oracle-gemini（Oracle Gemini 搜索）

Gemini Oracle

Use Google Gemini CLI as a planning oracle and code reviewer. Gemini provides analysis and recommendations; Claude synthesizes and presents to the user.

Critical: This skill is for planning and review ONLY. Never use Gemini to implement changes.

Prerequisites

Before invoking Gemini, validate availability:

~/.claude/skills/oracle-gemini/scripts/check-gemini.sh

If the script exits non-zero, display the error message and stop. Do not proceed without Gemini CLI.

Configuration Defaults

Timeout Guidelines

When invoking gemini via the Bash tool, always set an appropriate timeout:

• Minimum: 5 minutes (300000ms) for any Gemini operation
• Simple queries (single file review, focused question): 5 minutes (300000ms)
• Moderate complexity (multi-file review, feature planning): 10 minutes (600000ms)
• High complexity (architecture analysis, large codebase planning): 15 minutes (900000ms)

Model Selection

Choose model based on task requirements:

Selection heuristics:

• gemini-2.5-pro: Task involves multiple files, requires deep analysis, or architectural thinking
• gemini-2.5-flash: Simple queries, single file review, or when speed is prioritized

For detailed model capabilities, consult references/gemini-flags.md.

Workflow

1. Validate Prerequisites

Run the check script. On failure, report the installation instructions and abort.

2. Determine Mode

• Planning mode: User wants architecture, implementation approach, or design decisions
• Review mode: User wants code analysis, bug detection, or improvement suggestions

3. Construct Prompt

Build a focused prompt for Gemini based on mode:

Planning prompt template:

Analyze this codebase and provide a detailed implementation plan for: [user request]

Focus on:
- Architecture decisions and trade-offs
- Files to create or modify
- Implementation sequence
- Potential risks or blockers

Do NOT implement anything. Provide analysis and recommendations only.

Review prompt template:

Review the following code for:
- Bugs and logic errors
- Security vulnerabilities
- Performance issues
- Code quality and maintainability
- Adherence to best practices

[code or file paths]

Provide specific, actionable feedback with file locations and line references.

4. Select Model and Execute Gemini

Before executing, assess task complexity to select appropriate model:

1. Count files involved in the query
2. Evaluate scope (single module vs cross-cutting)
3. Consider depth (surface review vs architectural analysis)

Use positional prompt syntax (Gemini CLI does NOT support HEREDOC/stdin). Always use the Bash tool's timeout parameter (minimum 300000ms / 5 minutes).

For short prompts, pass directly as a positional argument:

# Short prompt - direct positional argument
gemini -m gemini-2.5-pro --sandbox -o text "Your prompt here" 2>/dev/null

For long prompts, write to a temp file first, then use command substitution:

# Step 1: Generate unique temp file paths and write prompt
GEMINI_PROMPT="/tmp/gemini-${RANDOM}${RANDOM}.txt"
GEMINI_OUTPUT="/tmp/gemini-${RANDOM}${RANDOM}.txt"
cat > "$GEMINI_PROMPT" <<'EOF'
[constructed prompt with code context]
EOF

# Step 2: Execute Gemini with prompt from file
# Bash tool timeout: 300000-900000ms based on complexity
gemini \
  -m "${MODEL:-gemini-2.5-pro}" \
  --sandbox \
  -o text \
  "$(cat "$GEMINI_PROMPT")" \
  2>/dev/null > "$GEMINI_OUTPUT"

Important flags:

• -m: Model selection (gemini-2.5-pro or gemini-2.5-flash)
• --sandbox: Prevents any file modifications (non-negotiable)
• -o text: Plain text output (use json if user requests structured output)
• 2>/dev/null: Suppresses error messages and stderr noise

Bash tool timeout: Estimate based on task complexity (see Timeout Guidelines above). Never use the default 2-minute timeout for Gemini operations.

5. Present Gemini Output

Read the analysis from the temp file and display to the user with clear attribution:

cat "$GEMINI_OUTPUT"

Format the output with clear attribution:

## Gemini Analysis

[Gemini output from /tmp/gemini-analysis.txt]

---
Model: gemini-2.5-pro

For very large outputs (>5000 lines), summarize key sections rather than displaying everything.

6. Synthesize and Plan

After presenting Gemini output:

1. Synthesize key insights from Gemini analysis
2. Identify actionable items and critical decisions
3. If Gemini's analysis presents multiple viable approaches or significant trade-offs, consider using AskUserQuestion to clarify user preferences before finalizing the plan
4. Write a structured plan to ~/.claude/plans/[plan-name].md
5. Call ExitPlanMode to present the plan for user approval

When to use AskUserQuestion:

• Gemini proposes multiple architectures with different trade-offs
• Technology or library choices need user input
• Scope decisions (minimal vs comprehensive) are ambiguous

Skip clarification when:

• Gemini's recommendations are clear and unambiguous
• User's original request already specified preferences
• Only one viable approach exists

Error Handling

Usage Examples

Planning Request

User: "Ask Gemini to plan how to add authentication to this app"

1. Validate Gemini CLI available
2. Gather relevant codebase context
3. Assess complexity → auth spans multiple modules → use gemini-2.5-pro
4. Construct planning prompt with auth requirements
5. Execute Gemini with gemini-2.5-pro model
6. Present Gemini's architecture recommendations
7. Synthesize into Claude plan format
8. Write to ~/.claude/plans/ and call ExitPlanMode

Code Review Request

User: "Have Gemini review the changes in src/auth/"

1. Validate Gemini CLI available
2. Read files in src/auth/ directory
3. Assess complexity → single directory, focused review → use gemini-2.5-flash for speed
4. Construct review prompt with file contents
5. Execute Gemini review
6. Present findings with file/line references
7. Summarize critical issues and recommendations

Model Override Request

User: "Ask Gemini with flash model to review this function"

1. Validate Gemini CLI available
2. Read the target function
3. User explicitly requested flash model → use gemini-2.5-flash
4. Construct focused review prompt
5. Execute Gemini with flash model
6. Present quick feedback

Additional Resources

Reference Files

• references/gemini-flags.md - Complete model and flag documentation

Scripts

• scripts/check-gemini.sh - Prerequisite validation (run before any Gemini command)