Shep

Shep solves a real pain: running multiple AI coding agents at once without them stepping on each other. The core innovation is worktree isolation—each feature gets its own git worktree, so parallel sessions never clobber files or collide on ports, and your main checkout stays clean. The CI watch feature is a standout: after push, Shep monitors your pipeline and the agent reads failures to push fixes, up to three configurable retries. That closes the loop from prompt to merged PR. Where Shep shines is the spec-driven mode. For complex features, you flag --no-fast and Shep writes requirements, research, and a plan as YAML you approve before any code runs. That approval gate keeps humans in control without slowing down simple tasks. Shep is agent-agnostic by design—it works with Claude Code, Cursor CLI, Gemini CLI, or any terminal agent, and you can swap per feature. That’s a big flexibility win over Cursor’s built-in orchestration, which locks you into Cursor. The downsides are real. Shep is not a standalone AI—you must bring your own agent subscription, and it runs locally with Node 22+, so you need CLI and git comfort. The dashboard is a local web app at localhost:4050, not a SaaS team dashboard. Merging is manual unless you use --allow-merge, which is a feature for safety but a hurdle for full automation. There’s no built-in LLM, so non-developers or teams not using AI coding agents won’t get value. Where it fits: dev teams shipping many features concurrently, solo developers with agent subscriptions, CI/CD teams wanting automated fix loops. Where it doesn’t: non-developers, teams wanting a managed cloud service, or those expecting zero human review.