| agents | ||
| commands | ||
| skills | ||
| gemini-extension.json | ||
| GEMINI.md | ||
| README.md | ||
Agent Code Extension
A Gemini Extension for structured feature planning and implementation with Codex-driven reviews. This extension provides a disciplined engineering workflow to ensure high-quality, testable, and well-documented code changes.
Overview
The Agent Code Extension facilitates a professional software development lifecycle within the Gemini CLI. It breaks down complex feature requests into manageable, plan-driven tasks, leveraging specialized sub-agents for implementation and expert code review.
Core Workflows
📝 Plan (/plan)
Collaborative feature planning that produces a verified implementation plan.
- Requirements Discussion: Refine scope and constraints with the user.
- Initial Documentation: Draft
plan.mdandtasks.mdin a dedicated session folder. - Codex Review Loop: Mandatory plan scrutiny by the
codex-analyzersub-agent. - Final Approval: User sign-off on the hardened plan before implementation begins.
🛠️ Implement (/impl)
Iterative implementation of an approved plan with automated tasks and continuous review.
- Plan Analysis: Internalize the session plan and verify constraints.
- Task Breakdown: Divide work into small, independent tasks (1-3 files each).
- Iterative Cycle: For each task:
- Implement changes using
task-implementer. - Run automated tests to verify behavior.
- Perform Codex-driven code review for quality and security.
- Commit changes with standardized messages.
- Update session documentation (
tasks.mdandplan.md).
- Implement changes using
🚀 Commit (/commit)
Standardized git commit generation following the Conventional Commits specification with emoji prefixes.
- Format:
<emoji> <type>(<optional scope>): <description> - Types:
feat,fix,docs,style,refactor,perf,test,build,ci,chore.
Specialized Sub-Agents
🔍 codex-analyzer
An expert code analysis orchestrator that uses the Codex CLI to perform deep technical reviews. It identifies:
- Logic bugs and edge cases.
- Security vulnerabilities.
- Performance bottlenecks.
- Architectural inconsistencies.
👷 task-implementer
A disciplined implementation specialist focused on delivering incremental, high-quality code. It adheres to:
- Pattern-First Development: Matches existing codebase conventions.
- Test-Driven Approach: Ensures every change is accompanied by verification logic.
- Scope Discipline: Limits changes to 1-3 files per task to maintain reviewability.
File Structure
agents/: Instruction sets for specialized sub-agents.commands/: TOML definitions for top-level extension commands.skills/: Detailed procedural guidance for the development lifecycle.gemini-extension.json: Extension configuration and metadata.GEMINI.md: Core mandate and workflow summary for the agent.
Getting Started
- Plan a Feature:
/plan "Add user authentication using JWT" - Implement the Plan:
/impl .agents/sessions/2026-02-26-user-auth - Commit Changes:
/commit "Add login endpoint"
Built for disciplined engineering and high-quality software delivery.