Unified IMA OpenAPI

Problem to solve

When an agent uses the IMA OpenAPI for text and files, “notes” and “knowledge-base” are easy to conflate. A request like “save this into the knowledge base” may actually need append_doc; “add this note to the knowledge base” may require fetching note_id first. Encoding is another hidden failure: PowerShell 5.1 can silently convert the request body to GBK, and notes writes can create irreversible garbled text if content or title is not valid UTF-8.

How the skill works

• Module routing: route to notes or knowledge-base by intent; for cross-module actions such as “view original content” or “add a note to a knowledge base”, read both submodules before executing.
• Credential and API flow: prefer ~/.config/ima/client_id and api_key, then send JSON POST requests to https://ima.qq.com through ima_api.cjs.
• Error handling: check both script-level stderr codes -100/-200 and backend stdout responses where code != 0.
• Write safety: validate UTF-8 before notes text writes; keep uploaded files original, with unchanged filenames and binary contents.

Boundaries

Useful for note search, creation, appending, knowledge-base file upload, link addition, and media info retrieval. It should not process video, Bilibili/YouTube URLs, or file:// resources directly; in those cases, point users to the IMA desktop client. If intent is unclear, such as “save this”, confirm whether to create a new note or append to an existing one.