Ski Platform Exporter

Platform Exporter Skill

Converts a system exported for Google Antigravity to other platforms, applying the necessary transformations according to each platform's conventions.

Input / Output

Input:

• Path of the system in exports/{name}/ (reads .agents/ within it)
• Destination platform: chatgpt, claude-ai, dust, gemini, codex
• System name (extracted from path or provided)

Note: Claude Code and Codex exports are generated automatically as part of the default triple-platform structure (exports/{name}/.claude/ and exports/{name}/.codex/). This skill is only needed for application platform exports.

Output:

• Files generated in exports/{name}/{platform}/
• Summary of exported entities

Procedure

1. Validate input

• Check that exports/{name}/.agents/ exists
• Verify that the destination platform is supported (chatgpt, claude-ai, dust, gemini, codex)
• If the platform already exists in exports/{name}/{platform}/, ask whether to overwrite

2. Read Antigravity structure

Scan .agents/ and register all entities found:

.agents/
├── workflows/      → list all .md files
├── agents/         → list all .md files
├── skills/         → list all subdirectories (ski-name/)
├── rules/          → list all .md files
├── knowledge-base/ → list all .md files
├── commands/       → list all .md files
├── scripts/        → list all scp-* files
├── hooks/          → list all hok-*.md files
└── process-overview.md

3. Apply mapping according to destination platform

Detailed mapping tables per platform (Claude Code reference, Applications, Codex invocation):

../../resources/res-platform-exporter-mappings.md

Read the section for the requested destination platform and apply it.

4. Execute the conversion

For each entity in the list:

1. Read the source file from .agents/{type}/{name}
2. Apply transformations if applicable (none needed for most)
3. Write to the destination path according to the mapping

Maintain the subdirectory structure (e.g. skills maintain their internal folder).

5. Present summary

✅ Export to {platform} complete.
Location: exports/{name}/{platform}/

Exported entities:
- {N} workflows
- {N} agents
- {N} skills
- {N} rules
- {N} knowledge-bases
- {N} commands

[Platform-specific instructions]

Instructions by platform:

• Claude Code: "Open this directory in Claude Code to use the complete system."
• ChatGPT: "Upload the .md files to your project in ChatGPT."
• Claude.ai: "Create a new project and attach the .md files."
• Dust / Gemini: "Upload the files according to the platform's conventions."

Examples

Example 1 — Export to Claude Code

Input:

{
  "system": "exports/customer-onboarding/",
  "platform": "claude-code"
}

Expected output:

• Directory exports/customer-onboarding/ with .claude/ structure
• CLAUDE.md generated with system context
• settings.json created
• 2 workflows converted to commands in .claude/commands/
• 3 agents copied to .claude/agents/
• 1 skill copied to .claude/skills/
• 2 rules copied in .claude/rules/ and referenced in CLAUDE.md

Example 2 — Export to ChatGPT

Input:

{
  "system": "exports/email-classifier/",
  "platform": "chatgpt"
}

Expected output:

• Directory exports/email-classifier/chatgpt/ with folders aligned to the architecture
• 1 agent in workflows/age-spe-email-classifier.md
• 2 skills in skills/ski-xxx/SKILL.md
• 1 rule in rules/rul-xxx.md
• process-overview.md at root

Error Handling

• System not found: Verify that exports/{name}/ exists. If not, list available systems in exports/ and ask the user to specify correctly.

• Platform not supported: Show list of supported platforms: claude-code, chatgpt, claude-ai, dust, gemini, codex.

• Destination export already exists: Ask:

  The export to {platform} already exists at exports/{name}/{platform}/.
  
  Do you want to overwrite it?
  A) Yes, overwrite
  B) No, cancel
  C) Generate in a different directory ({platform}-v2)
  

• Corrupt or unreadable file: Notify and skip that file. Continue with the rest. List skipped files in the summary.