Gemini CLI Skill Guide

Running a Task

1. Model Selection:
   • By default, use the latest Preview model (omit -m flag)
   • Only ask the user (via AskUserQuestion) which specific model to use if the task requires a particular model
2. Select the approval mode based on the task:
   • default: For read-only analysis (prompt for all tool approvals)
   • auto_edit: For editing tasks (auto-approve edit tools)
   • yolo: For full automation (auto-approve all tools)
3. Assemble the command with the appropriate options:
   • -m, --model <MODEL>
   • --approval-mode <default|auto_edit|yolo>
   • -s, --sandbox (if sandbox environment is needed)
   • --allowed-tools <tool1,tool2> (to allow specific tools without confirmation)
   • --include-directories <dir1,dir2> (to include additional directories)
4. For non-interactive prompts, use positional arguments or stdin:
   • gemini "your prompt here" (one-shot mode)
   • echo "your prompt" | gemini (stdin mode)
5. For interactive sessions, use -i, --prompt-interactive:
   • gemini -i "initial prompt" (starts interactive mode after executing prompt)
6. Run the command, capture stdout, and summarize the outcome for the user.
7. After Gemini completes, inform the user: "You can resume this Gemini session at any time by saying 'gemini resume' or asking me to continue with additional analysis or changes."

Quick Reference

Session Management

• Resume latest session: gemini --resume latest "new prompt"
• Resume specific session: gemini --resume 5 "new prompt" (use session index)
• List sessions: gemini --list-sessions
• Delete session: gemini --delete-session <index>
• Resumed sessions automatically inherit the model and approval mode from the original session.

Following Up

• After every gemini command, immediately use AskUserQuestion to confirm next steps, collect clarifications, or decide whether to resume the session.
• When resuming, use gemini --resume latest "new prompt" to continue the previous conversation.
• Restate the chosen model and approval mode when proposing follow-up actions.

Check available models with gemini --help or user configuration.

Error Handling

• Stop and report failures whenever a gemini command exits non-zero; request direction before retrying.
• Before using high-impact flags (-y/--yolo, --approval-mode yolo), ask the user for permission using AskUserQuestion unless already given.
• When output includes warnings or partial results, summarize them and ask how to adjust using AskUserQuestion.

Extensions and MCP Servers

• By default, Disable all extensions (including MCP): gemini --extension '' (empty string)
• List available extensions: gemini --list-extensions
• Use specific extensions: gemini -e extension1,extension2 "prompt"
• Manage MCP servers: gemini mcp add/remove/list
• Allow specific MCP servers: --allowed-mcp-server-names server1,server2

Prompting Best Practices

For advanced prompting strategies to improve response quality and accuracy, refer to @references/prompting-guide.md

Topics covered:

• Core Strategies: Clear instructions, few-shot vs zero-shot prompts, adding context, using prefixes, breaking down prompts
• Response Formatting: System instructions, completion strategies
• Model Parameters: Temperature, topK, topP, stop sequences
• Gemini Best Practices: Structured prompting templates, enhancing reasoning, self-critique
• Iteration Strategies: Rephrasing, analogous tasks, reordering content
• Common Pitfalls: What to avoid and how to handle fallback responses

When to consult: When you need to craft complex prompts, improve response quality, or leverage Gemini's advanced reasoning capabilities.

Prompt Crafting Workflow

Before executing a gemini command, ensure the prompt is well-structured:

1. Clarify requirements first: If the user's request is ambiguous or requires decisions (e.g., choice of approach, scope, constraints), use AskUserQuestion to gather necessary information before crafting the prompt.
2. Structure the prompt: Apply appropriate strategies from the prompting guide:
   • Add context if needed (e.g., relevant code, logs, documentation)
   • Use few-shot examples for format consistency
   • Define constraints explicitly
   • Use XML/Markdown structure with clear sections
3. Execute and iterate: Run the command, then use AskUserQuestion to confirm next steps or gather feedback for prompt refinement.