BytesFromToby

Executes one slice of the blueprint — writes the code, checks off steps, logs deviations, and stops if the blueprint contradicts the spec. Run after foreman, one slice at a time.

Proves a slice or feature is done by running it. Reads the spec's Done when items, runs the tests that encode them (driving the software where no test exists), captures evidence, stamps the blueprint, and produces a signed report. Runs with fresh eyes — ideally a separate subagent. Run after builder — per slice or final sign-off.

Autonomous maintenance walkthrough — baseline, spec drift (via surveyor), coverage, docs, and a prioritized recommendations list. Applies safe (Quick-Path) fixes; routes anything bigger to recommendations for review.

Interviews the user to fully define a feature or project, then writes the spec. Run at the start of any new feature or project, or to update an existing spec. Also flags (does not perform) a restructure when the spec or a single feature grows too large.

Reads the spec and produces the blueprint — a step-by-step execution plan for the builder, with a committed test planned for every automated Done-when item. Run after architect, before builder.

Bootstrap a project's structure — convention folders, a CLAUDE.md contract (test command, run/demo command, blueprint format), and a starter spec with the "Done when" format. Run once per project.

Static check of code against the spec — finds drift, unbuilt features, undocumented code, and automated Done-when items with no backing test. Reads and compares; never runs the software. Run before a feature, after a refactor, or when spec and code feel out of step.