A real application, built to a regulated standard — almost entirely by AI.
foragerr is a self-hosted manager for a comic library you own. Its code, tests, and documentation are written almost entirely by Claude, under a spec-driven, requirement-traced process — directed, reviewed, and approved by a single human operator. This site is the record of how it's built.
Discipline is the product, as much as the software is.
Four practices, borrowed from regulated engineering, applied to an AI-authored codebase.
Every behavior foragerr is meant to have is written down first, with a stable ID. IDs are allocated in the registry when a change is proposed, and a commit hook rejects any commit citing an ID that is not registered.
Each requirement links to the spec that defines it, the tagged tests that prove it, and the commits that built it. The matrix is regenerated from repository state — spec headings, test tags, commit trailers — and inconsistencies fail the run.
The owner approves every proposal before implementation begins, and no change merges until its gate passes: the full test suite green, traceability and dependency checks at zero, and a multi-angle review of the diff.
Every tagged release carries the same four artifacts: a CHANGELOG entry citing the requirements it moved, an annotated tag, the pushed history, and a GitHub Release. Tags are immutable — a mistake gets a new patch release, never a moved tag.