doctor is one 1,159-line file: model, registry, runner, thirteen checks and renderer with no seam between them

[MongLong0214]

src/commands/doctor.ts is one file carrying the model, the registry, the runner, thirteen checks and the renderer — 1,159 lines with the in-flight #458 check. Adding a check means editing the file every other check lives in, and nothing but convention stops one check from reaching into another. ADR-0032's context section names this file size as part of what forced the ADR.

Why this exists

• PRD §9.4 fixes the layout, and the ADR's consequences section states the outcome: "src/commands/doctor.ts splits into a registry, a runner, per-check modules, and a renderer; the PRD fixes the layout. The command's observable text output for existing checks does not change."
• The contributor story depends on it: one registry entry plus one test against an injected context, in a file of its own.

Scope

Move-only. No logic change, no behaviour change.

• Create src/commands/doctor/ with registry.ts, runner.ts, report.ts, render.ts, and checks/<category>-<id>.ts (PRD §9.4), one file per check:
  checks/runtime-cli-runtime.ts, checks/runtime-git-trailers.ts, checks/transport-notes-refspec.ts, checks/transport-notes-push.ts, checks/capture-commit-msg-hook.ts, checks/capture-hook-runtime.ts, checks/capture-pending-backlog.ts, checks/delivery-inject-runtime.ts, checks/delivery-inject-version.ts, checks/delivery-mcp-lifecycle.ts, checks/history-history-depth.ts, checks/history-squash-conservation.ts, checks/index-index-health.ts.
• src/commands/doctor.ts becomes a re-export shim so every existing import path keeps working (PRD §9.4: "The public import surface (runDoctor, formatReport, register) is re-exported so callers (init, tests) do not change"). The full surface in use today: runDoctor, formatReport, register, evaluateInjectRun, and the types CheckStatus, DoctorCheck, DoctorReport, DoctorOptions — src/commands/init.ts and test/doctor.test.ts / test/doctor-inject-version.test.ts import through ./doctor.js and must not change.
• Comments move with their code — the doc comments on these checks carry the defect history (doctor probed dist/cli.js while distribution ships dist/commitlore.mjs; CI red #40, the hook could be redirected to any executable and doctor still said ok #49, doctor: PreToolUse hook runtime reports exit 127 for a working hook on a binary-only install (checks a synthesised path, not the recorded command) #128, doctor prescribes 'commitlore hooks install' for a stale commitlore.bin, but that command cannot repair it #296, doctor reports a healthy install on a repository where every capture has been silently dropped for 8 days #458) and are part of what is being preserved.

Acceptance criteria

• Layout exactly as PRD §9.4 plus the shim; src/commands/init.ts untouched; every existing test passes unedited.
• Text output and --json byte-identical (the pinned snapshots stay green with zero updates).
• No file under checks/ imports another file under checks/ — the seam is real: a check depends on the model and src/core/, never on a sibling. Enforced by a test that walks the import graph, not by review.
• npm run build still produces dist/commitlore.mjs, and the CI fresh-clone step ("A fresh clone runs, and the plugin entry points resolve", .github/workflows/ci.yml) passes.
• git log --follow/rename detection recognises the moves (move-only diffs, reviewable as such).

Tests that must exist

case	what it catches
no check module imports a sibling check module (import-graph walk over src/commands/doctor/checks/)	dependency-by-import creeping back — the coupling the registry's declared edges exist to replace. This is the fails-if-wrong case: the split that leaves checks entangled has not produced the seam PRD §9.4 is for
text and JSON snapshots unchanged	any behaviour smuggled into the move
the existing full suite, unedited	a broken re-export — init and the doctor tests are the consumers PRD §9.4 promises stability to

Open question — flagged, not decided here

PRD §9.4 names four module files plus checks/, but does not name a home for the model and factory (DoctorCheck, check(...), the unions). A model.ts beside them is the obvious reading; report.ts is arguable. Implementer's choice, stated in the PR, not silently mixed into runner.ts.

Depends on

• One dead dependency surfaces as repeated findings: doctor's two check dependencies are a function parameter and a comment #466 (PRD §9 orders the split after the declared-dependency step; splitting first would move code the collapse ticket is about to rewrite).

Out of scope

• Any behaviour, ordering, rendering, or export-name change; any new file beyond the layout; fixing anything found during the move (file it instead).

Traceability: PRD §9.4; ADR-0032 consequences.