PenguinHarness：一句话让 Agent 自动构建 Agent，$0.02 生成完整 RAG 应用

by Mycelium Protocol

---

构建一个 Agent 应用，通常的路径是：写 prompt、接工具、调参数、调试循环……一圈下来少则几小时，多则几天。

PenguinHarness（Prism-Shadow）把这个过程倒过来：你写一句话，Agent 帮你构建完整的 Agent 应用。

TypeScript，Apache-2.0，260 stars（2026 年 7 月 19 日开源，两周内）。

---

三个核心能力

1. Agent 自动构建 Agent（$0.02 完成一个完整 RAG 应用）

输入一句话：

Collect the docs from https://github.com/ericbuess/claude-code-docs 
and build a RAG app that answers Claude Code questions as a 
configuration expert, citing its sources.

PenguinHarness 会自动：

• 抓取目标文档
• 搭建 RAG 检索架构
• 写完整代码和运行说明
• 生成带引用来源的问答界面

整个过程在 DeepSeek V4 Pro 上消耗 $0.02（约 ¥0.2） 的 token。

2. 成本碾压：数据分析精度最高，成本 1/70

官方 benchmark（同任务对比）：

Harness	数据分析	编程任务	相对成本
PenguinHarness	最高	≈ OpenAI Codex	1/70
Claude Code	对照	对照	1×

关键设计选择：最小工具集 + 干净的底层接口。每次任务的 tool call 数量和 token 消耗都更少，专门针对 DeepSeek 等开放模型做了深度调优。

3. Agent 自进化：每轮跑完更强

内置 Agent Tuning Skills，让 Agent 对自己做基准测试、找失分点、自动发布下一版：

• 运行基准 → 找出哪些任务失败了
• 自动修改自身的 prompt/工具逻辑
• 打快照（每轮改前都保存）
• 在 Trace 视图里观察每一次请求

这套循环可以无人值守运行，每次迭代后 Agent 的能力都在增强。

---

内置 Skills

四组 Skill，覆盖从办公生产力到 Agent 自优化：

分组	Skills
办公生产力	data-analysis, firecrawl
软件开发	web-design, software-engineering
AI 应用开发	penguin-sdk, penguin-cli, llamafactory, ollama, vllm…
Agent 调优	agent-creation, benchmark-design, agent-evaluation, agent-optimization

Agent 也可以自己写 Skill 并优化它。

---

支持的模型

涵盖当前主流 Frontier 和开放模型：

模型	支持渠道
DeepSeek V4	DeepSeek、OpenRouter、SiliconFlow 等
Kimi K3	Moonshot AI、OpenRouter
GLM 5.2	Z.AI、OpenRouter、SiliconFlow 等
Qwen 3.8 Max	Qwen Token Plan（预览）
GPT 5.6	OpenRouter
Gemini 3.6 Flash	Google Gemini、OpenRouter
Claude 5	Anthropic、OpenRouter

也支持任意 OpenAI 协议兼容端点，本地部署（Ollama/vLLM）同样可用。

---

一行安装

# Linux / macOS
curl -fsSL https://penguin.ooo/install.sh | sh
penguin web   # 启动 Web UI，访问 http://127.0.0.1:7364

# Windows (PowerShell)
irm https://penguin.ooo/install.ps1 | iex
penguin web

# npm（需要 Node >= 24）
npm install -g @prismshadow/penguin-cli
penguin web

Web UI 第一次登录：用户名 admin，密码 penguin-2026（记得立即修改）。

CLI 快速上手

# 配置模型
penguin config model add \
  --provider deepseek \
  --model-id deepseek-v4-pro \
  --api-key sk-... \
  --set-default

# 一次性任务
penguin run -m "Create a Python script that parses CSV and outputs charts"

# 交互 REPL
penguin chat

SDK（给 Agent 用 Agent）

import { createAgent, userText } from "@prismshadow/penguin-core";

const agent = await createAgent({ agentId: "default_agent" });
const session = await agent.createSession({ workspaceDir: process.cwd() });

for await (const output of session.run([userText("Build a RAG app for these docs: ...")], {
  approve: async () => "allow",
})) {
  // 流式接收输出
}

支持离线安装包（GitHub Releases 提供 Linux/macOS/Windows 各架构的 self-contained bundle），适合气隙环境部署。

---

为什么值得关注

“Agent 构建 Agent”正在成为真实的工程能力，而不是 PPT 概念。PenguinHarness 给这个概念加了三层约束：成本可控（$0.02 级别）、可观测（Trace 视图）、可进化（内置基准+自优化循环）。

DeepSeek 作为主力推理引擎的选择不只是性价比——官方 benchmark 显示它在数据分析任务上已经超过闭源竞品，而成本是 1/70。这个剪刀差在 Agent 密集调用的场景下会被放大。

Apache-2.0 开源，可以直接嵌入商业产品。

两周 260 stars，Roadmap 还列了桌面应用、Agent 公司模板、公司级自进化——项目还在早期，但方向感很清晰。

仓库：github.com/Prism-Shadow/penguin-harness · 官网：penguin.ooo · 文档：penguin.ooo/docs

---

Mycelium Protocol — 追踪 AI 系统的底层演化

---

关于 Mycelium

菌丝协议。持续追踪 AI 工具、系统和实验的内容节点。

---

PenguinHarness: An Agent That Builds Agents for $0.02

by Mycelium Protocol

Building an agent application normally means writing prompts, wiring tools, tuning parameters, and iterating through debug loops — hours or days of work.

PenguinHarness (Prism-Shadow) inverts the process: describe what you want in one sentence, and an agent builds the complete agent application for you.

TypeScript, Apache-2.0, 260 stars (open-sourced July 19, 2026 — two weeks).

Three Core Capabilities

1. Agent builds agent ($0.02 for a complete RAG application)

Input a single sentence:

Collect the docs from https://github.com/ericbuess/claude-code-docs 
and build a RAG app that answers Claude Code questions as a 
configuration expert, citing its sources.

PenguinHarness autonomously fetches the docs, builds the retrieval architecture, writes complete code and run instructions, and generates a QA interface with cited sources. Cost on DeepSeek V4 Pro: $0.02.

2. Cost compression: highest data-analysis accuracy at 1/70 the cost

Official benchmark (same tasks, head-to-head):

Harness	Data Analysis	Coding	Relative Cost
PenguinHarness	Best	≈ OpenAI Codex	1/70
Claude Code	Baseline	Baseline	1×

Design choice: a deliberately minimal toolset over clean low-level interfaces — fewer tool calls, fewer tokens — deeply tuned for open models like DeepSeek.

3. Self-evolution: each round makes it stronger

Built-in Agent Tuning Skills run the benchmark, find where the agent loses points, auto-modify its own prompt/tool logic, take a snapshot before each change, and surface every request in the Trace view. This loop can run unattended, with capability improving after each iteration.

Built-in Skills

Group	Skills
Office Productivity	data-analysis, firecrawl
Software Development	web-design, software-engineering
AI App Development	penguin-sdk, penguin-cli, llamafactory, ollama, vllm…
Agent Tuning	agent-creation, benchmark-design, agent-evaluation, agent-optimization

Agents can also write and optimize their own skills.

One-Line Install

# Linux / macOS
curl -fsSL https://penguin.ooo/install.sh | sh
penguin web   # Web UI at http://127.0.0.1:7364

# Windows (PowerShell)
irm https://penguin.ooo/install.ps1 | iex

# npm (Node >= 24)
npm install -g @prismshadow/penguin-cli

First login: admin / penguin-2026 (change immediately).

Why This Matters

“Agent building agents” is becoming a real engineering capability, not a marketing claim. PenguinHarness adds three constraints that make it practical: controllable cost ($0.02-scale), observable (Trace view), and self-improving (built-in benchmark + optimization loop).

The DeepSeek integration isn’t just a cost play — the official benchmark shows it outperforms closed-source competitors on data analysis at 1/70 the cost. That gap compounds dramatically in agent-intensive workloads.

Apache-2.0 means it can be embedded in commercial products. Two weeks in, 260 stars, with desktop app, agent company templates, and company-level self-evolution on the roadmap.

Repository: github.com/Prism-Shadow/penguin-harness · Website: penguin.ooo · Docs: penguin.ooo/docs

---

Mycelium Protocol — tracking the deep evolution of AI systems

© 2026 Mycelium Protocol. All rights reserved.