Software Factory

>the_stations

Six Claude Code sessions, launched together, each briefed into a lane it owns and a set of paths it must not touch.

Every station is pinned to an exact model version in the registry — never a moving alias — shown here in short form. Each station's model is verified accepted by a live call before it lands, the same rule the product's own model catalog follows. The conductor drops to low effort deliberately: it reads JSON on a timer and never writes code. The scarce resource is shared rate-limit budget, not dollars.

>the_three_that_live_in_github

Not launched with the others, not on the laptop, and easy to forget — which is backwards, because one of them is the stage between "an agent wrote it" and "a human merges it".

• The reviewer. Reads a pull request against the original wording of the issue it claims to close — skipping drafts, bot and docs-only changes — offline and read-only, and fails the check if it has nothing to say. It runs a different model family on purpose: every station is a Claude agent, and a reviewer that shares the authors' blind spots is not a reviewer.
• Error triage. Files one deduplicated ticket per new production error, with evidence attached, on a schedule.
• Dependency updates. Opens the upgrade pull requests.

Two of those three run no model at all. They are deterministic — and they are the two that never surprise anyone.

>how_it_holds_together

• One registry. A single JSON file is both the source of truth and the configuration: who exists, which model, which effort, which working directory, which inbox — and the exact prompt each agent is launched with. Adding a station is a registry entry, a card and a brief, never an edit to the launcher.
• A card per station. What it owns, what it must not touch, what it publishes to peers, what it needs from them — and, load-bearing, what it may do on a peer's word versus what it must bring to a human first.
• A conductor, not a manager. This factory runs on frozen contracts — a station announces, the consumer acknowledges, then it lands — so stations block on each other by design. Which means its characteristic failure is not a dead agent but two healthy agents out of time with each other, and a liveness check reports that as six green lights. The first health run found nineteen live unanswered handoffs, the oldest about five and a half days, and it was a contract acknowledgement — while every station was alive and busy. The metaphor draws its own limit: a conductor cues entrances, it does not write the score. It never assigns work, sets priority, restarts a station, or merges anything.
• One report format. Every agent reports as DECIDE → DO → FYI. Rulings first, because they are blocking another agent and cost a human seconds to answer; then work only a person can do — secrets, production, a physical device; then what happened. The rule that gives it teeth: a decision an agent took that was the human's to take goes in DECIDE, retroactively. Nothing gets moved to FYI because it is awkward.
• A bridge, not a socket. The stations coordinate over a local agent-to-agent bridge modeled on Google's Agent2Agent (A2A) protocol — the same vocabulary and task lifecycle: agent cards, tasks moving through submitted, working, completed, failed or input-required, message parts, artifacts. It runs entirely on one machine over append-only files: no agent servers, no HTTP, no discovery endpoints, nothing leaving the laptop. That is also why it survives a crashed agent, a closed terminal and a reboot — the medium is files. Borrowing A2A's shape rather than inventing one means it can port to real A2A transport later without redesigning the conversation.