# Delegate Code Research to Gemini CLI

Large codebases are difficult to inspect within one assistant context. This skill delegates scoped research to Gemini CLI and returns structured findings for Claude, Codex, or Claude Code.

## Install

```bash
npx skillstore add dasblueyeddevil/gemini-research-subagent
```

## Metadata

- Status: approved
- Slug: dasblueyeddevil-gemini-research-subagent
- Skillstore revision: r1
- Version status: missing
- Tree hash: fd9c43262b4a4d18fe606b1d4d7866c7472252138293c542cb941e7e47bf7f8f
- Author: DasBluEyedDevil
- GitHub username: DasBluEyedDevil
- License: MIT
- Repository: https://github.com/DasBluEyedDevil/Multi-Agent-Workflow/tree/main/.claude/skills/gemini-research
- Ref: 7e9e368dc1a370f2040369ebcd0db06d90cb48e5
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/dasblueyeddevil-gemini-research-subagent
- Manifest: https://skillstore.pages.dev/api/skills/dasblueyeddevil-gemini-research-subagent/manifest

## Capabilities

- Runs documented Gemini CLI wrapper commands for directory, diff, bug, and verification research.
- Offers role presets for review, debugging, planning, security, auditing, explanation, migration, documentation, dependency mapping, and onboarding.
- Defines custom role naming patterns under \`.gemini/roles/\`.
- Requests structured Gemini responses with summary, files, analysis, and recommendations.
- Separates research, implementation, and verification steps in the workflow.

## Use Cases

- Trace Complex Bugs: Developers can ask Gemini to inspect related files before changing code in Claude, Codex, or Claude Code.
- Plan Architecture Changes: Technical leads can collect cross-file context before approving a multi-component implementation plan.
- Review Security Risks: Security reviewers can request broad codebase analysis before deciding what the main assistant should fix.

## Prompt Templates

### Ask for Context

```
Use the Gemini research subagent to summarize the main modules related to this feature before I edit anything.
```

### Trace a Bug

```
Use Gemini with the debugger role to trace why this behavior happens across the relevant files. Return files, cause, and fix options.
```

### Plan a Change

```
Use Gemini with the planner role to propose an implementation plan for this multi-file change. Include risks, sequencing, and verification steps.
```

### Audit and Verify

```
Use Gemini with the security role to review this area, then use the verify template after I make changes. Flag data exposure risks.
```

## Limitations

- Requires Gemini CLI and a wrapper script that is not included in this package.
- May expose source code or diffs to an external AI service.
- Does not write code; the main assistant must implement changes.
- Role quality depends on local \`.gemini/roles/\` files.

## Best Practices

- Scope directory inputs tightly before invoking Gemini.
- Review Gemini findings before changing code.
- Redact secrets and sensitive customer data from prompts and diffs.

## Anti Patterns

- Running the wrapper automatically on untrusted repositories.
- Letting Gemini write final code without local review.
- Passing full projects when a smaller directory is enough.

## Security Audit

- Audited at: 2026-07-05T08:55:19.77\+00:00
- Summary: Most static detections are Markdown backtick false positives, but the command examples intentionally run a relative Gemini wrapper. The main concerns are untrusted relative script execution, possible source disclosure to Gemini CLI, and prompt injection through local role files.

## Stats

- Views: 231
- Downloads: 11
- Favorites: 0
- Popularity score: 0