Improve ChatGPT Deep Research progress reporting

[lavapapa]

Summary

• report non-completed ChatGPT Deep Research widget states instead of returning only opaque unavailable
• require a real widget_state.report_message body before returning status=completed
• expose progress metadata such as async task id, widget session id, async status, Venus status, plan title/id, and wait-for-user deadline
• keep polling running / not_ready states with --wait true, but return immediately for waiting_for_user / needs_user_action

Why

ChatGPT Deep Research runs inside a cross-origin internal widget, so the iframe DOM is not readable by OpenCLI. The reliable completed-report path is the conversation payload: metadata.chatgpt_sdk.widget_state.report_message.content.parts[0].

However, the outer ChatGPT /stream_status endpoint can say COMPLETE while the Deep Research widget itself has not produced a report yet. In live validation this appeared as a conversation with an internal Deep Research iframe and widget metadata such as venus_widget_state.status = waiting_for_user_response_on_plan, but no report_message. The old command surfaced this as unavailable, which hid useful state and made --wait misleading.

Behavior

Completed historical reports are unchanged: if report_message exists and looks like a report, the command returns status=completed, report text, and sources.

If a Deep Research widget is present but no report is ready yet, the command can now return structured non-completed statuses:

• running
• not_ready
• waiting_for_user
• needs_user_action

with fields such as:

• progress
• asyncTaskConversationId
• widgetSessionId
• asyncStatus
• venusMessageType
• venusStatus
• waitingForUserUntil
• planTitle
• planId

Test plan

/Users/marvin/.nvm/versions/node/v24.4.1/bin/node --check clis/chatgpt/utils.js
/Users/marvin/.nvm/versions/node/v24.4.1/bin/node --check clis/chatgpt/deep-research-result.js
/Users/marvin/.nvm/versions/node/v24.4.1/bin/node ./node_modules/.bin/vitest run --project adapter clis/chatgpt/utils.test.js clis/chatgpt/commands.test.js
git diff --check

Result: targeted adapter tests passed: 2 files, 113 tests.

Notes

This keeps the previous completed-report extraction behavior and adds progress-aware semantics for report-not-ready widget states. The outer /stream_status: COMPLETE remains diagnostic only; report completion requires report_message.