claude-agent-converter

Files (1)

• SKILL.md

SKILL.md (first lines)

# Claude Agent Converter

Convert Claude Code subagents to standard Agent Skills format for portability across AI coding assistants.

## When to Use

- Migrating existing `.claude/agents/*.md` subagent files to `skills/*/SKILL.md` format.
- Extracting reusable workflows from subagent definitions.
- Creating portable skills from Claude Code-specific agents.
- Standardizing agent definitions for use with Claude Code, Codex CLI, Cursor CLI, and other runtimes.

## Inputs

- Source agent file path (e.g., `.claude/agents/my-agent.md`).
- Optional: target skill name (defaults to agent name from frontmatter).
- Optional: mode to extract (if agent supports multiple modes).

If input is missing, ask for the source agent file path.

## Format Differences

### Claude Code Subagent Format

Location: `.claude/agents/<agent-name>.md`

```yaml
---
name: agent-name
description: When Claude should delegate to this subagent
tools: Read, Grep, Glob, Bash
disallowedTools: Write, Edit
model: fast | default | advanced | inherit
permissionMode: default | acceptEdits | dontAsk | bypassPermissions | plan
skills: skill-1, skill-2
hooks:
  PreToolUse:
    - matcher: "Bash"
      hooks:
        - type: command
          command: "./scripts/validate.sh"
---
[System prompt instructions for the subagent...]
```

### Standard Agent Skill Format

Location: `skills/<skill-name>/SKILL.md`

```yaml
---
name: skill-name
description: Complete description including what the skill does and when to use it.
---

# Skill Title

## When to Use

- Scenario 1
- Scenario 2

## Inputs

- Required input 1
- Optional input 2

## Workflow

1. Step 1
2. Step 2
...

## Outputs

- Output 1
- Output 2
```

## Conversion Workflow

1. **Read the source agent** from `.claude/agents/`.

2. **Analyze agent structure**:
   - Identify if agent has multiple modes (e.g., ask, exec, review, search).
   - Determine if agent should become one skill or multiple skills.
   - Note tool restrictions for skill documentation.

3. **Extract metadata**:
   - `name` from YAML frontmatter.
   - `description` for triggering context.
   - `tools` and `disallowedTools` for capability constraints.
   - `model` preference if specified.
   - `skills` dependencies to reference.
   - `hooks` for validation logic to document.

4. **Determine skill name(s)**:
   - Single-mode agent: use agent name directly.
   - Multi-mode agent: create separate skills per mode (e.g., `agent-ask`, `agent-exec`).
   - Use kebab-case for multi-word names.

5. **Create skill directory**: `skills/<skill-name>/`

6. **Transform content** to SKILL.md format:
   - **Frontmatter**: Keep only `name` and `description`.
   - **Enhance description**: Expand to include when to use the skill.
   - **Extract "When to Use"**: From mode selection criteria in agent prompt.
   - **Document Inputs**: What information the skill needs to operate.
   - **Structure Workflow**: Convert agent instructions to numbered steps.
   - **List Outputs**: Expected artifacts or results.
   - **Note Tool Usage**: Document which tools the workflow uses.
   - **Document Constraints**: Read-only modes, permission requirements.

7. **Handle multi-mode agents**:
   - If agent has distinct modes (ask/exec/review/search), create separate skills.
   - Each skill gets the mode-specific workflow and constraints.
   - Shared context (CLI usage, verification) goes in each skill as needed.

8. **Remove runtime-specific content**:
   - Remove `tools`, `disallowedTools`, `model`, `permissionMode` from frontmatter.

From dceoy/ai-coding-agent-skills, AGPL-3.0.

More from dceoy/ai-coding-agent-skills