.png&w=256&q=75)
🤝 Top Collaborators
.png&w=640&q=75)
.png&w=640&q=75)
🤓 Latest Submissions
GECKO
Gecko — Git for AI Coding Agents AI coding agents can plan, write, modify, test, and debug software rapidly. But when multiple agents work on the same GitHub repository, coordination becomes difficult. Two agents may attempt the same task, modify overlapping files, work on dependent tasks in the wrong order, lose progress after failures, or create conflicts that developers must manually resolve. Gecko solves this problem. Gecko is an AI Agent Coordination Layer — essentially “Git for AI coding agents.” It turns a GitHub repository into a live control center for planning, coordinating, monitoring, and reviewing multiple AI coding agents across different developer machines. A hosted Gecko control plane coordinates local Gecko CLIs, while GitHub remains the source of code history and the local CLI handles agent execution. The workflow begins with a GitHub repository and development goal. Gecko can generate a reviewable plan containing tasks, dependencies, acceptance criteria, expected files, and verification commands. Humans can review and approve the plan before agents begin execution, keeping control over what work is created. Developers register local AI agents and configure providers. Gecko supports IBM Bob and OpenAI-compatible providers. The CLI handles initialization, authentication, agent identity, providers, planning, tasks, worktrees, reservations, checkpoints, verification, event synchronization, logging, and pull requests. The core mechanism is atomic task ownership. When an agent claims a task, the Gecko coordination server determines ownership and prevents another active agent from simultaneously claiming it. Ownership is maintained centrally, allowing agents to work in parallel without duplicate execution. Each task can run inside an isolated Git worktree and branch, giving every agent its own working environment. Gecko supports dependencies and file reservations for blocked or overlapping work. During execution, the CLI sends regular heartbeats, watches file changes, records checkpoints, synchronizes events, and stops execution when a task claim expires. Verification commands run inside the assigned worktree, with output captured as evidence. Verified work can be pushed to a GitHub pull request with task and verification context. Failed work is preserved for inspection, retry, or requeue. Target Users Gecko is designed for software developers, engineering teams, and organizations using multiple AI coding agents. The web dashboard shows projects, plans, tasks, dependencies, agents, reservations, events, verification status, and pull requests. What Makes Gecko Different Gecko does not replace AI coding agents. It provides the coordination layer they are missing. Instead of treating agents as isolated coding assistants, Gecko treats them as members of a distributed software team. It combines task ownership, Git worktrees, dependencies, file coordination, heartbeats, checkpoints, verification, event synchronization, human approval, and pull requests into one Git-native system. Our key demonstration uses two AI coding agents on two separate developer devices connected to the same Gecko project. Both attempt to claim the same task: one obtains ownership while the other is rejected because the task is already claimed. The second device can work on an independent task in its own worktree. This demonstrates distributed coordination rather than simply running multiple agents on one computer. IBM Bob was also important in building Gecko. We used IBM Bob to build the Gecko CLI and web application/server, implement and refine functionality, test and debug the system, configure and test IBM Bob as an agent/provider within Gecko, and create the AI-generated project presentation. Gecko transforms multiple independent AI coding agents into one coordinated, observable, Git-native software team—enabling parallel development while maintaining ownership, isolation, verification, synchronization, and human control.
27 Sep 2026
