Agent-Skills.md

Agent Skills: Gemini Deep Research

Run Google Gemini deep research and search-grounded Gemini research tasks with bounded scope and captured outputs. Use when research should rely on Gemini, Google Search grounding, or Google-hosted document context.

UncategorizedID: closedloop-technologies/awesome-deep-researchers/deep-research-gemini

Repository

closedloop-technologies/awesome-deep-researchers
closedloop-technologiesLicense: MIT
71

Install this agent skill to your local

pnpm dlx add-skill https://github.com/closedloop-technologies/climbhill/tree/HEAD/.agents/skills/deep-research-gemini

Skill Files

Browse the full folder contents for deep-research-gemini.

Download Skill

Loading file tree…

.agents/skills/deep-research-gemini/SKILL.md

Skill Metadata

Name
deep-research-gemini
Description
Run Google Gemini deep research and search-grounded Gemini research tasks with bounded scope and captured outputs. Use when research should rely on Gemini, Google Search grounding, or Google-hosted document context.

Gemini Deep Research

Use this skill for Gemini Deep Research Agent or Gemini search-grounded research flows. Gemini Deep Research is useful when research should be grounded in Google Search or uploaded documents.

Environment

op run --env-file .env.climbhill -- python .agents/skills/deep-research-gemini/scripts/gemini_research.py \
  --prompt "Research current deep research APIs. Keep the answer concise and cite sources." \
  --mode grounded \
  --model gemini-3.5-flash

Required 1Password-backed variables:

  • GOOGLE_API_KEY
  • GOOGLE_CLOUD_PROJECT when project-scoped Gemini or Google Cloud APIs need a project resource name

Cost Controls

  • Use search grounding or Deep Research Agent only when fresh web/document research is required.
  • Cap prompt scope and request concise reports.
  • For benchmark parity, record whether the run used the Deep Research Agent, generateContent with grounding, or a manual two-step retrieval workflow.
  • The benchmark command uses grounded Interactions API mode by default because Gemini Deep Research Agent tasks are documented as agentic workflows that can cost around or above the repository's $1 per-call budget. Use --mode deep-research-start only for a deliberate live Deep Research Agent experiment.

Current Source Links

  • https://ai.google.dev/gemini-api/docs/interactions/deep-research
  • https://ai.google.dev/gemini-api/docs/google-search
  • https://gemini.google/overview/deep-research/