Your AI coding sidecar
Organizing every agent thread into a task tree, summarizing progress, and guiding (or automating) the next steps.
The architectural overview.
Handle Terminal automatically until done or paused.
You are not alone
When AI coding scales, control breaks first.
One or two threads are manageable. Ten or fifty are not. Visibility drops, status drifts from reality, and your next action becomes guesswork.
- You lose track of who is doing what across concurrent agent threads.
- Task status and terminal reality diverge after context switches.
- The moment you step away, progress can stall with no clear owner.
What is Shellman? Shellman is an AI coding sidecar that turns scattered agent conversations into a structured execution system.
Feature system
Build order from chaos with a TODO Tree + Sidecar Modes.
TODO Tree
Parallel agent work fails first at visibility. TODO Tree restores control with one operational view: parent-child structure, runtime state, flags, and priority in a single scan.
It turns the left sidebar from a passive list into an execution surface, so your next action is always obvious.
Sidecar Modes
- Advisor: use it during planning or risky edits when you want ideas and review, but keep every terminal action under your explicit control.
- Observer: use it when many threads run in parallel and you need Task Tree to stay trustworthy without constantly checking each terminal.
- Autopilot: use it for execution-heavy delivery windows where routine steps should run continuously until done, with automatic progress reporting back to the team.
Browser Access
Access your workspace from any browser, not just a fixed local setup.
Continue task coordination from anywhere on PC or mobile, with the same operational context.
Best practices
Align team behavior with AGENTS-SIDECAR.md.
Use the file as your operational contract for sidecar collaboration: role boundaries, control-plane discipline, and coordination rules across Codex processes.
./AGENTS-SIDECAR.md