Taskmaster

# Gemini CLI-Specific Instructions > **Note:** This file works alongside `AGENTS.md` (generic AI agent instructions). AGENTS.md contains the core Task Master commands and workflows for all AI agents. This file contains only Gemini CLI-specific features and integrations. ## MCP Configuration for Gemini CLI Configure Task Master MCP server in `~/.gemini/settings.json`: ```json { "mcpServers": { "task-master-ai": { "command": "npx", "args": ["-y", "task-master-ai"] } } } ``` **Note:** API keys are configured via `task-master models --setup`, not in MCP configuration. ## Gemini CLI-Specific Features ### Session Management Built-in session commands: - `/chat` - Start new conversation while keeping context - `/checkpoint save <name>` - Save session state - `/checkpoint load <name>` - Resume saved session - `/memory show` - View loaded context Both `AGENTS.md` and `GEMINI.md` are auto-loaded on every Gemini CLI session. ### Headless Mode for Automation Non-interactive mode for scripts: ```bash # Simple text response gemini -p "What's the next task?" # JSON output for parsing gemini -p "List all pending tasks" --output-format json # Stream events for long operations gemini -p "Expand all tasks" --output-format stream-json ``` ### Token Usage Monitoring ```bash # In Gemini CLI session /stats # Shows: token usage, API costs, request counts ``` ### Google Search Grounding Leverage built-in Google Search as an alternative to Perplexity research mode: - Best practices research - Library documentation - Security vulnerability checks - Implementation patterns ## Important Differences from Other Agents ### No Slash Commands Gemini CLI does not support custom slash commands (unlike Claude Code). Use natural language instead. ### No Tool Allowlist Security is managed at the MCP level, not via agent configuration. ### Session Persistence Use `/checkpoint` instead of git worktrees for managing multiple work contexts. ### Configuration Files - Global: `~/.gemini/settings.json` - Project: `.gemini/settings.json` - **Not**: `.mcp.json` (that's for Claude Code) ## Recommended Model Configuration For Gemini CLI users: ```bash # Set Gemini as primary model task-master models --set-main gemini-2.0-flash-exp task-master models --set-fallback gemini-1.5-flash # Optional: Use Perplexity for research (or rely on Google Search) task-master models --set-research perplexity-llama-3.1-sonar-large-128k-online ``` ## Your Role with Gemini CLI As a Gemini CLI assistant with Task Master: 1. **Use MCP tools naturally** - They integrate transparently in conversation 2. **Reference files with @** - Leverage Gemini's file inclusion 3. **Save checkpoints** - Offer to save state after significant progress 4. **Monitor usage** - Remind users about `/stats` for long sessions 5. **Use Google Search** - Leverage search grounding for research **Key Principle:** Focus on natural conversation. Task Master MCP tools work seamlessly with Gemini CLI's interface. --- *See AGENTS.md for complete Task Master commands, workflows, and best practices.*