Gemini Deep Research

**Judgement:** A strong, specialized skill for rigorous, multi-source research and synthesis using Google’s Gemini Deep Research Agent. It is best suited for substantial, non-real-time research tasks where depth and structure matter more than speed. **What it does well** - Breaks complex queries into sub-questions, systematically searches the web, and synthesizes findings into comprehensive reports. - Produces both Markdown (human-readable report) and JSON (full interaction metadata), making it easy to consume results programmatically or in UIs. - Supports custom report formats and optional File Search integration, enabling tailored executive summaries, competitive matrices, and internal-vs-public comparisons. - Streaming mode allows callers to monitor progress on long-running jobs. **Key risks & limitations** - **External dependency & credentials:** Requires a valid `GEMINI_API_KEY` from Google AI Studio and does *not* support Antigravity OAuth tokens. If the key is missing, misconfigured, or rate-limited, the skill will fail. - **Latency:** Deep research runs can take minutes to hours depending on query complexity, making it unsuitable for low-latency chat-style interactions. - **Cost exposure:** Heavy or frequent use may incur non-trivial API costs, depending on the project’s Google quota and billing. - **Content quality & hallucinations:** While Gemini’s research agent is designed for multi-source grounding, outputs may still reflect web noise, outdated sources, or occasional hallucinations; downstream systems should treat outputs as high-quality drafts, not ground truth. **Recommended use scenarios** - When a user or workflow explicitly requests **deep, multi-source research** (e.g., market landscapes, technical surveys, or policy overviews) rather than quick factual answers. - For **structured reports**: executive summaries, competitive overviews, risk assessments, or multi-section technical documents where the custom format option is valuable. - For **background tasks** that can run asynchronously, with results later surfaced to the user or stored in knowledge bases. - In **enterprise or analyst workflows** that can tolerate higher latency and cost in exchange for more thorough research and documentation. Avoid using this skill for simple Q&A, quick lookups, or any interaction where users expect near-instant responses.