| name | Gemini Research Subagent |
| description | Delegates large-context code analysis to Gemini CLI. Use when analyzing codebases, tracing bugs across files, reviewing architecture, or performing security audits. Gemini reads, Claude implements. |
| dependencies | gemini-cli |
Gemini Research Subagent
You have access to Gemini CLI as a large-context research assistant. Use Gemini for reading large codebases, use yourself for implementation.
When to Invoke Gemini
ALWAYS use Gemini BEFORE:
- Reading files >100 lines
- Understanding unfamiliar code areas
- Tracing bugs across multiple files
- Making changes that affect multiple components
- Performing security or architecture reviews
DO NOT use Gemini for:
- Simple single-file edits you can handle
- Writing code (that's your job)
- Files you've already analyzed this session
How to Invoke
./skills/gemini.agent.wrapper.sh -d "@src/" "Your research question"
Available Roles (loaded from .gemini/roles/)
| Role | Command | Use Case |
|---|---|---|
reviewer |
-r reviewer |
Code quality, bugs, security |
debugger |
-r debugger |
Bug tracing, root cause analysis |
planner |
-r planner |
Architecture, implementation planning |
security |
-r security |
Security vulnerabilities audit |
auditor |
-r auditor |
Codebase-wide analysis |
explainer |
-r explainer |
Code explanation |
migrator |
-r migrator |
Migration planning |
documenter |
-r documenter |
Documentation generation |
dependency-mapper |
-r dependency-mapper |
Dependency analysis |
onboarder |
-r onboarder |
Onboarding guide |
Custom Roles
Add custom roles in .gemini/roles/<name>.md. Examples:
| Role | Focus |
|---|---|
kotlin-expert |
Kotlin/Android, coroutines |
typescript-expert |
TypeScript type safety |
python-expert |
Python async, type hints |
api-designer |
REST API design |
database-expert |
Query optimization |
Templates
# Implementation-ready output
./skills/gemini.agent.wrapper.sh -t implement-ready -d "@src/" "Add user profiles"
# Fix-ready output for bugs
./skills/gemini.agent.wrapper.sh -t fix-ready "Login fails with 401"
# Post-implementation verification
./skills/gemini.agent.wrapper.sh -t verify --diff "Added password reset"
Response Format
Gemini returns structured output you can parse:
## SUMMARY
[1-2 sentence overview]
## FILES
[file:line references]
## ANALYSIS
[detailed findings]
## RECOMMENDATIONS
[actionable items]
Workflow Pattern
- Research: Invoke Gemini to understand context
- Implement: You write code based on Gemini's analysis
- Verify: Invoke Gemini to verify your changes
Example Usage
# Pre-implementation research
./skills/gemini.agent.wrapper.sh -r planner -d "@src/" "How should I add caching?"
# Post-implementation verification
./skills/gemini.agent.wrapper.sh -t verify --diff "Added caching layer"