Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Google Gemini's Deep Research Agent autonomously breaks down complex queries, searches the web systematically, and produces structured markdown reports with citations. It handles the kind of multi-source synthesis that would take a human hours of browsing.
.claude/skills/gemini-deep-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | — | — |
| case-14 | ✗→✓ | ▲ Improved | — | — |
| case-18 | ✗→✓ | ▲ Improved | — | — |
| case-07 | ✗→✓ | ▲ Improved | — | — |
| case-19 | ✗→✓ | ▲ Improved | — | — |
Google Gemini's Deep Research Agent autonomously breaks down complex queries, searches the web systematically, and produces structured markdown reports with citations. It handles the kind of multi-source synthesis that would take a human hours of browsing.
GEMINI_API_KEY environment variable must be set (obtain from Google AI Studio)requests library installedThe script is at scripts/deep_research.py relative to this skill's directory (i.e., the directory containing this SKILL.md). Resolve the full path from the skill's location before running.
bashpython3 <this-skill-directory>/scripts/deep_research.py \ --query "<research query>" \ --stream \ --output-dir ./reports
| Flag | Purpose | Default | |------|---------|---------| | --query | (required) The research question | — | | --stream | Print progress updates while waiting | off | | --output-dir | Where to save the report files | current dir | | --format | Custom output structure (see example below) | free-form | | --file-search-store | Gemini file-search store name | none | | --api-key | Override GEMINI_API_KEY env var | env var |
GEMINI_API_KEY: Run echo $GEMINI_API_KEY to see if it's set. If empty, ask the user whether they'd like to provide a Gemini API key (they can get one from https://aistudio.google.com/apikey). If the user provides one, pass it via --api-key. If the user declines, do not use this skill — fall back to other research approaches and let the user know why.requests is installed: python3 -c "import requests". If missing, install it: pip3 install requests.Basic research:
bashpython3 <this-skill-directory>/scripts/deep_research.py \ --query "Current state of quantum error correction techniques" \ --stream --output-dir ./reports
Custom output format:
bashpython3 <this-skill-directory>/scripts/deep_research.py \ --query "Competitive landscape of EV batteries" \ --format "1. Executive Summary\n2. Key Players (data table)\n3. Technology Comparison\n4. Supply Chain Risks" \ --stream --output-dir ./reports
The script produces two timestamped files in the output directory:
deep-research-YYYY-MM-DD-HH-MM-SS.md — the final markdown reportdeep-research-YYYY-MM-DD-HH-MM-SS.json — full interaction metadataThe report is also printed to stdout so you can capture it directly.
--stream so the user can see progress..md report to the user. Summarize key findings and point them to the full report file.https://generativelanguage.googleapis.com/v1beta/interactionsdeep-research-pro-preview-12-2025x-goog-api-key header| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 18 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.