/deep-research

name: deep-research
description: Run actual ChatGPT and Gemini Deep Research through their signed-in browser interfaces, retrieve complete exports, and synthesize the reports. Use for browser Deep Research, resuming its runs, importing its completed reports, or setting up browser access right after installing it. Ordinary agentic research belongs to research-council.

Deep Research

Launch the actual **ChatGPT Deep Research** and **Gemini Deep Research** products, follow their progress, export the full reports, and save the research with the project. Use supported browser/computer tools to do this for the user; manual prompt copying is a fallback for a concrete access problem.

Read [browser routes](references/browser-routes.md) and your host's adapter ([Claude Code](adapters/claude-code.md) or [Codex](adapters/codex.md)) to choose how to operate the browser. Read [browser procedures](references/browser-procedures.md) before touching either product and [reports and synthesis](references/reports-and-synthesis.md) before capturing results. All references are relative to this skill folder. No private scripts, separate synthesis skill, or API keys are required.

**Runs are autonomous.** Browser access is set up once, when the skill is installed. Later runs reuse the saved route (`local/settings.json`) and need the user only for a sign-in, CAPTCHA, quota block or new permission prompt.

Setup

**Right after installing or updating this skill, run [setup](references/setup.md) in the same session**, without waiting for the first research request. Also run it on `/deep-research setup` or "recheck setup", and before a research run that finds no `local/settings.json`. Setup detects the browser route, puts every one-time step into one checklist, verifies Deep Research mode without sending anything, and saves the result. It launches no research.

Select products

**Default: ChatGPT and Gemini Deep Research.** A named product or explicitly saved preference overrides that default. A current request outranks project preferences, which outrank personal preferences. Choosing products needs no questionnaire; browser access is handled once by setup. Save future preferences in the user's existing instructions only when asked.

• “Only ChatGPT” or “only Gemini” runs one product. “Both” runs both browser products.
• Check the signed-in browser, visible Deep Research availability and any displayed allowance before starting. Never assume free access, a fixed monthly quota, or that an app subscription guarantees access. Record an unknown balance as unknown; an enabled mode may still be used within the user's request, without promising it is free.
• If a default product is unavailable, say why and run the available one, reporting a single-product result. If the user requires both or a particular product as a prerequisite, honor that dependency rather than silently omitting it.
• Do not switch to ordinary ChatGPT chat, Codex search, Claude agents, Antigravity, Gemini CLI, Grok, Perplexity, or a paid API as substitutes. Those are outside this skill. `/research-council` provides ordinary agentic research separately.
• Invocation authorizes one research job per selected product, including routine plan approval within the stated brief. Do not buy credits, upgrade plans or change account/security settings. Ask only when a needed choice, sign-in, permission or plan change requires the user.

Prepare and save the request

Use the active project/task's research folder, or create `research/<date>-<topic>-<unique-id>/` there. Ask for the destination only when ambiguous. Do not put the only copy in a temporary directory.

Preserve an existing research brief verbatim unless the user asks to refine it. For an informal question, turn the supplied context into a concise brief: the question/decision, relevant background, scope, comparisons, constraints and desired report. Ask only for consequential missing information. Do not require a separate prompt-review stage or manufacture answers to provider clarifications.

Compare the brief with the original request: preserve the user's objective, supplied fact status, uncertainty, constraints, evidence and intended operation. Distinguish necessary assumptions from user requirements; do not silently add a method, deliverable or approval step.

For full prompt preparation requested by the user or needed to resolve consequential ambiguity, use the installed Prompt skill in preparation/hold mode before freezing the brief. It returns the brief to this workflow, which retains dispatch authority and continues under the existing authorization. Discover Prompt through the host's installed skills; do not assume a neighboring folder. If it is unavailable, use the concise preparation here and disclose any unresolved ambiguity; if the user explicitly requires Prompt itself, report the missing dependency instead of silently substituting. A clear request, including a dictated one, may pass unchanged without a full Prompt pass. Research workers and synthesizers do not restart preparation. Resume, import and synthesis-only requests keep their existing operation and must not become new research launches.

Save the common dispatched text as `PROMPT.md`. Request source-linked evidence, counterevidence and uncertainty. Both products receive the same question; record any provider-specific follow-ups separately. Do not give one product the other's answer. Use only authorized material; a local path in the prompt is not an uploaded file. Verify actual attachments or record their absence. Ask before expanding to unrelated private sources.

Create `RUN.md` before submission. Use the fields and states in [reports and synthesis](references/reports-and-synthesis.md). On every resume, inspect this record and reconnect to existing tasks before opening new ones.

File chats in Deep Research

Use a **Deep Research** container in each app: a ChatGPT Project and a Gemini Notebook where supported. Find