Watchbill

Available today, free: github.com/Sovereign-Labs-AU/watchbill

Watchbill coordinates many AI agent sessions — and one human — sharing a single working tree for months. It is plain markdown and git plus a few small scripts: no server, no database, no vendor, no account. We extracted it from our own lab, where it has run for roughly ninety days across concurrent sessions from more than one vendor.

The problem it solves

Run more than one autonomous agent against shared work and you reliably get three failures: collision (two agents silently clobber each other), encroachment (an agent wanders into owned work, because nothing said so), and unseen stalls (an agent dies holding work, and nobody notices for days). Rules alone don’t fix this — agents forget across compacted context windows. Visible state does.

What’s in it

• CLAIMS.md — ownership on time-bounded leases that expire visibly; takeover only on the human Operator’s word
• DIARY.md — a live NOW (heartbeat-stamped, stale-flagged) plus an append-only log that is never rewritten
• notebooks/ — one private task-head per session; objectives written before work; reconciled and deleted after
• The instruments — a claims auditor, an ownership guard, a heartbeat, and a test suite whose fixtures are real traps from our production history

Honest scope

Watchbill makes violations visible; it does not make them impossible. A determined or broken agent can ignore a markdown file, and we say so on the front page rather than pretending otherwise. The repo also ships its own pre-release adoption audit — including the findings — at docs/ADOPTION_AUDIT.md.

License

Apache-2.0. Use it commercially, modify it, keep your changes — the license asks only that attribution travels with the code.