End-to-end zcp dev flow for any non-trivial change — analyze, live-verified plan, sliced AFK implementation, full verification, owner retest pack, spec reconciliation. Routes small safe fixes to a LITE path.
Debug with hypotheses and cheap experiments instead of shotgun edits — find the root cause, prove it, then fix it with a regression test.
Restate the problem, surface invariants and edge cases, and cut the work into thin verifiable slices before writing any code.
Before claiming any task done, re-read the full diff, run everything, hunt orphans, and verify each claim you are about to make.
Ship the smallest change that delivers value, keep the tree releasable at every step, and let working software drive the next decision.
Drive every behavior change through a failing test first — red, green, then refactor with the tests as a safety net.
How this project builds and evolves an app for someone who can't review code — on the first build AND every later change, feature, fix, or restyle. Use whenever guided mode is on, however technical or specific the request sounds (e.g. "track my workouts", "make the tickets more kanban-style", "add login"). Read the request, resolve or extend the architecture, write/update a compact PRD + thin vertical slices, build each slice on the dev runtime with tests, verify it, and narrate progress against a live URL. Resolve everything silently; never interview the user with questions they can't answer.
Verified code audit for one Go package — finds real issues, presents max 3, waits for approval before any edits