yalla-skills
AI-agent skills for yalla — the CLI that deploys and operates apps on Dokploy.
This repository is a Claude Code plugin marketplace plus standalone variants for other agents. The skills teach the model how to plan, ship, verify, and recover deploys end-to-end through yalla — from a fresh docker-compose stack to a path-routed monorepo with custom certs and persistent volumes.
Prerequisite for runtime use: the yalla CLI must be installed and authenticated against your Dokploy instance. The skill itself is documentation — it teaches an agent which yalla commands to run; it does not include the binary. See the yalla repo for installation.
Install
Claude Code (recommended path)
Two commands inside any Claude Code session:
/plugin marketplace add JuribaDev/yalla-skills
/plugin install yalla-dokploy-deploy@yalla-skills
After install the skill loads automatically in every conversation. It activates on phrases like "deploy this to my Dokploy", "ship this dist folder", "tear down the staging project", "add a postgres to this app", and many others — see plugins/yalla-dokploy-deploy/skills/yalla-dokploy-deploy/SKILL.md for the full trigger list (English + Arabic).
To update later:
/plugin update yalla-dokploy-deploy@yalla-skills
Claude.ai
The .skill single-file package isn't shipped in v0.1.0 (the description is over Claude.ai's current 1024-char cap — trimming it is on the v0.2.0 roadmap). For now, install manually by uploading the skill folder via Claude.ai's Skills UI, or use the Claude Code install path above if you have Claude Code.
Codex CLI
Codex doesn't use the Claude Code marketplace format. Install manually:
mkdir -p ~/.codex/skills
git clone --depth 1 https://github.com/JuribaDev/yalla-skills /tmp/yalla-skills-checkout
cp -R /tmp/yalla-skills-checkout/codex/yalla-dokploy-deploy ~/.codex/skills/
rm -rf /tmp/yalla-skills-checkout
(Or copy the codex/yalla-dokploy-deploy/ folder directly into your project's .codex/skills/ if you prefer per-project install.)
Manual install (any other agent runtime)
The skill is a plain folder containing SKILL.md + references/ + scripts/. Copy plugins/yalla-dokploy-deploy/skills/yalla-dokploy-deploy/ into wherever your agent expects skill folders. The SKILL.md frontmatter (name, description) is what gates triggering.
What's in this repo
yalla-skills/
├── .claude-plugin/
│ └── marketplace.json ← declares the repo as a Claude Code marketplace
├── plugins/
│ └── yalla-dokploy-deploy/
│ ├── .claude-plugin/plugin.json ← plugin metadata
│ └── skills/yalla-dokploy-deploy/ ← the Claude / Claude.ai skill
│ ├── SKILL.md ← entry point (~240 lines, always loaded)
│ ├── references/ ← 18 reference docs loaded on-demand
│ │ ├── composite-verbs.md ← yalla deploy compose, teardown, wait, rescue, audit
│ │ ├── databases.md ← postgres/mysql/mariadb/mongo/redis + backups
│ │ ├── lifecycle.md ← the canonical ordered API-call recipe
│ │ ├── drop-deploy.md, mounts.md, ... ← per-scenario refs
│ ├── scripts/
│ │ └── detect_stack.py ← build-type + DB detection heuristics
│ └── README.md
└── codex/
└── yalla-dokploy-deploy/ ← same content + agents/openai.yaml for Codex
Versioning
Released versions match the yalla CLI release they were built and tested against. The version is in .claude-plugin/marketplace.json and plugins/yalla-dokploy-deploy/.claude-plugin/plugin.json. The skill bakes in CLI flag names and operation IDs, so pin yalla CLI within the same minor version when possible.
Trigger surface — quick reference
The skill auto-triggers on these intents (and many phrasings of each):
