Skills Compatibility Layer

Description

Ensures the OpenClaw SKILL.md format works seamlessly across nanobot, overstory, and Ollama. Discovers and loads skills, parses frontmatter, builds a unified tool registry, and exports tool definitions in each system's native format.

Architecture

┌─────────────────┐      ┌───────────────┐
│  skill_loader.py │─────▶│ SkillLoader   │──▶ Parses SKILL.md + _meta.json
└─────────────────┘      └───────────────┘
         │
         ▼
┌─────────────────┐      ┌───────────────┐
│ tool_registry.py │─────▶│ ToolRegistry  │──▶ Unified tool catalog
└─────────────────┘      └───────────────┘
                                │
                    ┌───────────┼───────────┐
                    ▼           ▼           ▼
               nanobot      overstory     Ollama
               format       format       format

• skill_loader.py — Discovers skills directories, parses YAML frontmatter from SKILL.md files, extracts tool definitions from _meta.json, returns structured SkillDefinition objects.
• tool_registry.py — Singleton tool registry. Registers tools from skills, MCP servers, or manual definitions. Exports the full catalog in nanobot, overstory, or Ollama JSON-schema format.

Requirements

• Python 3.9+
• No external dependencies (stdlib only — uses json, sqlite3, pathlib, re for YAML-subset parsing)

Commands

Discover skills

python3 scripts/skill_loader.py discover --dir /path/to/skills --json

Load a single skill

python3 scripts/skill_loader.py load --skill /path/to/skills/agent-swarm --json

List registered tools

python3 scripts/tool_registry.py list --json
python3 scripts/tool_registry.py list --capability code --json

Export tool definitions

python3 scripts/tool_registry.py export --format nanobot
python3 scripts/tool_registry.py export --format overstory
python3 scripts/tool_registry.py export --format ollama

Register tools from a skill

python3 scripts/tool_registry.py register --skill agent-swarm --dir /path/to/skills

Usage as Python Module

from skill_loader import SkillLoader
from tool_registry import ToolRegistry

loader = SkillLoader()
loader.load_all("/path/to/skills")

skill = loader.get_skill("agent-swarm")
print(skill.name, skill.version, skill.tools)

registry = ToolRegistry.instance()
registry.register_skill_tools("agent-swarm", "/path/to/skills")

for tool in registry.list_tools():
    print(tool["name"], tool["skill"])

nanobot_tools = registry.export_for_nanobot()
overstory_tools = registry.export_for_overstory()
ollama_tools = registry.export_for_ollama()

Skill Definition Format

Skills are discovered by looking for directories containing a SKILL.md file. The frontmatter block (between --- delimiters) is parsed as key-value pairs:

---
name: my-skill
displayName: My Skill
description: What this skill does
version: 1.0.0
---

If a _meta.json file exists alongside SKILL.md, tool names and additional metadata are extracted from it.

Export Formats

nanobot

[{"name": "tool_name", "skill": "skill-name", "description": "...", "parameters": {...}}]

overstory

[{"tool": "tool_name", "source_skill": "skill-name", "description": "...", "input_schema": {...}}]

Ollama

[{"type": "function", "function": {"name": "tool_name", "description": "...", "parameters": {...}}}]