gemini-deep-research

Gemini Deep Research — Platform Knowledge

Standalone Invocation

If invoked directly via /gemini-deep-research, spawn the browser-researcher agent:

Agent(
  subagent_type="browser-researcher",
  name="gemini-researcher",
  prompt="skill_path: <skill-path>/SKILL.md\nresearch_prompt: <from user>\noutput_path: <from user or .work/gemini-report.md>\ncaller: <your name>",
  run_in_background=true
)

Then relay messages between user and agent via SendMessage. If loaded by browser-researcher via Read, ignore this section.

Platform Configuration

• url: https://gemini.google.com/app
• backend: playwright-cli
• submit_method: enter_key
• plan_review: gated
• hard_timeout: 90
• completion_signal: "I've completed your research"

Playwright-CLI Backend

All browser interaction is driven by scripts/pw-driver.py via Bash. The agent calls subcommands; the driver handles Chrome lifecycle, CDP attach, and all DOM interaction internally.

Phase	Subcommand
Login check	python3 scripts/pw-driver.py status (exit 10 = login required)
One-time login	python3 scripts/pw-driver.py login (user runs interactively)
Research activation + submit	python3 scripts/pw-driver.py start-research --prompt-file=<path>
Plan capture	python3 scripts/pw-driver.py wait-plan --timeout=120
Plan approval	python3 scripts/pw-driver.py approve-plan
Plan revision	python3 scripts/pw-driver.py revise-plan --text-file=<path> (or --text=<str>)
Research monitoring	python3 scripts/pw-driver.py wait-complete --timeout=5400
Report extraction	python3 scripts/pw-driver.py extract --output=<path>
Session close	python3 scripts/pw-driver.py close

Exit codes: 0 success · 10 login required · 20 timeout · 30 extraction failure · 40 DOM mismatch.

DOM Reference (verified March 2026)

• Report container: div.markdown.markdown-main-panel (select largest by text length)
• Prompt input: textbox "Enter a prompt for Gemini"
• Deep Research toggle: button "Deselect Deep research" (X on chip)
• Model picker: button "Open mode picker"
• Completion text: "I've completed your research. Feel free to ask me follow-up questions or request changes."

These selectors may change as Google updates Gemini's UI.

Platform Quirks

• Clarifying forms: Gemini may show refinement questions (checkboxes, dropdowns) before research. Handle autonomously based on the research prompt context. Only escalate genuinely ambiguous questions.
• Pro model selection: Always select Pro for best results. If already selected, dismiss picker.
• Report panel: Opens on the right side after completion. If not visible, look for the report link card in chat and click it.
• Research activation: 4-step sequence — Tools button → Deep Research item → model picker → Pro model.