2 months ago
Railway Agent Skill
Making Railway coding-agent-friendly has been a priority for us. We've already shipped the Railway MCP Server, which works well — but now we're adding something new: Agent Skills.
Agent Skills are a simple, open format for giving agents new capabilities. They're folders of instructions, scripts, and resources that agents can discover and use to work more accurately and efficiently. Unlike MCP, there's no server to run and they consume less of the LLM's context window.
We've built the Railway Skill so your coding agent can spin up projects, deploy services, and manage infrastructure without you ever leaving your editor.
If you’re using Claude Code, you can install the Railway plugin
```
claude plugin marketplace add railwayapp/railway-claude-plugin
claude plugin install railway@railway-claude-plugin
```
Every major coding agent (e.g. Cursor, Codex, OpenCode, etc.) supports the Agent Skills spec. so you all you need to do is clone or copy the skill directory into your project, and you're set.
We're actively improving the Skill and we'd love to hear what's working and what's not. You can use this thread to share your feedback
0 Threads mention this feature
7 Replies
2 months ago
Thanks for this!
One thing I'm really missing is allowing Claude code to access my Railway logs to debug issues.
Imagine this flow:
An issue arrives from Sentry (or is reported by a user).
Claude can check the logs around the specified time.
Claude can check the code.
That would be wonderful.
If already possible somehow let me know, I considered the CLI but there is no time based filter as far as I've understood?
2 months ago
It didn't work for me, but Claude then figured out how to make it work using this command instead:
claude plugin install railway@railway-skills
a month ago
would be fun if we allow the agent to "optimize my hosting cost this month" and it will search the usage, the load and then suggest it for you. and suggest you deployment templates and open source stuff available for you problem
railiway already allow my org to not hire ci/cd .. I hope I t can allow my devs to talk to their infra using natural language with business and/or technical context
a month ago
I'm sorry, i really don't like the new skills -_-
They work, but they:
1. clutter my skills dir with 12 skill ( i know, probably for the agent to better use the things it needs instead of needing everything, but still... Should't there be a more optimized solution for this?
2. THE SKILL ACTIONS ARE NOT NAMESPACED!!! -_-
Example:
One of your skills is called `railway-new`. but since the name of this skill in frontmatter, it shows in opencode +oh-my-opencode as `/new` (because oh-my-opencode loads all skills as slash commands. This shadows the /new action for creating a new chat session.
I know, my example is very specific, but:
- all your skills have this problem: railway-status only is "status" in frontmatter, railway-service is only "service", etc.
- if agent harnesses decide that showing skills as slash commands is great (for better visibility), its confusing as hell.
3. Installing them via the script is nice at first, but then you think: ok, how do i backup my skills config?
I could symlink my skills somewhere, but this falls apart quickly.
I know, this is an issue of the ecosystem not having standardized on a skill marketplace / standardized skill location on disk.
But it still sucks.
So for now, i'll teach my agent how to manage those skills in my dotfiles repo and maybe go and simply wrap your cli with a SKILL file, which is probably enough for now.
bjesuiter
I'm sorry, i really don't like the new skills -_- They work, but they: 1. clutter my skills dir with 12 skill ( i know, probably for the agent to better use the things it needs instead of needing everything, but still... Should't there be a more optimized solution for this? 2. THE SKILL ACTIONS ARE NOT NAMESPACED!!! -_- Example: One of your skills is called `railway-new`. but since the name of this skill in frontmatter, it shows in opencode +oh-my-opencode as `/new` (because oh-my-opencode loads all skills as slash commands. This shadows the /new action for creating a new chat session. I know, my example is very specific, but: - all your skills have this problem: railway-status only is "status" in frontmatter, railway-service is only "service", etc. - if agent harnesses decide that showing skills as slash commands is great (for better visibility), its confusing as hell. 3. Installing them via the script is nice at first, but then you think: ok, how do i backup my skills config? I could symlink my skills somewhere, but this falls apart quickly. I know, this is an issue of the ecosystem not having standardized on a skill marketplace / standardized skill location on disk. But it still sucks. So for now, i'll teach my agent how to manage those skills in my dotfiles repo and maybe go and simply wrap your cli with a SKILL file, which is probably enough for now.
a month ago
Okay, having the skills in your git repo is ok, this is manageable by an agent to simply go there and copy the files.
So point 32 of my list is mostly resolved now.
Maybe we, as a industry, should adopt some kind of "run md" spec where an agent reads an md file and executes the instructions like code 
bjesuiter
I'm sorry, i really don't like the new skills -_- They work, but they: 1. clutter my skills dir with 12 skill ( i know, probably for the agent to better use the things it needs instead of needing everything, but still... Should't there be a more optimized solution for this? 2. THE SKILL ACTIONS ARE NOT NAMESPACED!!! -_- Example: One of your skills is called `railway-new`. but since the name of this skill in frontmatter, it shows in opencode +oh-my-opencode as `/new` (because oh-my-opencode loads all skills as slash commands. This shadows the /new action for creating a new chat session. I know, my example is very specific, but: - all your skills have this problem: railway-status only is "status" in frontmatter, railway-service is only "service", etc. - if agent harnesses decide that showing skills as slash commands is great (for better visibility), its confusing as hell. 3. Installing them via the script is nice at first, but then you think: ok, how do i backup my skills config? I could symlink my skills somewhere, but this falls apart quickly. I know, this is an issue of the ecosystem not having standardized on a skill marketplace / standardized skill location on disk. But it still sucks. So for now, i'll teach my agent how to manage those skills in my dotfiles repo and maybe go and simply wrap your cli with a SKILL file, which is probably enough for now.
a month ago
Hey! Thank you so much for sharing your feedback
For #1 and #3. You can actually try using skills.sh
Running npx skills add railwayapp/railway-skills will give you an interactive experience where you can pick and choose which skills get installed
#2 makes sense to me. We'll rethink the naming convention
mahmoud-railway
Hey! Thank you so much for sharing your feedbackFor #1 and #3. You can actually try using skills.shRunning npx skills add railwayapp/railway-skills will give you an interactive experience where you can pick and choose which skills get installed#2 makes sense to me. We'll rethink the naming convention
a month ago
Thanks for that! I'll try skills sh for the interactive experience, waiting for openskills v2 right now, looks very promising!