Skill System

The skill ecosystem

The Neurova Skill system supports dynamic registration, event triggering, permission control and sandboxed execution, and is compatible with the OpenClaw and Qwenpaw protocols.

Core architecture

Skill system
├── Built-in skills
│   ├── memory_manage — add, query and delete memories
│   ├── web_search — search the web
│   ├── file_operation — read and write files (under path protection)
│   ├── kb_builder — import and manage knowledge bases
│   └── browser_read — turn JS pages into Markdown (Playwright)
│
├── External skills
│   ├── OpenClaw-compatible skills
│   ├── Qwenpaw-compatible skills
│   ├── Skill marketplace (market_registry / sources / store)
│   └── Custom skills
│
└── Composite skills
    ├── Chained skills (several skills in sequence)
    └── Parallel skills (several skills at once)

The Skill base class

def execute(self, **kwargs) -> SkillResult:
        """Execute the skill. Subclasses must implement this."""
        raise NotImplementedError
```
## SkillRegistry
- Registers and unregisters skills
- Alias support (memory → memory_manage)
- Events fired before and after execution (PRE_EXECUTE / POST_EXECUTE)
- Usage statistics (execution_count, last_used)
- Permission control, isolated per agent
## Skill marketplace
Skills can be installed from a marketplace:
- **market_registry** — the marketplace registry (slug → source)
- **market_sources** — remote sources (258 skills from Aliyun, 1000 from iFlytek)
- **market_store** — local storage and caching
- **skill_doc_executor** — generates an executable skill from a skill description document
## Sandboxed execution (optional)
- Memory limit (256MB by default)
- CPU limit (50% by default)
- Timeout control (30 seconds by default)
- Network and file access controls