for answering a specific question about a codebase -> think of it as a toolcall that spawns a subagent to answer a specific question of a codebase and then returns the summarized answer for it to the main agent (the one you are typing with)

[PaulRBerg]

I haven't tried it; I find Claude Code's built-in context management pretty darn good.

I recommend chatting offline about 3rd party tools in order to keep this issue thread lean and let the Effect team extract the essentials out of our feedback.