Agentic-Z

An AI Agent Stack for DayZ Modding — clone it, and your repo comes pre-loaded with specialist agents, slash commands, and a local RAG index that turn Claude Code, Codex, or Gemini CLI into a competent DayZ modding co-author.

• 11 DayZ specialist agents covering scripts, configs, assets, maps, UI, server admin, debugging, and Workbench plugins.
• 26 slash-command skills that drive DayZ Tools end-to-end — preflight, scaffold, build PBOs, launch a local diag server + client, manage types.xml, scope the agent to a single mod, etc.
• A semantic-search RAG over the vanilla DayZ source (Enforce Script, configs, layouts, materials) plus the Bohemia community wiki, exposed to every agent through the dayz-rag MCP server. Embeddings via Voyage AI (voyage-code-3, 200M-token free tier covers ~3 full rebuilds). Or skip the build entirely with /dayz-search-download and pull the prebuilt index from GitHub releases.
• Three-CLI support out of the box. The same agents and skills work in Claude Code, Codex CLI, and Gemini CLI. One sync-skills run wires them all up.

There are two ways to use Agentic-Z. Pick one — running both side-by-side will load every skill twice (once unprefixed, once as agentic-z:<name>).

Option A — Clone as a template (recommended for starting a new mod project):

git clone <this-repo> my-dayz-mod
cd my-dayz-mod
python .claude\skills\sync-skills\sync.py

That symlinks the repo's .claude/skills/ into each agent CLI's home directory so all three discover the same slash commands. Inside Claude Code you can also run /sync-skills instead. Skills appear unprefixed (/dayz-build-pbo).

Option B — Install as a Claude Code plugin (for adding the toolkit to an existing project without forking):

/plugin marketplace add DayZ-n-Chill/Agentic-Z
/plugin install agentic-z@dayz-n-chill

Skills appear with the agentic-z: prefix (/agentic-z:dayz-build-pbo). The dayz-rag MCP server registers automatically.

Then, from your mod project in Claude Code:

/dayz-init

That's it. The wizard asks what you're doing (new mod or import), where to scaffold, what map to test on, and whether to build/launch. It auto-fixes what it can (P:\ mount, junctions) and surfaces steam:// links for the rest. Every run after the first drops you into a mission-control hub for the cached project.

Power users can still call individual skills directly: /dayz-preflight, /dayz-build-pbo, /dayz-launch-test, etc. See the full slash-command list further down.

Full prerequisites, env-var overrides, and troubleshooting: docs/dayz-modding.md.

RAG embeddings run through Voyage AI's hosted API (voyage-code-3, code-tuned, 1024-dim). A free Voyage account includes 200M tokens — enough for ~3 full rebuilds of the vanilla DayZ corpus. Add VOYAGE_API_KEY=pa-… to .env at the repo root before running /dayz-search-index. If you'd rather skip the build entirely, /dayz-search-download pulls the maintainer's prebuilt index from GitHub releases (~1 minute, no API key needed for download — but query-time embedding still needs the key).

L3 files reference L2 in one line. The more specific layer wins ties.

<repo>/
├── .claude/
│   ├── agents/             # 11 DayZ specialists + agent-creator (L3)
│   ├── skills/             # 26 slash-command skills (L3) + L2 shared conventions
│   │   ├── _shared/dayz-conventions.md
│   │   ├── sync-skills/    # bootstrap: link skills into Claude/Codex/Gemini home dirs
│   │   └── dayz-*/         # one folder per DayZ skill
│   ├── mcp/dayz-rag/       # local RAG MCP server (search_dayz_source, get_dayz_file, ...)
│   ├── agent-memory/       # per-agent committed memory
│   ├── local-memory/       # gitignored, per-clone, user/machine notes only
│   └── settings.local.json # per-clone Claude Code settings
├── CLAUDE.md / AGENTS.md / GEMINI.md   # L1 rules (one per CLI, same content)
├── docs/                   # deep docs (DayZ workflow, model routing)
├── wiki/                   # Docusaurus mirror of docs/agents/skills
├── workspace/              # in-progress mods (workspace/<ModName>/) and _server/
└── output/                 # one-shot deliverables

Most skills gate on /dayz-preflight first per L2 conventions. Documented exceptions: /dayz-stop-test (escape-hatch — must work even when env is broken) and /dayz-init (orchestrates other skills, runs preflight on its own steps).

Plus agent-creator for scaffolding new agent definitions.

The full text lives in CLAUDE.md / AGENTS.md / GEMINI.md.

• Communication — answer first, caveat after. Deliver the underlying answer via available tools; mention any limitation as a one-liner after, never as the lead.
• Tooling — pick the fastest tool for the job. Default Python; cmd/.bat for trivial Windows wrappers; PowerShell only when explicitly asked or genuinely faster; prefer dedicated Read / Edit / Write / Grep / Glob.
• Doc maintenance — plain copies. Edit CLAUDE.md / AGENTS.md / GEMINI.md together.
• Bootstrap — /sync-skills after cloning. Required so Codex and Gemini see the slash commands.
• Memory — local-memory/ only, never for rules. User/machine-specific notes only; rules go in the repo.
• Model routing — match model to task. Searches → Sonnet subagent; trivial file-find → Haiku; coding/design → main Opus thread. Details in docs/model-routing.md.

• docs/README.md — architecture overview, how to add skills/agents, local-memory rules.
• docs/dayz-modding.md — full DayZ workflow, prerequisites, env vars, troubleshooting.
• docs/model-routing.md — when to use Opus / Sonnet / Haiku and the subagent patterns.
• docs/prompt-conventions.md — why agent and skill files use caps the way they do (RFC 2119 directives, when to cap, when to stay lowercase).
• wiki/ — Docusaurus build of the docs (kept in sync via /docs-sync).

A new skill: create .claude/skills/my-skill/, add SKILL.md with name/description frontmatter and a "How to run" section, drop the script in. Run /sync-skills to make it discoverable in all agent CLIs.

A new agent CLI: append an entry to .claude/skills/sync-skills/agents.json and run /sync-skills. The new home gets links for every skill automatically.

Join us on Discord: discord.gg/dayznchill

Agentic-Z is built by and for the DayZ modding community, and contributions are highly encouraged. Every agent, skill, and convention in this repo started as a real problem someone hit while shipping a mod — the more modders contribute, the sharper the toolkit gets for everyone.

In the Discord you can:

• Learn how to become a contributor. New contributors are walked through the repo layout, the L1/L2/L3 rule structure, and how to land a first PR.
• Propose new agents and skills. If you keep solving the same problem by hand, that's a skill waiting to be written. Pitch it in Discord and we'll help shape it.
• Report bugs and rough edges. Path resolution quirks, AddonBuilder errors, agent prompts that miss the mark — all welcome.
• Share what you've built. Mods, server setups, custom agents, RAG indexes over your own assets. The toolkit improves fastest when people show what they shipped with it.
• Help refine and enhance the product. Docs improvements, troubleshooting entries, better defaults, new MCP integrations — every contribution compounds.

Whether you're a seasoned Enforce Script developer, a server admin, a 3D artist, or just learning DayZ modding for the first time, there's a place for your work here. The goal is a tool that makes DayZ modding accessible and enjoyable for everyone — and that only happens with community input.

The shipped configuration is the security baseline. Tighten it for your environment, never loosen it without thinking.

• .claude/settings.json (committed) — narrow Bash allowlist limited to running this repo's own skills (Bash(python .claude/skills/*), Bash(npm list *), etc.) plus an explicit deny list that blocks the dangerous escape hatches: Bash(python -c *), Bash(python -m *), Bash(node -e *), Bash(node -p *), Bash(npm install *), Bash(npx *). Editing this file affects every clone — keep allows narrow and prefer deny rules over removing them.
• .claude/settings.local.json (gitignored, per-clone) — your machine-specific overrides. Never set dangerouslySkipPermissions: true in a long-lived clone; it bypasses every prompt including the deny list. Use /fewer-permission-prompts to build a real allowlist incrementally as you hit prompts.
• .claude/agents/*.md — every agent declares an explicit tools: allowlist. Auditor/diagnostic agents (dayz-mod-reviewer, dayz-mod-debugger) cannot Edit/Write/Bash. Authoring specialists get Read/Write/Edit/Glob/Grep plus the dayz-rag MCP tools but no shell or network. Subagents inherit only what's listed — no implicit full-toolset escalation.

If you fork the repo to publish a downstream variant, run git diff main -- .claude/settings.json .claude/agents/ before pushing and convince yourself any widened permissions are intentional.

See LICENSE. Copyright (c) 2026 Brian Orr (DayZ n' Chill). Free to use for developing DayZ modifications.