dev-init

name	dev-init
argument-hint	[--force] [--skip-axial]
description	Initialize project — orchestrates env-setup, ci-setup, release-setup + axial ADR. Use when setting up a new repo or re-running Roxabi project init. Triggers: "dev-init" | "setup project" | "initialize project" | "/dev-init". Not the harness built-in /init (CLAUDE.md scaffold only).
version	0.9.2

Dev-init

Let: I_TS := ${CLAUDE_PLUGIN_ROOT}/skills/dev-init/init.ts F := --force flag present in $ARGUMENTS SKIP_AXIAL := --skip-axial flag present in $ARGUMENTS args := join(F ? "--force" : "", SKIP_AXIAL ? "--skip-axial" : "")

Full project initialization harness. Orchestrates three focused sub-skills in sequence, each independently re-runnable.

Invoke: /dev-init (skill name = plugin name). Claude Code / Grok built-in /init (scaffold CLAUDE.md only) is a different command.

Requires: dev-core installed + enabled (sub-skills + axial-adr-create live there).

Sub-skill	Concern
/dev-core:env-setup	stack.yml, CLAUDE.md rules, docs stubs, LSP
axial-adr-create (agent)	Axis of decomposition ADR — mandatory drift prevention (N×M trap). Skippable via --skip-axial for trivial single-axis projects. See shared/references/axial-decomposition.md
/dev-core:ci-setup	GitHub Actions, TruffleHog, Dependabot, hooks, marketplace plugins
/dev-core:release-setup	Commit standards (Commitizen), hook additions, release automation (semantic-release / Release Please)

Run sub-skills directly to reconfigure a single concern without re-running the full init.

Dual harness (Claude Code + Grok)

Keep this skill portable across hosts:

Do	Don't
Prefer semantic steps (what to run, what to check, what to write)	Hardcode host-only tool names in frontmatter (allowed-tools: Bash, Agent, …)
Use portable env: CLAUDE_PLUGIN_ROOT or GROK_PLUGIN_ROOT (Grok sets both)	Require one host's tool whitelist to load the skill
Invoke sub-skills by stable slash id (/dev-core:env-setup)	Assume Claude Skill / Agent tool shape is available
Spawn agents by role name (axial-adr-create / dev-core:axial-adr-create)	Embed Claude-only or Grok-only APIs as the only path
Shell via the host's bash tool (Claude Bash / Grok run_terminal_command)	Rely on allowed-tools for discovery

When a step needs a subagent, instruct: “spawn the project agent for role X with prompt …” — each host maps that to its Task / spawn_subagent / Agent tool.

Phase 1 — Parse Input + Idempotency

¬F → check existing: test -f .claude/dev-core.yml && echo "1" || grep -c 'dev-core' .env 2>/dev/null || echo "0". result > 0 → present choice Re-configure (≡F) | Skip (abort).

Phase 2 — Prerequisites

Run: bun $I_TS prereqs. Parse JSON → display ✅/❌ table for bun, gh, git remote.

∃ ❌ → show install links:

• bun: https://bun.sh/
• gh: https://cli.github.com/ then gh auth login
• git remote: git remote add origin <url>

→ present choice Abort | Continue anyway (warn: some features won't work).

Phase 3 — Orchestrate

Call sub-skills in order. Each runs its own phases, asks its own questions, displays its own progress.

skill: "dev-core:env-setup", args: "{args}"

Phase 3a — Axial ADR (mandatory drift prevention)

Foundational decision: which axis of variation is primary in this system. Project init is the only moment where the cost of asking is zero — post-scaffold, the axis is implicit in code structure and changing it costs a refactor. Without this ADR, projects drift N×M (target × concern duplication).

Reference: ${CLAUDE_PLUGIN_ROOT}/../shared/references/axial-decomposition.md

Skip path: SKIP_AXIAL ∨ F → emit D("Axial ADR", "⏭️ skipped via --skip-axial") (or --force for blanket override), continue. Trivial single-axis projects (single-purpose CLI utilities, libraries with no transport/adapter axis) legitimately exit here.

Gate path (¬SKIP_AXIAL ∧ ¬F):

1. Check existing:

   Grep tool: pattern="^axial: true|axis of decomposition", path="docs/architecture/adr/", -l, -i
   

2. ≥1 match → D("Axial ADR", "✅ Already present"), continue. (Singleton invariant — if >1 match, dispatch axial-adr-review later to surface the violation.)
3. ∅ → spawn the axial-adr-create sub-agent (host: Agent / Task / spawn_subagent):

   subagent_type: "axial-adr-create"   # or "dev-core:axial-adr-create" on Grok
   description:   "Elicit axial decomposition decision"
   prompt:        "Conduct the axial-decomposition interview for this project. Read ${CLAUDE_PLUGIN_ROOT}/../shared/references/axial-decomposition.md first. Output: ADR file in docs/architecture/adr/ with `axial: true` frontmatter (grep-discoverable canonical marker — singleton invariant)."
   

4. Agent exit status:
   • created ∨ superseded ∨ kept → D("Axial ADR", "✅ {status}"), continue.
   • cancelled ∧ ¬F → halt /dev-init:

     ⛔ Axial ADR required before scaffolding can continue.
     
        State: env-setup has already completed (idempotent on re-run).
               ci-setup, release-setup have NOT run.
     
        Options:
          • Re-run `/dev-init` — will redo env-setup (idempotent) + re-prompt axial-adr-create.
          • Invoke the agent standalone: spawn `axial-adr-create` directly, then re-run `/dev-init`.
          • Skip if this is a trivial single-axis project: `/dev-init --skip-axial` (re-runs env-setup; documents the skip in dev-core.yml).
     
        Rationale: shared/references/axial-decomposition.md
     

   • cancelled ∧ F → ⚠️ warn "axial ADR skipped via --force — drift risk acknowledged", continue.

skill: "dev-core:ci-setup", args: "{args}"

skill: "dev-core:release-setup", args: "{args}"

Phase 4 — Report

dev-init complete
=================

  Run /dev-core:checkup   to verify full configuration health
  Run /dev-core:seed-docs to populate docs stubs from CLAUDE.md + codebase

Next steps:
  /dev-core:checkup           Verify full configuration health
  /dev-core:seed-docs         Populate scaffolded docs with content from CLAUDE.md + codebase
  /dev-core:dev #N            Start working on an issue
  /dev-init --force           Re-configure anytime
  /dev-core:env-setup         Re-run environment setup only
  /dev-core:ci-setup          Re-run CI/CD setup only
  /dev-core:release-setup     Re-run release setup only

Safety Rules

1. Never commit secrets — .env must be gitignored (.claude/dev-core.yml contains only public repo configuration — commit it)
2. Always present choices and wait for user reply before destructive operations (delegated to sub-skills)
3. Idempotent — safe to re-run; sub-skills merge rather than overwrite

$ARGUMENTS