Agentic Engineering Starter Kit — Development skill for Claude Code

Agentic Engineering Starter Kit

Everything needed to set up **agentic engineering** in any software project — the practice of treating AI coding agents as team members with versioned conventions, guardrails, workflows, and feedback loops, so every agent session starts with the accumulated knowledge of all previous ones.

Works with any agentic coding tool — Claude Code, `AGENTS.md`-convention tools, Cursor, GitHub Copilot (guide § 3.10). Extracted from a mature production repository (called *the reference project* throughout the guide) and fully de-branded: its material is illustration, never prescription.

Quick start — three entry points

1. Agent-driven (recommended): paste `starter-kit/SETUP-PROMPT.md` into a fresh session of your AI coding agent at your repo's root. The agent detects your tool, mines your repo for ground truth, installs the setup, and verifies it — including a harmonize track for repos that already have agent configuration (preserve-by-default, evidence-based, approval-gated).
2. Script:

   ./starter-kit/bootstrap.sh --target      # day-1 defaults
   ./starter-kit/bootstrap.sh --with-meta ...          # + Level-4 self-improvement loop (later)

   Detects (or asks) which tool the repo uses, installs to that tool's expected locations, never overwrites, idempotent.
3. Manual: follow the file-to-target mapping table in `starter-kit/README.md`.

Then read [`AGENTIC-ENGINEERING-GUIDE.md`](AGENTIC-ENGINEERING-GUIDE.md) — the concepts, a measured reference anatomy, the generic blueprint with design rules for every artifact type, the adoption playbook, and the maintenance flywheel.

What's inside

Path	What it is
`AGENTIC-ENGINEERING-GUIDE.md`	The full guide: § 1 concepts → § 2 reference anatomy → § 3 blueprint → § 4 adoption → § 5 maintenance (+ § 5.6 agentic-graph-engineer