translate claude skills into chatgpt or codex skills only after auditing purpose, target host, capability parity, resource portability, tool assumptions, and unrealizable behavior. use when the user uploads or points to a claude skill, asks to port a claude skill to chatgpt, codex, or openai skills, or wants a compatibility review before translation. requires a compatibility report, risk score, capability matrix, and user approval before packaging when behavior cannot be preserved faithfully.
83
90%
Does it follow best practices?
Impact
89%
2.34xAverage score across 2 eval scenarios
Passed
No known issues
{
"context": "Tests whether the agent follows the required pre-translation workflow: producing a compatibility report with a capability matrix, correctly classifying MCP/Notion connector capabilities as host-dependent or unrealizable, assigning an appropriate risk score, stopping before packaging to seek approval, and avoiding overbroadening the skill's purpose when capabilities cannot be ported.",
"type": "weighted_checklist",
"checklist": [
{
"name": "Compatibility report produced",
"description": "A compatibility report file or clearly labelled section is produced before any translated skill files are generated",
"max_score": 10
},
{
"name": "Capability matrix present",
"description": "The compatibility report contains a capability matrix table with at least the columns: Capability, Status (or equivalent classification column), and Notes",
"max_score": 10
},
{
"name": "Connector classified host-dependent",
"description": "The Notion MCP connector capability (query_database, update_page, or equivalent read/write Notion access) is classified as host-dependent or unrealizable in the capability matrix",
"max_score": 12
},
{
"name": "Credential assumption documented",
"description": "The compatibility report explicitly lists the NOTION_API_KEY or Notion MCP server as a connector or credential assumption that does not transfer to ChatGPT",
"max_score": 10
},
{
"name": "Risk score high or blocked",
"description": "The compatibility report assigns a translation risk score of 'high' or 'blocked' (not 'low' or 'medium')",
"max_score": 10
},
{
"name": "Approval gate triggered",
"description": "The agent stopped before packaging and explicitly asked the user whether to proceed, rather than producing a translated skill.zip without approval",
"max_score": 14
},
{
"name": "Direct approval wording used",
"description": "The approval request uses wording that identifies specific capabilities that will not be equivalent in ChatGPT (not a generic 'do you want to continue?')",
"max_score": 8
},
{
"name": "No overbroadening",
"description": "The output does NOT instruct ChatGPT to 'use any available source', 'use available tools', or similar fallback that silently broadens the skill's purpose to compensate for missing Notion access",
"max_score": 12
},
{
"name": "Translation mode named",
"description": "The report or decision document names the translation mode as 'adaptation', 'minimum viable translation', or 'blocked' (not 'translation' as if the skill ports cleanly)",
"max_score": 8
},
{
"name": "Host-specific assumptions listed",
"description": "The compatibility report includes a 'Host-specific assumptions' or equivalent section that mentions Claude Desktop and/or MCP server requirements",
"max_score": 6
}
]
}