Skill Module Integration

What is a SKILL.md module?

A SKILL.md file is a self-contained, callable unit of AI capability. It defines:

• A skill name and invocation syntax
• The input parameters it accepts (required + optional)
• The output format it produces
• Error conditions and fallback behavior
• Compatibility notes (which models/frameworks support it)

Skills are the composable building blocks of agent workflows. A single agent might call 5–10 skills in sequence or in parallel depending on the task.

SKILL.md file structure

---
skill_id: AA-SKL-001
name: skill-name
version: 1.0.0
compatibility: claude-code, cursor, openai-functions, langchain-tools
---

# SKILL: skill-name

## Invocation
```
/skill-name {required_param} [optional_param]
```

## Input Schema
| Parameter | Type   | Required | Description        |
|-----------|--------|----------|--------------------|
| required_param | string | yes | Description of what this is |
| optional_param | string | no  | Defaults to X if omitted    |

## Output Schema
```json
{
  "status": "success | error | retry",
  "result": "...",
  "metadata": { "confidence": 0.0–1.0, "tokens_used": int }
}
```

## Error Conditions
- MISSING_PARAM: required_param not provided
- INVALID_TYPE: param type mismatch
- TIMEOUT: execution exceeded 30s

## System Prompt
[The actual skill instructions for the LLM]

Installation

# Single skill
cp your-skill.md ~/.claude/skills/

# Multiple skills from a bundle
cp bundle-folder/*.md ~/.claude/skills/

# Verify installation
ls ~/.claude/skills/

# Install to project scope (takes precedence over global)
mkdir -p .claude/skills/
cp your-skill.md .claude/skills/

# Skills in .claude/skills/ are project-scoped
# Only available when Claude Code runs from this directory

Invocation patterns

# Basic call
/skill-name "your input"

# With named params
/skill-name context="your context" task="specific task"

# Output of one skill as input to next
result = /skill-a "input"
/skill-b context={result.output} task="next step"

# Route to different skills based on task type
if task_type == "research":
  /research-skill {input}
elif task_type == "draft":
  /writer-skill {input}

IO validation

Always validate skill outputs before passing them to the next step. Each skill ships with a JSON schema you can use for programmatic validation.

import jsonschema
import json

output_schema = {
    "type": "object",
    "required": ["status", "result"],
    "properties": {
        "status": {"type": "string", "enum": ["success", "error", "retry"]},
        "result": {"type": "string"},
        "metadata": {"type": "object"}
    }
}

skill_output = json.loads(raw_output)
jsonschema.validate(skill_output, output_schema)  # raises if invalid

if skill_output["status"] == "error":
    handle_error(skill_output)
elif skill_output["status"] == "retry":
    retry_with_backoff(skill_call, max_retries=2)
else:
    pass_to_next_step(skill_output["result"])

Error & retry patterns

Next steps