UniSkill V4

极简AI Agent框架 - 3%解决方案。从8771行精简到1000行，保留苏格拉底探明+多模型辩论核心方法论。

Security Scan

The skill's files and runtime instructions are coherent with a minimalist AI-agent / debate engine: it doesn't request unrelated credentials or perform obvious exfiltration, but origin is unknown and there are a few integration/runtime details you should check before installing.

Purpose & Capability

Name/description (minimalist Socratic engine + multi-model debate) matches the provided code: socratic_engine_v4, idea_debater_v4, orchestrator_v4 and gateway implement the stated functionality. There are no unrelated required env vars, binaries, or surprising dependencies declared.

Instruction Scope

SKILL.md and gateway code describe keyword-based automatic triggering and provide a main entry (process_with_uniskill_v4). The runtime instructions and code do not read secrets or system configuration beyond inserting a package path and importing local modules. INTEGRATION.md and the gateway refer to a specific OpenClaw workspace path (~/.openclaw/workspace/skills/universal-skill/core_v4) — this is an integration hint (path insertion), not an explicit request for unrelated files, but it is something to be aware of. Comments/TODOs mention future integration with sessions_spawn / local or cloud models (i.e., external model calls are planned but not implemented).

Install Mechanism

There is no install spec in the registry metadata (instruction-only skill) and no remote download URLs. The package contains source files only. No high-risk download/extract steps are present in the manifest.

Credentials

The skill declares no required environment variables or credentials. The code uses model identifiers as role names (glm-5, qwen3.5-plus, kimi-k2.5) but the implementation currently uses heuristic/local scoring and contains TODOs where real model invocation could be added. No unexpected credential or secret names are requested.

Persistence & Privilege

always:false (normal). The skill is designed to be triggerable automatically by keyword matches (gateway.should_trigger) — this behavior can cause the agent to run the skill for matching user inputs without an explicit manual invocation. Autonomous invocation is the platform default; consider whether automatic keyword triggers are acceptable in your environment.

Assessment

What to check before installing - Source/trust: the skill's origin/homepage is unknown; prefer installing from a verified GitHub/PyPI source and confirm repository contents match. - Automatic triggering: the gateway auto-triggers on many keywords (CNC, decision words). If you don't want automatic runs, disable autonomous invocation or review agent trigger rules. - Future network/model calls: model-calling code is currently stubbed/heuristic with TODOs mentioning sessions_spawn and cloud/local models. If later versions add remote model calls, they may require credentials — only allow that from trusted sources. - Runtime dependencies: orchestrator imports psutil at runtime but requirements.txt does not include it; running in your environment may raise ImportError. Install and test in a sandbox or virtualenv and run the included tests (pytest) before enabling in production. - Path insertion: INTEGRATION.md and the gateway manipulate sys.path and reference an OpenClaw workspace location — ensure this won't inadvertently cause module shadowing or import of unexpected code in your environment. - Review logs and metadata: provided logs and scripts (github-release.sh) include author emails and repo URLs; verify these links before trusting published packages. Recommended actions: run the package in an isolated environment, run the provided tests, inspect the code for any later commits that add external network calls or credential usage, and only enable automatic invocation if you understand and accept the trigger rules.

UniSkill V4 - 极简AI Agent框架

3%解决方案：8,771行 → 1,000行

触发条件

自动触发：

• 用户说 "帮我评估"、"验证这个"、"哪个更好"、"点子王验证"
• 涉及开源/闭源策略决策
• CNC报价相关（材料、加工、零件）
• 多方案选择问题

使用方式

from core_v4 import process_with_uniskill_v4

result = process_with_uniskill_v4(user_input)
if result.triggered:
    if result.needs_more_info:
        return result.question  # 苏格拉底追问
    else:
        return result.recommendation  # 辩论推荐

UniSkill V4 - 极简AI Agent框架

3%解决方案：8,771行 → 260行

核心模块

1. 苏格拉底引擎 (60行)

• 5W2H需求锚定
• 收敛系数判断 (0.7阈值)
• 只在关键参数缺失时"咬人"

2. 高速辩论器 (80行)

• 异步多模型对抗
• 商业分析师/技术顾问/投资顾问
• 五维评分体系

3. 编排器 (120行)

• 整合探明+辩论
• 内存安全 (<100MB)
• 超时保护

快速开始

from uniskill_v4 import check_clarity, quick_debate

# 检查需求清晰度
is_clear, prompt = check_clarity("帮我加工10个TC4零件")

# 多模型辩论
result = quick_debate(
    "开源框架还是开源应用？",
    ["开源框架，闭源应用", "全部开源", "全部闭源"]
)

设计哲学

保留的（V2精华）

• ✅ 苏格拉底需求探明
• ✅ 多模型对抗辩论
• ✅ 收敛系数阈值
• ✅ 五维评分体系

剔除的（V2冗余）

• ❌ 独立收敛检查器
• ❌ 复杂提问模板
• ❌ 伪造日志
• ❌ 24个冗余模块

性能对比

版本	代码量	模块数	启动时间
V2	8,771行	27个	~5秒
V4	260行	3个	<0.5秒

作者

Timo - [e-mail]
海狸 (Beaver) - 靠得住、能干事、在状态

许可证

MIT License

---

靠得住、能干事、在状态 🦫

所有文件均由大帅教练系统生成/dashuai coach