dev-onboarding

name	dev-onboarding
description	Get a developer from a fresh clone to a running localhost — fast and diagnosably. Generates + runs a read-only `doctor` (detect missing tools/versions/secrets, print the exact fix) and an idempotent `bootstrap` (install → env → seed → run), designed for two paths: an internal team member and an external/from-scratch contributor. Use to set up, verify, or scaffold a local dev environment.

Dev Onboarding Skill

Turns "clone → running localhost" from tribal knowledge into a deterministic, self-diagnosing flow. The division of labor:

• Scripts do the reproducible steps (check, install, seed, run) — idempotent, config-driven, CI-usable, no LLM needed.
• You (the agent) diagnose and adapt — pick the profile, read the actual failure, cross-reference CLAUDE.md/.env.example, explain why, and fix it.

It composes with what a repo already has — it doesn't replace the run command (npm start), the .env.example contract, or the code-readability Getting-Started wiki page. The Getting-Started page describes setup for humans; this skill executes and verifies it from the same source facts, so they can't disagree.

The two paths (the core design)

Onboarding diverges almost entirely on secrets and data, never on tools/install:

	team — internal access	contributor — no access (fork/OSS)
Tools, install	identical	identical
Database	the real internal connection string	their own free-tier DB + a seed command
API keys	team/personal keys	personal free keys, or a clearly-labeled mock toggle
Goal	running on live services, fast	running on own/seed data, modify, open a PR

Honesty rule: a contributor "mock" path must never serve silent fake data behind the live code path. Seed a real free-tier DB, or load a clearly-labeled fixture in an explicit local mode. (Mirrors the data-integrity rule many repos keep.)

Modes

• doctor (default, read-only) — cold-start check: OS/shell, each required tool vs. its version range (.nvmrc/engines), .env presence and the shape of each secret for the active profile, optional service reachability (--services). Prints a checklist with the exact fix for every red line; exits non-zero on any blocker (CI-friendly). node scripts/doctor.mjs --profile team|contributor.
• bootstrap (idempotent, mutates) — runs the doctor, then install → create .env from the example (never clobbers) → the profile's data path (contributor: set mock toggles + point at the seed command) → hands off to the run command. Re-runnable. node scripts/bootstrap.mjs --profile ….
• scaffold (one-time, for a new repo) — infers a starter dev-onboarding.config.json from package.json/.nvmrc/.env.example, appends a "Contributor (no internal access)" block to .env.example, and prints the npm aliases + Getting-Started wiring. Review the generated config before committing.
• verify — doctor --services + the run command's smoke endpoints; report green/red. Good for the steward's freshness sweep and for CI.

Configure for your project

Add dev-onboarding.config.json at the repo root (or run scaffold to generate it):

{
  "project": "my-app",
  "runCommand": "npm start",
  "runUrl": "http://localhost:3001",
  "healthEndpoints": ["http://localhost:4000/api/health"],
  "requiredTools": [
    { "name": "node", "versionFrom": ".nvmrc" },
    { "name": "npm", "range": ">=10.8.0" },
    { "name": "git" },
    { "name": "docker", "optional": true, "reason"  
  
   
                  
             
  
   
               
            
  

Wire npm run doctor / npm run bootstrap to the scripts (point at the skill dir, like the other skills). The same config drives any repo.

How it relates to the other skills

• code-readability team owns the docs (Getting-Started + Skill-Inventory pages); this skill is what those pages tell you to run. They read the same package.json + .env.example.
• quality-steward keeps it honest: add doctor/the config to the weekly sweep so the checks never drift from package.json (the exact failure mode that rots hand- maintained setup scripts).

Don'ts

• Don't overwrite an existing .env (bootstrap copies only when missing).
• Don't bake machine-specific assumptions into the scripts — everything from config.
• Don't serve silent mock data for the contributor path (honesty rule above).
• Don't replace the project's run command — hand off to it.

Files

• scripts/config.mjs — config loader + helpers
• scripts/doctor.mjs — read-only cold-start check (the highest-leverage piece)
• scripts/bootstrap.mjs — idempotent setup
• scripts/scaffold.mjs — generate config + contributor block for a new repo