How the pipeline actually works

01

The team, rebuilt as agents

Four delivery roles, each scoped to one job and one artifact.

Open → 02

The pipeline

A fixed six-station line stamps an artifact on your ticket at each stop.

Open → 03

Orchestrator & specialists

A hub dispatches to specialists and reports gaps — it never improvises.

Open → 04

Handoff artifacts

The baton has to be well-formed, or the next leg refuses to start.

Open →

05

Human gates

Only one keyword opens the gate — no amount of clever phrasing does.

Open → 07

The review loop

Review and rework repeat with a cap — a stuck change reaches you.

Open → 08

What stays a human job

A split checklist: what a machine can verify, and what needs judgment.

Open →

06

Standards as guardrails

One rulebook scans every diff — blocking gets fixed, warnings get surfaced.

Open → 09

Right-sizing the model

Mechanical checks route cheap; judgment calls route to a stronger model.

Open →

10

Same workflow, different runtime

One blueprint, two machines — the tool underneath is swappable.

Open → 11

Resume, don't restart

A state file bookmarks every stage — a crash never repeats finished work.

Open →