AI Tool Converter

Convert agents, skills, commands, and rules between different AI coding assistant formats while preserving functionality and adapting format-specific features.

Supported File Types

Both Claude Code and Cursor support these concepts:

File Locations

.claude/agents/**/*.md

.cursor/agents/**/*.md

.claude/skills/<name>/SKILL.md

.cursor/skills/<name>/SKILL.md

.claude/commands/*.md

.cursor/commands/*.md

.claude/rules/*.md

CLAUDE.md

.cursor/rules/*.mdc

AGENTS.md

Compatibility: Cursor reads

.claude/

Parameters

• source-format

  - Source format (

  claude-code

  ,

  cursor

  )

• target-format

  - Target format (

  claude-code

  ,

  cursor

  )

• file-path

  - Path to file or directory to convert

Steps

Step 1: Identify File Type

Determine the file type from its path:

.claude/agents/foo.md      -> Agent
.claude/skills/foo/SKILL.md -> Skill
.claude/commands/foo.md    -> Command
.claude/rules/foo.md       -> Rule
CLAUDE.md                  -> Rule

If unclear, ask: "What type of file is this? (agent, skill, command, or rule)"

Step 2: Validate Input

1. Parse arguments to extract source format, target format, and file path
2. If parameters are unclear, ask user to clarify
3. Validate source and target formats are supported and different
4. Verify file/directory exists

Step 3: Read and Parse Source

1. Read the source file(s)
2. Parse YAML frontmatter (if present)
3. Extract Markdown body/prompt content
4. Identify format-specific fields

Step 4: Apply Conversion Rules

Agents

Claude Code -> Cursor:

• Preserve:

  name

  ,

  description

• Convert model:

  haiku

  ->

  fast

  ,

  sonnet

  /

  opus

  ->

  inherit

• Remove:

  tools

  ,

  disallowedTools

  ,

  permissionMode

  ,

  color

  ,

  skills

  ,

  hooks

  ,

  disabled

• Add if applicable:

  readonly: true

  ,

  is_background: true

Cursor -> Claude Code:

• Preserve:

  name

  ,

  description

• Convert model:

  fast

  ->

  haiku

• Remove:

  readonly

  ,

  is_background

• Add if applicable:

  tools

  ,

  permissionMode

Skills

Claude Code -> Cursor:

• Keep:

  name

  ,

  description

  ,

  disable-model-invocation

• Remove:

  argument-hint

  ,

  user-invocable

  ,

  allowed-tools

  ,

  model

  ,

  context

  ,

  agent

  ,

  hooks

• Add if needed:

  license

  ,

  compatibility

  ,

  metadata

• Warning:

  $ARGUMENTS

  and

  !`cmd`

  syntax won't work in Cursor

Cursor -> Claude Code:

• Keep:

  name

  ,

  description

  ,

  disable-model-invocation

• Remove:

  license

  ,

  compatibility

  ,

  metadata

• Add if needed:

  argument-hint

  ,

  allowed-tools

  ,

  model

Important: Copy entire skill directory (SKILL.md + references/ + scripts/), not just SKILL.md

Commands

CRITICAL DIFFERENCE: Cursor commands have NO YAML frontmatter - they are plain Markdown only!

Claude Code -> Cursor:

• Remove ALL frontmatter - Cursor doesn't support it
• Remove

  $ARGUMENTS

  ,

  $0

  substitutions - won't work
• Remove

  !`cmd`

  dynamic context - won't work
• Keep body as plain Markdown only

Cursor -> Claude Code:

• Add frontmatter (optional) to enhance functionality
• Consider adding:

  description

  ,

  argument-hint

  ,

  disable-model-invocation

• Keep body unchanged

Rules

Claude Code -> Cursor:

• File:

  CLAUDE.md

  ->

  .cursor/rules/main.mdc

  or

  AGENTS.md

• File:

  .claude/rules/*.md

  ->

  .cursor/rules/*.mdc

• Convert:

  paths

  ->

  globs

• Remove:

  @import

  syntax (not supported)
• Add:

  description

  ,

  alwaysApply

  if needed

Cursor -> Claude Code:

• File:

  .cursor/rules/*.mdc

  ->

  .claude/rules/*.md

• File:

  AGENTS.md

  ->

  CLAUDE.md

• Convert:

  globs

  ->

  paths

• Remove:

  description

  ,

  alwaysApply

Step 5: Update Internal References

CRITICAL: Scan file contents for references to agents, commands, and skills, then update them for the target format.

Common patterns to find:

.claude/agents/foo

.cursor/agents/foo

.claude/skills/bar

.cursor/skills/bar

.claude/commands/baz

.cursor/commands/baz

.claude/rules/qux.md

.cursor/rules/qux.mdc

/foo

/foo

CLAUDE.md

AGENTS.md

Update strategy:

1. Replace directory prefix:

   .claude/

   <->

   .cursor/

2. Replace global rule file:

   CLAUDE.md

   <->

   AGENTS.md

3. Update rule extensions:

   .md

   <->

   .mdc

   (for rules only)
4. Preserve slash command names (they don't change)
5. Convert agent invocation syntax (see below)

Agent Invocation Syntax

Claude Code -> Cursor conversion:

Before:

Use the Task tool to invoke the reviewer:
- Set `subagent_type: code-reviewer`
- Pass the file path in the prompt

After:

Invoke the reviewer agent using:
- Explicit: /code-reviewer review this code
- Natural: Use the code-reviewer subagent to review this code

Cursor -> Claude Code conversion:

Before:

Invoke the code reviewer:
- Use /code-reviewer followed by your request
- Or: Have the code-reviewer subagent check for security issues

After:

Use the Task tool to invoke the code reviewer:
- Set `subagent_type: code-reviewer`
- Include your request in the prompt parameter
- Example: Task tool with subagent_type: code-reviewer, prompt: "Check for security issues"

Step 6: Determine Output Path

Mirror the source structure in the target format:

.claude/agents/foo.md

.cursor/agents/foo.md

.claude/skills/foo/SKILL.md

.cursor/skills/foo/SKILL.md

.claude/commands/foo.md

.cursor/commands/foo.md

.claude/rules/foo.md

.cursor/rules/foo.mdc

Ask user if they want a different output location.

Step 7: Write Converted File

1. Create target directory with

   mkdir -p

   if needed
2. For skills: copy entire directory structure
3. Write converted file with appropriate frontmatter (or none for Cursor commands)
4. Verify file was created successfully

Step 8: Report Results

Converted: [source-file]
Type: [agent|skill|command|rule]
From: [source-format]
To: [target-format]
Output: [output-file]

Changes made:
- [Field changes]
- Removed: [list]
- Added: [list]

Limitations:
- [Features that won't work in target format]

Test the converted file:
[Instructions for target tool]

Examples

Example 1: Convert Agent

claude-code cursor .claude/agents/reviewer.md

Result:

• Input:

  .claude/agents/reviewer.md

  (with

  model: sonnet

  ,

  tools: [Read, Grep]

  )
• Output:

  .cursor/agents/reviewer.md

  (with

  model: inherit

  ,

  readonly: true

  )
• Removed:

  tools

  field
• Added:

  readonly: true

  (since original had restricted tools)

Example 2: Convert Skill

claude-code cursor .claude/skills/code-review/

Result:

• Copies entire skill directory to

  .cursor/skills/code-review/

• Converts SKILL.md frontmatter
• Preserves references/ and scripts/ subdirectories
• Warning:

  $ARGUMENTS

  substitutions won't work

Example 3: Convert Command

claude-code cursor .claude/commands/deploy.md

Result:

• Input:

  .claude/commands/deploy.md

  (with frontmatter and

  $ARGUMENTS

  )
• Output:

  .cursor/commands/deploy.md

  (plain Markdown, no frontmatter)
• Removed: All frontmatter,

  $ARGUMENTS

  references
• Warning: Lost functionality - user must manually pass arguments

Key Differences Summary

$ARGUMENTS

$0

!`cmd`

haiku

sonnet

opus

fast

inherit

tools

disallowedTools

readonly

argument-hint

paths

globs

Validation Checklist

After conversion:

• Correct file type identified
• YAML frontmatter valid (or correctly removed for Cursor commands)
• Required fields present for target format
• No source-format-specific fields remain
• Internal references updated (

  .claude/

  <->

  .cursor/

  ,

  CLAUDE.md

  <->

  AGENTS.md

  )
• Agent invocation syntax converted
• Markdown links point to correct target paths
• Markdown body preserved (except for reference updates)
• File in correct target directory
• User warned about lost functionality

Error Handling