Gemini CLI Additional FAQ & Answers

Use gemini 'your prompt' --output-format json to output machine-parseable structured data instead of human-readable text. Response format: {response: "AI text", toolCalls: [{name, parameters, results}], metadata: {model, tokens, latency}}. Parse with jq: gemini 'count functions' --output-format json | jq -r '.response' extracts text, jq '.toolCalls[].name' lists tools used. Essential for CI/CD integration: parse JSON to count issues by severity, fail build if critical issues >0. Example: RESULT=$(gemini --output-format json 'analyze code'), COUNT=$(echo $RESULT | jq '.issues | length'), exit 1 if COUNT exceeds threshold. Enables automation with reliable structured output vs fragile regex parsing of markdown.

Production patterns: (1) CI/CD quality gates - GitHub Actions analyzes PR code, parses JSON for issue counts, fails build if critical >0. (2) Security scanning - nightly cron analyzes codebase for OWASP Top 10, outputs findings to JSON, Slack notification if vulnerabilities detected. (3) Log analysis - process server logs in batch, extract error data (timestamp, type, stack trace), aggregate with jq -s 'add' into single report. (4) Documentation generation - weekly automation generates OpenAPI specs from TypeScript routes, converts JSON to YAML, commits to docs/ folder. (5) Parallel processing - use xargs -P 4 for concurrent Gemini calls respecting rate limits (60/min free tier). Best practices: validate JSON with jq '.', implement retry logic with exponential backoff, check exit codes.

Search grounding connects Gemini to real-time web data via Google Search API, providing current information beyond January 2025 training cutoff. Mechanism: Gemini autonomously performs Google searches when needed, retrieves top 10 results with titles/snippets/URLs, grounds answers in cited sources with inline citations (Source: [1] example.com). Activation: gemini config set grounding true (global) or --grounding flag (per-query). Benefits: improves factual correctness 45% for post-training queries (2024-2025 content), accurate dependency versions, framework comparisons, recent security advisories. Performance impact: adds 3-8 seconds latency (search API call + processing), increases token consumption 30-50% (search results added to context), counts toward rate limits. Cost: grounded queries use 2-3x more tokens.

Enable grounding for: (1) Version/release queries - current framework versions, recent changelogs (training cutoff limitation). (2) Security research - latest CVEs, security advisories, penetration testing techniques. (3) Comparison shopping - SaaS pricing, features, benchmarks for tools updated 2024-2025. (4) Troubleshooting errors - Stack Overflow solutions, GitHub issues from recent releases. (5) Learning new technologies - tutorials for tools launched 2024-2025. Disable for: code generation/refactoring (no benefit, adds latency), codebase analysis (irrelevant to private code), well-known information (established best practices), high-frequency automation (slows CI/CD 2-3x), offline environments (requires internet). Best practice: keep global grounding disabled, use --grounding flag selectively when needed. Prevents accidental quota waste.

Create ~/.gemini/commands.json or .gemini/commands.json (project-local): {"commands": {"security-review": {"prompt": "Review for OWASP Top 10 vulnerabilities...", "files": true, "outputFormat": "markdown"}}}. Usage: gemini security-review --file src/auth.ts executes standardized prompt. Advanced: parameterized commands with ${VAR} placeholders, multi-step workflows with checkpoints, conditional logic based on file type. Team collaboration: commit .gemini/commands.json to repo (version control commands alongside code), document in README. Production use cases: security reviews before deployment, API doc generation (OpenAPI from code), test generation (>80% coverage), code migrations (React Class → Hooks). Best practices: name as verb-noun (review-security, generate-tests), include examples in prompts (few-shot learning improves output 40%), version commands semantically.