The public proof of coding-agent use is unusually direct: the KISS Sorcar paper says the system's own agent has been used to develop itself, and the GitHub repo exposes the engineering receipts behind that claim.
This clears the Builder Work bar because it teaches a transferable skill: write down the invariants your agents must preserve, then make the repo structure, tests, and runtime paths enforce those invariants before scaling agent autonomy.
The Engineering Receipt
The repo's INVARIANTS.md is a long list of durable behavior rules: worktree defaults, running-agent lifecycle, tab restoration, sub-agent isolation, cost accounting, merge behavior, and update rules. The Sorcar runtime code then shows why the list is not decorative. It builds tools for Bash, Read, Edit, Write, web use, skills, MCP servers, model switching, parallel execution, and budget attribution.
Why It Is Not A Five-Minute Demo
A shallow generated wrapper would call a model and stream output. KISS Sorcar carries agent work through git worktree isolation, auto-commit, manual merge or discard, sub-agent usage accounting, UI notifications, chat/web/VS Code surfaces, and a record of what must never regress.
What Builders Can Steal
Use self-hosting as a stress test, not a slogan. Let the agent work on the agent, but keep public constraints close: invariants first, isolated worktrees, repeatable evaluations, explicit cost accounting, and caveats when public proof shows engineering discipline rather than adoption.