Kon

The deciding trait is that a session is a thing you can come back to. Work resumes, continues, hands off to a fresh one, exports, and compacts when it gets long, which means a thread of work survives a night, a meeting, or a crashed terminal. That sounds small until the first time it saves you an hour of re-explaining.

You are the wrong buyer if you want the tool to already know your codebase, because context is layered on when you ask rather than gathered for you. Pick it for continuity. Pick something heavier for omniscience.

Read the headless mode carefully. One flag runs a prompt without a person present and approves the tool calls automatically, and the row records no git operations at all. So the configuration meant for automation is the configuration with no approval step and no commit marking where the changes began. Exit codes are documented, which tells you the run failed and not what it touched.

What it does right is document those exit codes. Most tools in this class return zero and hope.

1. The interesting number is published rather than implied: a core prompt under 270 tokens and a fixed harness around 1,000. Stating that figure invites verification, which is the point of stating it. 2. Project context is layered in on request rather than gathered in advance, so the window holds what the task needed instead of what the tool guessed.

2. That design makes the token cost of the scaffolding a known constant instead of a mystery, and a known constant is what lets a reader reason about cost per task at all.

353 stars and a single author. There is nothing here to invest in, which is a neutral fact rather than a criticism: the project is a demonstration of taste, and taste travels through reputation rather than through revenue.

Moat: none. Pricing power: not applicable at zero. Likely acquirer: none, though the author is the kind of person a lab hires, and that is the most common way a project like this stops. Position: excellent to use, and never the thing sixty people's process depends on.

Two facts decide this for me. It returns documented exit codes, so a pipeline can branch on the result, which is rare enough at this size to note. And the project states that Windows is untested, which excludes a third of my engineers from a tool I would otherwise be asked to standardise on.

Nothing per seat and provider spend uncapped. No SSO, no audit log, no console, no retention policy. Approved with conditions: the platforms it claims to support, and no unattended runs against anything we ship.

The flag I appreciate most is the one that lets me turn the auth check off for an endpoint that never had a key to begin with. That detail tells me somebody actually ran this against a server on their own network rather than assuming every endpoint is a vendor with a billing page.

Both compatible wire formats work, the install goes through a tool manager I already use, and the licence is MIT. There is no MCP, so my servers stay outside, and skills are the extension point I get instead.