AgentGit

This Skill addresses the pain point of "trusting sub-agent self-reported results" in multi-agent collaborative development. In practice, multiple sub-AI agents generate code, configurations, or documentation on separate branches, claiming "task completed," but lack objective evidence. AgentGit introduces a mandatory gating mechanism—merging is only allowed if the exit code of a real command (e.g., test, lint) is 0—completely eliminating the failure mode of relying on sub-agent self-reporting.

Usage is straightforward: simply run agentgit init in the project root to initialize, then create an independent task for each sub-agent's assignment: agentgit create-task "description". After a sub-agent completes work on its automatically created branch, run agentgit validate <task-id> to trigger the pre-configured validation command (e.g., npm test). If the command exits with code 0, the task is marked as approved; otherwise, it's rejected. Finally, only approved tasks can be merged into the main branch via agentgit merge <task-id>.

It is ideal for teams or individuals who need to coordinate multiple AI coding agents, ensure code quality, or have strict compliance requirements. This is especially useful in development scenarios where AI-assisted code generation is already used but there's a struggle to automatically verify the correctness of outputs—for example, gating code, configuration files, or documentation generated by sub-agents through automated tests or lint rules before merging into the main branch.

It is recommended to configure validation commands (e.g., test suites, static analysis, schema validation) in .agentgit/config.yaml early in the project. Note that it primarily relies on Node.js and git environments, and validation commands must reflect results via exit codes. If the project lacks suitable automated checks, prioritize adding them rather than skipping the validation step.