deepresearch-conversation

# Deep Research Conversation

This skill allows OpenClaw agents to conduct in-depth research discussions with users on a given topic. The API Key is automatically loaded from the OpenClaw config — no manual setup is needed.

## API Table
|    name    |               path              |            description                |
|------------|---------------------------------|---------------------------------------|
|DeepresearchConversation|/v2/agent/deepresearch/run|Multi-round streaming deep research conversation (via Python script)|
|ConversationCreate|/v2/agent/deepresearch/create|Create a new conversation session, returns conversation_id|
|FileUpload|/v2/agent/file/upload|Upload a file for the conversation|
|FileParseSubmit|/v2/agent/file/parse/submit|Submit an uploaded file for parsing|
|FileParseQuery|/v2/agent/file/parse/query|Query the status of a file parsing task|

## Workflow

### Path A: Topic discussion without files
1. Call **DeepresearchConversation** directly with the user's query. A new conversation is created automatically.

### Path B: Topic discussion with files
1. Call **ConversationCreate** to get a `conversation_id`.
2. Call **FileUpload** with the `conversation_id` to upload files.
3. Call **FileParseSubmit** with the returned `file_id`.
4. Poll **FileParseQuery** every few seconds until parsing succeeds.
5. Call **DeepresearchConversation** with the `query`, `conversation_id`, and `file_ids`.

### Multi-round conversation rules
- The DeepresearchConversation API is a **SSE streaming** interface that returns data incrementally.
- After the first call, you **must** pass `conversation_id` in all subsequent calls.
- If the response contains an `interrupt_id` (for "demand clarification" or "outline confirmation"), the next call **must** include that `interrupt_id`.
- If the response contains a `structured_outline`, present it to the user for confirmation/modification, then pass the final outline in the next call.
- Keep calling DeepresearchConversa...