Gemini Agent Booster (2026 Edition — Gemini 2.5 Pro/Flash)

English | Bahasa Indonesia

---

English

Orchestration & Integration

Connects and orchestrates with relevant domain skills like brainstorming, zero-to-prod-orchestrator, and project-context-mapper to ensure cohesive execution.

Description

Master optimization protocol for the Gemini Agent (Antigravity) to leverage native Gemini 2.5 Pro/Flash capabilities — including 1M+ token context window, multimodal vision analysis, deep research mode, thinking chains, and high-speed code generation.

Trigger Conditions

• Analyzing very large codebases, logs, or documents that require 1M+ token context.
• Performing multimodal UI audits (analyzing screenshots of running applications).
• Running deep research tasks requiring web search + reasoning synthesis.
• Generating large, complex code artifacts in a single pass.
• Delegating complex multi-step tasks to browser subagents.

Gemini 2.5 Pro/Flash Capabilities (2026)

Capability	Gemini 2.5 Pro	Gemini 2.5 Flash
Context Window	1M tokens	1M tokens
Thinking / Reasoning	✅ Extended thinking	✅ Flash thinking
Multimodal (Image/Video)	✅	✅
Code Generation	Best-in-class	Very fast
Web Search (Grounding)	✅	✅
Deep Research	✅ (up to 30 min)	✅
Audio	✅	✅
Speed	Slower	5-10x faster
Cost	Higher	Lower

1M+ Token Long-Context Strategies

When working with very large inputs (codebases, documents, logs):

1. Pass full file trees — use list_dir to get the complete file tree, then pass relevant files to maximize context.
2. Read whole files, not snippets — with 1M context, read entire source files rather than grepping for fragments.
3. Cross-file analysis — trace data flows, imports, and interfaces across multiple files simultaneously.
4. Log analysis — pass entire server logs (up to hundreds of thousands of lines) for pattern detection.
5. Codebase onboarding — ingest an entire unfamiliar repo in one pass for deep architectural understanding.

Multimodal Vision — UI Audit Protocol

Use the browser_subagent + generate_image tools for visual feedback:

1. Screenshot the running app via browser subagent.
2. Analyze the screenshot for layout issues, accessibility problems, design inconsistencies, Core Web Vitals opportunities.
3. Generate a target design using generate_image for visual comparison.
4. Iterate code based on visual delta between current and target design.

Thinking Mode — Mandatory Extended Reasoning (o1-Style)

For any complex problem, architecture decision, or large refactor, you MUST engage Gemini's extended thinking or generate a mandatory explicit reasoning chain (<thought>) before generating code:

• Architecture decisions with multiple tradeoffs.
• Debugging hard-to-reproduce race conditions.
• Designing multi-step agentic workflows.
• Cryptographic or security-sensitive implementations.
• Protocol: State the problem clearly → engage deep reasoning chain → critique your own initial assumptions → validate the conclusion → execute tool calls.

Deep Research Mode

Gemini 2.5 Pro's deep research can spend up to 30 minutes actively searching, reading, and synthesizing from hundreds of sources. Use for:

• Market research before building a new feature.
• Security vulnerability research.
• Competitive analysis of existing tools.
• Collecting the latest API documentation for a framework.

Speed Optimization — Flash for Iteration

Use Gemini 2.5 Flash for:

• Quick code generation, small file edits.
• Rapid back-and-forth iteration (linting, formatting, small bug fixes).
• High-volume batch processing.

Use Gemini 2.5 Pro for:

• Complex architecture decisions.
• Long-context full-codebase analysis.
• Security audits requiring deep reasoning.
• Generating comprehensive documentation.

Agent Parallelism

Gemini Agent natively supports parallel tool calls — run independent operations simultaneously:

• Read multiple files at once.
• Run multiple searches in parallel.
• Spawn multiple browser subagents for concurrent tasks.
• Write multiple files simultaneously.

---

Bahasa Indonesia

Integrasi Orkestrasi

Terhubung dan mengorkestrasi skill domain yang relevan seperti brainstorming, zero-to-prod-orchestrator, dan project-context-mapper untuk memastikan eksekusi yang kohesif.

Deskripsi

Protokol optimasi utama untuk Gemini Agent (Antigravity) memanfaatkan kemampuan native Gemini 2.5 Pro/Flash — termasuk context window 1M+ token, analisis visi multimodal, mode penelitian mendalam, rantai pemikiran, dan generasi kode berkecepatan tinggi.

Kondisi Pemicu

• Menganalisis codebase, log, atau dokumen yang sangat besar yang memerlukan konteks 1M+ token.
• Melakukan audit UI multimodal (menganalisis screenshot aplikasi yang berjalan).
• Menjalankan tugas penelitian mendalam yang memerlukan pencarian web + sintesis penalaran.
• Menghasilkan artefak kode yang besar dan kompleks dalam satu lintasan.
• Mendelegasikan tugas multi-langkah kompleks ke browser subagent.

Kemampuan Gemini 2.5 Pro/Flash (2026)

Kemampuan	Gemini 2.5 Pro	Gemini 2.5 Flash
Context Window	1M token	1M token
Pemikiran / Penalaran	✅ Extended thinking	✅ Flash thinking
Multimodal (Gambar/Video)	✅	✅
Generasi Kode	Terbaik di kelasnya	Sangat cepat
Pencarian Web (Grounding)	✅	✅
Penelitian Mendalam	✅ (hingga 30 mnt)	✅
Kecepatan	Lebih lambat	5-10x lebih cepat
Biaya	Lebih tinggi	Lebih rendah

Strategi Long-Context 1M+ Token

Manfaatkan jendela konteks besar dengan membaca seluruh file (bukan hanya fragment), melewatkan seluruh pohon file proyek, menganalisis aliran data lintas file secara bersamaan, dan menganalisis log server lengkap untuk deteksi pola.

Visi Multimodal — Protokol Audit UI

Gunakan browser_subagent untuk mengambil screenshot aplikasi yang berjalan, analisis secara visual untuk masalah tata letak atau aksesibilitas, buat desain target dengan generate_image, lalu iterasi kode berdasarkan delta visual.

Mode Pemikiran — Penalaran Mendalam (Mandatori)

Untuk masalah kompleks (keputusan arsitektur, debugging race condition, alur kerja agentik), Anda DIWAJIBKAN untuk mengaktifkan mode pemikiran diperpanjang (extended reasoning) atau menuliskan rantai pemikiran (<thought>) secara eksplisit untuk mengevaluasi asumsi Anda sebelum menghasilkan kode solusi. Jangan bertindak impulsif.

Mode Penelitian Mendalam

Gemini 2.5 Pro dapat menghabiskan hingga 30 menit secara aktif mencari, membaca, dan mensintesis dari ratusan sumber. Ideal untuk riset pasar, kerentanan keamanan, analisis kompetitif, atau pengumpulan dokumentasi API terbaru.

Flash untuk Iterasi, Pro untuk Kedalaman

• Flash: Edit cepat, iterasi bolak-balik, pemrosesan batch.
• Pro: Keputusan arsitektur, analisis codebase penuh, audit keamanan, dokumentasi komprehensif.

Paralelisme Agen

Gemini Agent mendukung pemanggilan tool paralel secara native — jalankan operasi independen secara bersamaan untuk memaksimalkan throughput.