Platform Compatibility
What each export looks like and where it falls short. Drafts are not certified by any provider; no universal interoperability is claimed.
OpenAI / ChatGPT
Draft export readyCustom GPT instructions + Assistants/Responses tool JSON (draft)
Structure
- instructions (system)
- tools[] as function JSON schema
- conversation starters from triggers
Caveats
- Tool definitions are drafts; you must implement each function.
- Custom GPT knowledge files are not included.
- Not submitted to or certified by OpenAI.
Live provider validation: not connected
Anthropic Claude
Draft export readyClaude Skill folder draft (SKILL.md with front matter)
Structure
- SKILL.md name/description front matter
- Markdown instructions body
- examples section
Caveats
- Tool use requires your own tool definitions in the API call.
- Front matter fields may change with Anthropic updates.
- Not submitted to or certified by Anthropic.
Live provider validation: not connected
Google Gemini
Partial — manual stepsGemini system instruction + functionDeclarations (draft)
Structure
- systemInstruction text
- tools.functionDeclarations[]
- safety notes as text
Caveats
- Gems UI does not import JSON; copy instructions manually.
- Function declarations need your implementation.
- Not certified by Google.
Live provider validation: not connected
Generic agent spec
Draft export readySteadyForge portable schema sf-skill/v1 (JSON)
Structure
- Full skill object
- agentMeta (risk, capabilities)
- Flow hook names
Caveats
- Target runtimes must map fields themselves.
- Portable ≠ universally interoperable.
Live provider validation: not connected