Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add majiayu000/spellbook --skill gemma4-local-deploygit clone --depth 1 https://github.com/majiayu000/spellbookWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/majiayu000/spellbook/gemma4-local-deploy)<a href="https://agentmods.dev/skills/majiayu000/spellbook/gemma4-local-deploy"><img src="https://agentmods.dev/badge/skills/majiayu000/spellbook/gemma4-local-deploy/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/majiayu000/spellbook/gemma4-local-deploy"><img src="https://agentmods.dev/badge/skills/majiayu000/spellbook/gemma4-local-deploy.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00172 | $0.01778 |
| Opus 5 | $0.00086 | $0.00889 |
| Sonnet 5 | $0.00034 | $0.00356 |
| Haiku 4.5 | $0.00017 | $0.00178 |
Grade A, and why
gemma4-local-deploy scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Gemma 4 12B 本地部署
把 Gemma 4 12B 的 GGUF 版本部署成本机模型服务。默认使用 llama.cpp / llama-server、Apple Metal、Q4_K_M 和 tmux,只监听 loopback;用户明确要求 QAT、256K、对比演示或 Ollama 时才切换路线。
Operating Contract
- Direct actions: 读取本机硬件、磁盘、端口、进程和模型缓存;在用户已要求本地部署时,安装或升级明确的软件包、下载选定模型、创建专用模型目录和 tmux 会话,并只绑定
127.0.0.1。 - Escalate before: 停止不属于本 Skill 的现有进程、覆盖已有模型或配置、删除用户数据、监听公网地址、改变防火墙,或下载用户未选择的大型模型变体。
- Evidence-backed pushback: 如果用户指定的模型标签、上下文、内存预算或本机能力与当前可验证状态冲突,先展示命令输出并提出可运行的 profile,不伪造支持状态。
- Feedback loop: 现状检查 → 选择并复述 profile → 执行一条部署路线 → 当前会话完成健康、模型和聊天验证 → 报告端点、资源与限制。
默认选择
- 默认模型仓库:
ggml-org/gemma-4-12B-it-GGUF - 默认量化:
Q4_K_M - 默认模型名:
gemma-4-12b-it - 默认端点:
http://127.0.0.1:8080 - 默认上下文:
32768 - 12B 长上下文:用户明确要求时选择
65536或131072 - QAT 仓库:
google/gemma-4-12B-it-qat-q4_0-gguf - QAT profile:
Q4_0、262144上下文 - 默认后台会话:
gemma4-12b - 默认关闭 thinking:
--reasoning off,避免 OpenAI API 的message.content为空 - Ollama:只在用户明确要求 Ollama 或需要 Ollama 生态时使用
QAT 是训练时模拟量化,不等于无损。关键任务仍要用当前会话的真实响应验证。用户明确要更高质量时,优先建议 Q6_K 或 Q8_0;除非用户接受更高内存和更慢加载,不默认使用 bf16。
Profile 选择
| Profile | 适用场景 | Model / quant | Context | Port / alias |
|---|---|---|---|---|
daily-q4km-32k |
默认日常聊天、编码、低风险本地 API | ggml-org/...:Q4_K_M |
32768 |
8080 / gemma-4-12b-it |
long-q4km-128k |
明确需要更长上下文,但保留默认 GGUF 路线 | ggml-org/...:Q4_K_M |
65536 或 131072 |
8080 / gemma-4-12b-it |
qat-q4_0-256k |
明确要求 QAT、Q4_0、256K 或低内存长上下文 | google/...qat-q4_0-gguf:Q4_0 |
262144 |
8080 / gemma-4-12b-it-qat-q4_0 |
compare-32k-vs-256k |
录屏、演示或 A/B 比较资源与速度 | 左 Q4_K_M,右 QAT Q4_0 |
32768 + 262144 |
8080 + 8081 |
最终回复必须说明选定 profile、端口、上下文和选择依据。不要把 256K 当作日常默认值。
执行流程
1. 搜索并确认现状
先检查已有安装、进程、端口、缓存、硬件和磁盘,避免重复部署:
command -v llama-server || true
llama-server --version || true
tmux has-session -t gemma4-12b 2>/dev/null && tmux display-message -p -t gemma4-12b '#S #{pane_pid}' || true
lsof -nP -iTCP:8080 -sTCP:LISTEN || true
ls -lh "$HOME/Library/Caches/llama.cpp/"*gemma-4-12B-it*Q4_K_M*.gguf 2>/dev/null || true
find "$HOME/Library/Caches/llama.cpp" "$HOME/Models" \( -name '*gemma-4-12b-it-qat-q4_0*.gguf' -o -name '*gemma-4-12B-it-qat-q4_0*.gguf' \) 2>/dev/null || true
system_profiler SPHardwareDataType | sed -n '1,30p'
df -h "$HOME"
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 98 lines · 172 tokens per session scan A 022825c3434c
gemma4-local-deploy is a skill published in the GitHub repository majiayu000/spellbook (277 stars, last pushed 2d ago), licensed MIT. It adds 172 tokens to every session and 1,778 once invoked, about $0.0009 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
pneuma-session
Instructions for renaming an active Pneuma session and replacing its default preview with a useful title and summary. A Pneuma session is one work area inside a larger project.
feishu-openapi-skill
Operate Feishu or Lark IM APIs through UXC with a curated OpenAPI schema, tenant-token bearer auth, and chat/message guardrails.
slack-openapi-skill
Operate Slack Web API through UXC with a curated OpenAPI schema, bearer-token auth, and messaging-core guardrails.
spring-shell
Spring Shell for building interactive CLI applications. Covers @ShellComponent, @ShellMethod, input validation, tables, and command groups. USE WHEN: user mentions "spring shell", "@ShellComponent", "@ShellMethod", "CLI application Spring", "interactive shell", "command line tool Spring" DO NOT USE FOR: simple scripts…
huggingface-community-evals
Run evaluations for Hugging Face Hub models using inspect-ai and lighteval on local hardware. Use for backend selection, local GPU evals, and choosing between vLLM / Transformers / accelerate. Not for HF Jobs orchestration, model-card PRs, .evalresults publication, or community-evals automation.
session-handoff
Use when the user wants to hand off, transfer, pause, or continue the current session in a new session or with another agent — asks for a "session handoff", a "prompt para a próxima sessão", to "continuar de onde paramos", or invokes /session-handoff; also when context is running low and in-flight work must survive a…