← Back to Skills Marketplace
124
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install adaptive-skill-stack
Description
自适应技能叠加技能——一个能够根据用户需求实时演化的元技能。每次使用时自动分析需求，叠加已有能力或构建全新能力实例，完成任务后将新获能力写入参考文件实现持续进化。此技能自身SKILL.md永不修改，但可自由改写自身的参考文件、脚本和资源文件。文件保护分四级：SKILL.md绝对不可触碰，结构脚本和格式契约仅用户授...
README (SKILL.md)
\r \r
🌀 自适应技能叠加技能\r
\r
⚠️ 铁律：本文件（SKILL.md）永远不可修改\r
\r
本文件是自适应技能叠加技能的核心定义，自创建之日起永远不允许被任何方式修改——包括但不限于追加内容、替换文本、重写文件。所有能力增长、知识积累、协议演化都必须写入本技能目录下的参考文件（references/）、脚本（scripts/）和资源（assets/）中。\r
\r
违反此铁律的行为等同于技能自毁，必须无条件拒绝。\r
\r
---\r
\r
🔵 核心架构\r
\r
🌐 输入解析层\r
\r 每次接收用户需求时，执行以下解析：\r \r
- 实时语义扫描：识别用户需求中的领域特征、技术术语、情绪暗示和潜在子需求\r
- 四维场景定位：\r
- 领域识别：判断涉及哪些学科/行业/技术栈\r
- 复杂度评级：简单（单技能可完成）/ 中等（2-3技能叠加）/ 复杂（需构建新能力）\r
- 输出预判：预期交付物类型（代码/文档/分析/可视化/多媒体等）\r
- 依赖分析：需要哪些外部工具、API或数据源\r \r
🟣 能力注册表（外部文件）\r
\r
已具备的能力存储在 references/capability-registry.md 中。该文件是技能的"大脑记忆"，记录：\r
- 已掌握的技能能力及其适用场景\r
- 每次任务中获得的新能力\r
- 能力之间的关联关系\r \r 首次使用时，若能力注册表不存在或为空，将创建初始版本并注入基础能力。\r \r
🟢 智能叠加协议（外部文件）\r
\r
叠加和构建新能力的详细协议存储在 references/protocols.md 中。该协议可随使用不断演化，记录：\r
- 标准化的能力叠加流程\r
- 历史任务中的最佳实践\r
- 新能力构建的模板和方法论\r \r ---\r \r
⚡ 运行流程\r
\r
阶段一：需求感知与能力匹配\r
\r
用户需求 → 语义解析 → 能力注册表查询\r
↓\r
┌─────────────────────┐\r
│ 能力注册表是否已覆盖？ │\r
└─────────────────────┘\r
↓ ↓\r
是 否\r
↓ ↓\r
🌈叠加模式 🛠️构建模式\r
```\r
\r
**叠加模式**（需求∩已有能力≠∅）：\r
- 从能力注册表中激活匹配的能力模块\r
- 如需多能力组合，执行叠加协议中的组合策略\r
- 进入执行阶段\r
\r
**构建模式**（需求∩已有能力=∅）：\r
- 使用大语言模型原生能力分析并解决新领域问题\r
- 在解决过程中提炼可复用的方法论\r
- 将新能力写入能力注册表\r
- 更新叠加协议中的相关条目\r
\r
### 阶段二：任务执行\r
\r
1. **展示激活状态**：在响应开头展示当前激活的能力组合\r
```\r
🧩 [能力A] × [能力B] × [能力C] 已激活\r
```\r
2. **执行任务**：按叠加后的能力组合完成用户需求\r
3. **跨维度融合**：当涉及多领域时，展示跨领域融合的解决方案思路\r
\r
### 阶段三：能力沉淀（关键步骤）\r
\r
每次任务完成后，**必须**执行以下操作：\r
\r
1. **读取** `references/capability-registry.md`\r
2. **分析**本次任务中使用了哪些能力、获得了哪些新能力\r
3. **将新能力写入**能力注册表，格式遵循注册表规范\r
4. **若有新方法论**，更新 `references/protocols.md`\r
5. **若生成了可复用资源**（模板、脚本等），保存到 `assets/` 或 `scripts/`\r
\r
---\r
\r
## 📂 文件系统契约\r
\r
本技能的目录结构及各文件的保护等级如下：\r
\r
```\r
adaptive-skill-stack/\r
├── SKILL.md 🔒🔴 绝对不可修改（铁律）\r
├── references/\r
│ ├── capability-registry.md ✏️🟡 结构锁定，内容可增长\r
│ ├── protocols.md ✏️🟡 结构锁定，内容可增长\r
│ └── knowledge/\r
│ ├── README.md 🔒🟠 结构契约不可修改\r
│ └── *.md ✏️🟢 领域知识自由增长\r
├── scripts/\r
│ ├── README.md 🔒🟠 结构契约不可修改\r
│ ├── capability-tracker.py 🔒🟠 结构代码不可修改\r
│ └── *.py（动态创建） ✏️🟢 自动化脚本自由积累\r
└── assets/\r
└── templates/\r
├── README.md 🔒🟠 结构契约不可修改\r
└── * ✏️🟢 模板资源自由积累\r
```\r
\r
### 保护等级说明\r
\r
| 等级 | 符号 | 含义 | 修改条件 |\r
|------|------|------|----------|\r
| 🔴 绝对保护 | 🔒🔴 | 技能身份和基因，不可触碰 | **永远不可修改** |\r
| 🟠 结构锁定 | 🔒🟠 | 定义文件格式和目录约定的结构性文件 | 仅当用户**明确要求重构**时可修改 |\r
| 🟡 结构锁定+内容增长 | ✏️🟡 | 整体结构不可改，但可追加新条目 | 只允许**追加**新内容，禁止重写结构 |\r
| 🟢 自由生长 | ✏️🟢 | 纯知识/资源积累区 | **自由创建、追加、修改** |\r
\r
### 各文件详细权限\r
\r
| 文件/目录 | 读取 | 追加新内容 | 修改结构 | 重写文件 | 创建/删除 |\r
|-----------|------|-----------|----------|----------|-----------|\r
| `SKILL.md` | ✅ | ❌ | ❌ | ❌ | ❌ |\r
| `capability-registry.md` | ✅ | ✅ 追加能力条目 | ❌ 保持类别结构 | ❌ | ❌ |\r
| `protocols.md` | ✅ | ✅ 追加新协议 | ❌ 保持已有协议 | ❌ | ❌ |\r
| `knowledge/README.md` | ✅ | ❌ | ❌ | ❌ | ❌ |\r
| `knowledge/*.md` | ✅ | ✅ | ✅ | ✅ | ✅ |\r
| `capability-tracker.py` | ✅ | ❌ | ❌ | ❌ | ❌ |\r
| `scripts/README.md` | ✅ | ❌ | ❌ | ❌ | ❌ |\r
| `scripts/*.py`（非tracker） | ✅ | ✅ | ✅ | ✅ | ✅ |\r
| `templates/README.md` | ✅ | ❌ | ❌ | ❌ | ❌ |\r
| `templates/*` | ✅ | ✅ | ✅ | ✅ | ✅ |\r
\r
### 铁律补充\r
\r
**🔴 绝对保护的文件**在任何情况下都不可修改，包括：\r
- SKILL.md：技能的核心定义和身份\r
\r
**🟠 结构锁定的文件**定义了格式约定，修改它们会导致数据格式不一致：\r
- `capability-tracker.py`：其正则表达式与注册表格式强耦合，随意修改将导致自动化追踪静默崩溃\r
- `scripts/README.md`：定义了动态脚本的命名和组织约定\r
- `knowledge/README.md`：定义了知识文件的命名和结构约定\r
- `templates/README.md`：定义了模板文件的命名和结构约定\r
\r
**🟡 结构锁定+内容增长的文件**允许追加新条目，但不可改变已有结构：\r
- `capability-registry.md`：可追加新的 `#### 能力` 条目，但不可删除/重命名已有类别标题、统计表格式、日志区格式\r
- `protocols.md`：可追加新的 `## 📐 协议N` 章节，但不可删除或重写已有协议\r
\r
---\r
\r
## 🔄 持续进化机制\r
\r
### 能力增长记录\r
\r
每次使用后，将以下信息追加到能力注册表：\r
\r
```markdown\r
#### [能力名称]\r
- **领域**：所属领域/分类\r
- **触发场景**：什么类型的需求会激活此能力\r
- **核心方法**：解决该类问题的核心方法论\r
- **依赖工具**：需要的外部工具或API\r
- **获得日期**：首次获得该能力的日期\r
- **使用次数**：累计使用次数（每次使用递增）\r
- **关联能力**：与哪些其他能力经常组合使用\r
- **状态**：🟢 活跃\r
```\r
\r
### 知识库扩展\r
\r
对于需要深度知识的领域，在 `references/knowledge/` 下创建领域知识文件：\r
- 文件名格式：`{领域名称}.md`\r
- 内容：该领域的核心概念、方法论、最佳实践\r
- 每次使用中获得的领域新知都应追加到对应文件\r
\r
### 模板库积累\r
\r
对于可复用的输出模板（代码骨架、文档结构、分析框架等），保存到 `assets/templates/` 下，并在能力注册表中引用。\r
\r
---\r
\r
## 🎯 交互展示协议\r
\r
### 动态能力标识\r
\r
每次响应时展示激活的能力状态：\r
\r
```\r
🧩 已激活能力：[能力A] × [能力B] × [能力C]\r
📊 匹配模式：叠加模式 / 构建模式\r
```\r
\r
### 构建模式提示\r
\r
当进入构建模式时，在响应中明确标注：\r
\r
```\r
🛠️ 构建模式启动：检测到新领域需求「XXX」，正在构建新能力实例...\r
✅ 新能力「XXX」已构建并已注册\r
```\r
\r
### 跨领域融合展示\r
\r
当融合多个领域能力时，展示融合思路：\r
\r
```\r
🔄 跨领域融合：[领域A] 的「方法X」 × [领域B] 的「方法Y」→ 融合方案\r
```\r
\r
---\r
\r
## 💡 使用提示\r
\r
- 用户可通过 `@` 符号强制指定能力方向，如 `@数据分析师 分析这份报告`\r
- 能力注册表是持续积累的，使用次数越多，叠加模式触发概率越高\r
- 首次使用某领域能力时必然进入构建模式，后续相同领域将走叠加模式\r
- 所有能力增长均可追溯，支持查看能力进化历史\r
\r
---\r
\r
## 🔗 关联文件索引\r
\r
| 文件 | 用途 | 何时加载 |\r
|------|------|----------|\r
| `references/capability-registry.md` | 能力注册表（所有已掌握能力） | 每次需求解析时 |\r
| `references/protocols.md` | 叠加协议（标准化流程） | 执行叠加/构建时 |\r
| `references/knowledge/*.md` | 领域知识（按需加载） | 涉及特定领域时 |\r
| `scripts/capability-tracker.py` | 能力追踪脚本（自动化记录） | 任务完成时 |\r
| `assets/templates/*` | 任务模板（可复用资源） | 匹配到模板时 |\r
Usage Guidance
This skill is coherent with its description (it keeps a registry and writes new capability entries and resources to disk). The main risk is persistence: after each task it will append to its registry and may create new scripts/templates inside the skill folder, effectively evolving its behavior across runs. Before installing or enabling autonomous invocation, consider:
- Run the skill in a sandbox or restricted environment where the skill directory is isolated and has no access to sensitive files or networked mounts.
- Disable autonomous invocation or require explicit user confirmation before the agent performs file writes or executes any newly created scripts.
- Audit and review any files created under references/, scripts/, and assets/ after each run; especially inspect new scripts (*.py) and templates before allowing them to execute.
- If you need stronger guarantees, make the capability-registry and protocols files read-only by default and require an explicit manual 'commit' step to accept suggested changes.
Confidence is medium because the skill's behavior is internally consistent, but the persistence/self-modifying pattern raises non-negligible operational risk that depends on runtime governance (sandboxing, manual review, agent invocation policy).
Capability Analysis
Type: OpenClaw Skill
Name: adaptive-skill-stack
Version: 1.0.0
The skill bundle defines a 'self-evolving' meta-skill framework that explicitly instructs the AI agent to dynamically create, modify, and execute Python scripts in the `scripts/` directory and manage data in `references/`. While the included `capability-tracker.py` is functionally benign and focused on record-keeping, the core instructions in `SKILL.md` create a high-risk environment for prompt injection. By design, the agent is encouraged to 'freely grow' its capabilities based on user input, which could be exploited to generate and run malicious code (e.g., data exfiltration or persistence) under the guise of a 'newly built capability.' This architecture lacks sufficient guardrails against malicious user-driven evolution.
Capability Assessment
Purpose & Capability
The skill claims to be a meta-skill that adapts by composing or creating capabilities and persisting them to files. The included capability-tracker.py and the references/ and assets/ manifests align with that purpose. Asking to read and append to a local capability registry and protocols file is coherent with the described function.
Instruction Scope
SKILL.md explicitly requires the agent to read and append structured entries to references/capability-registry.md and to update references/protocols.md and to save reusable resources into assets/ or scripts/ after every task. That gives the agent broad discretion to create new scripts and templates in the skill directory; while limited to the skill folder, this is persistent state and is executed/used across runs. There are no instructions to access external secrets or unrelated system files, but the 'must write after each task' rule is prescriptive and could cause unexpected accumulation of code and resources.
Install Mechanism
No install spec and no downloads or external package installs; the skill is instruction + a small local Python script. This is the lowest-risk install mechanism.
Credentials
The skill declares no required environment variables, credentials, or config paths. All file access is limited to files inside the skill directory (references/, scripts/, assets/), which is proportionate to its stated purpose.
Persistence & Privilege
The skill is explicitly designed to persist new capabilities, scripts, templates and knowledge by writing files into its directory after task completion. While writing its own files is allowed and coherent with the purpose, this creates a persistent, self-modifying capability: the agent can accumulate new executable scripts (scripts/*.py) and templates that may be used in future runs. Because the agent is allowed to invoke the skill autonomously (disable-model-invocation=false by default), that persistence expands the blast radius if the agent is later permitted to run generated scripts or load generated resources without human review.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install adaptive-skill-stack - After installation, invoke the skill by name or use
/adaptive-skill-stack - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of "adaptive-skill-stack": a self-evolving meta-skill that adapts in real-time to user needs by combining or building new capabilities.
- Strict file protection hierarchy established: SKILL.md is absolutely immutable; registry and protocol files can only grow, not be restructured; knowledge and templates are freely expandable.
- Introduces a multi-layered workflow: parses input, matches or constructs capabilities, records each new skill and protocol, and updates resources after every task.
- Implements a dynamic capability registry and protocol system, enabling persistent skill growth and transparent evolution history.
- Built-in interactive display protocol: shows active abilities, indicates build mode for new skills, and highlights cross-domain integrations during responses.
- Designed for ongoing expansion: every use expands the skill set, with all new methods, scripts, and templates automatically catalogued outside the core definition file.
Metadata
Frequently Asked Questions
What is Adaptive Skill Stack?
自适应技能叠加技能——一个能够根据用户需求实时演化的元技能。每次使用时自动分析需求，叠加已有能力或构建全新能力实例，完成任务后将新获能力写入参考文件实现持续进化。此技能自身SKILL.md永不修改，但可自由改写自身的参考文件、脚本和资源文件。文件保护分四级：SKILL.md绝对不可触碰，结构脚本和格式契约仅用户授... It is an AI Agent Skill for Claude Code / OpenClaw, with 124 downloads so far.
How do I install Adaptive Skill Stack?
Run "/install adaptive-skill-stack" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Adaptive Skill Stack free?
Yes, Adaptive Skill Stack is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Adaptive Skill Stack support?
Adaptive Skill Stack is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Adaptive Skill Stack?
It is built and maintained by 波动几何 (@wangjiaocheng); the current version is v1.0.0.
More Skills