testsAndMisc/refactor_claude_todo_resume.md

Resume: enforce the 250-line file cap

Paste this whole file to a fresh Claude session opened in ~/testsAndMisc. It is self-contained. Continues refactor_claude_todo.md, which is still the spec — read it too, but the decisions below override it where they differ.

Where things stand

92 files over 250 lines (was 183 — exactly half cleared). Everything is committed and pushed to main at 776e04d; the tree is clean.

python_pkg/ is DONE — 50 violations → 0. Do not reopen it. Cleared one package per tranche, pushing after each: brother_printer (15→0, suite pinned at 402), code_tutor (13→0, 216→217), then android_ui, app_icons, token_audit, focus_policy, random_jpg (57/44/81/137/14), then wsg_grabber's six test modules (277).

linux_configuration is 84 → 49, in eleven batches. All 12 prose files are done (8 split, 2 deleted as completed-work records on the user's ruling). kcd2_dice_solver is 12 → 4, all four remaining being source files. The shell pattern is settled and scripted — read docs/shell-split-recipes.md before touching another one.

Four deferred for shared state across the seam in kcd2_dice_solver and linux_configuration alike (see the table below); the TypeScript ones are search.ts (findBestSet ⇄ hillClimb) and badgeValue.ts (valueBadge ⇄ format). In TS a bad seam compiles clean and fails at runtime — an import cycle leaves a constant undefined at module-init. Always run the suite.

Earlier sessions had already done: the ~/utils exemptions, the no-inline-python hook, CLAUDE.md, poker-stakes/, and every wsg_grabber source file.

git log --oneline --grep='cap' is the authoritative list.

What is left

Directory	Count
linux_configuration	49
phone_focus_mode	12 (do LAST)
kcd2_dice_solver	4
focus_owner	11
billsplit	6
reverse_survivors / meta / docs	4 each
.github / bucket_catch	2 each

Four linux_configuration files are deliberately over the cap, all for one reason: state crosses the seam. Pass it explicitly first, then split; never suppress. See "When to give up on a seam".

File	Why
steam_compatibility.sh (663)	CACHE_MAP assoc array written in a lib, read by main
diagnose_pacman_hook_stall.sh (493)	assigns LAST_ELAPSED the caller reads, reads PACMAN_BIN it defines
clean_audio.sh (419)	probe-result globals threaded through two libs
setup_passwordless_system.sh (374)	no main(); top-level block must be kept by hand

The pre-commit shellcheck hook lints each lib in isolation, with no view of the caller's globals. That is stricter than shellcheck -x on the entry script and is the check that decides whether a seam is real — run pre-commit run shellcheck --files <all touched> before believing a split.

Next: the remaining main()-shaped shell scripts, then kcd2_dice_solver / focus_owner / billsplit (TS+Dart+Kotlin, untouched so far), then the four prose files — zero execution risk.

The live worklist

Do not work from a list in this file — it goes stale. Run:

bash ~/utils/scripts/check_file_length.sh --all

Every count here is a snapshot and will be wrong by the time you read it. Run the script. The table above tells you where the bulk sits, nothing more.

Do the source files in a package before its tests. Proven on wsg_grabber and again on all seven python_pkg packages: the test files then split along seams the source already established, instead of you inventing a new boundary for them. On brother_printer this meant four of the five source splits needed no test edit at all.

Decisions already made (do not re-ask)

1. Full clearance to --all exit 0. When context runs out, write a fresh handoff like this one rather than stopping mid-way.
2. CI checks only files in the push/PR range, not --all.
3. docs/superpowers/plans/** + sessions/** are exempt — already done.
4. poker-stakes/ is tracked — already done.
5. For prose: automate what can be automated, delete the automated lines, keep only HOW, never WHY.
6. Shell verification is bash -n + shellcheck + systemctl cat path checks. Never execute enforcement scripts that mutate the phone or the live system.
7. One commit per logical unit.
8. Editing test files to follow moved code is allowed (2026-08-16), bounded by: no assertion changed, nothing deleted/skipped/weakened, and an identical pass count before and after. See docs/python-split-recipes.md. Do not re-ask this — an earlier session wrongly treated tests as untouchable and stalled on brother_printer because of it.
9. Per-commit gate is pre-commit run --files <changed>; the real git commit is the full gate (it adds jscpd). Push per directory tranche, not per commit.
10. When a seam splits code covered by a per-file-ignore, fix the lint — do not copy the ignore onto the new file (user's call, 2026-08-16). Done for code_tutor/cli.py: S607 fixed with shutil.which, PERF203 by extracting the probe so the poll loop carries no try/except. Both entries deleted from meta/pyproject.toml; the repo now has two fewer suppressions. The which() fix added a branch, so a covering test was written (216 → 217) rather than excluding it. Expect the same question on any remaining per-file-ignore you split through — ask, don't assume.

The constraint that will bite you

Wire the file-length gate LAST, only once --all already exits 0.

meta/scripts/ci_mirror.sh:109 runs pre-commit run --all-files as a pre-push hook, and .github/workflows/pre-commit.yml:105 does the same in CI. Registering the hook while violations remain makes every git push fail until the last file is fixed. Do not scope around it with exclude:/files: — that is the allowlist the spec forbids.

Final commit, once and only once the tree is clean:

- id: file-length
  name: file length <= 250 lines
  entry: bash /home/kuhy/utils/scripts/check_file_length.sh
  language: system

Plus .github/workflows/file-length.yml modelled on ~/todo/.github/workflows/file-length.yml (checks out kuhyx/utils into .utils), scoped to changed files. Then prove it fails: stage a deliberately 251-line file, confirm git commit aborts, delete it.

Language recipes

• docs/python-split-recipes.md — the seam-selection rule, the __all__ re-export form, the TC001 trap, the mixin shape, and the coverage behaviour of new sibling modules. python_pkg is done; read it if a stray .py turns up, or for the seam rule, which generalises to any language.
• docs/shell-split-recipes.md — read this before the next linux_configuration split. What replaces the pass count when there is no test suite, which scripts escape the repo, the lib shape that satisfies both shellcheck and the shebang hook, and six traps found the hard way. The stubbed-run check is scripted at meta/scripts/verify_shell_split.sh; run it on every split whose script ends in main "$@".

Each rule in both cost a failed gate run or a revert to establish.

Split recipes

• Shell — extract into lib/*.sh sourced by a thin entry script. See docs/shell-split-recipes.md for the lib shape; note it corrects the old advice to repeat set -euo pipefail in every lib.
• Python — extract cohesive helpers into sibling modules; keep the public API stable by re-exporting.
• TS/Dart — extract components; re-export from the original path so no importer changes.
• Tests — split by describe/test-group. Coverage must not drop, and verify by the pass count, not a green run: a split that loses a whole module still reports "passed". In python_pkg, shared fixture builders go in the package's tests/conftest.py — the name-tests-test hook rejects any other non-test_*.py file under tests/, so a _helpers.py will fail the gate. python_pkg/syncyomi_guard/tests/conftest.py is the precedent, and meta/pyproject.toml omits */tests/* from coverage by path so nothing there is counted as production code (the opposite of the poker-stakes trap below). Prefer one shared conftest over another partN file: the repo already carries four, and each new one that redeclares the same fixture scaffolding pushes jscpd toward its 2% duplication threshold — which only fires on the real git commit.

Never game the cap: no one-lining, no deleting tests, no moving code into an exempt extension, no suppressions.

Never edit files while a push is running

git push runs pre-commit, which stashes unstaged changes and restores them afterwards. Editing a file in that window makes the restore fail with patch does not apply, and the push aborts — while git push | tail still reports exit 0, because the pipe masks it. This cost two silent failed pushes and one silently reverted split.

Rules that follow:

• Stage (git add) before running pre-commit run --files ... on a work-in-progress tree, or the same stash cycle can revert the edit.
• Push only with a clean tree, and do not start other file edits until it returns.
• After any push, confirm with git status --short --branch that it does not still say ahead N. Never trust the exit code alone.

Traps that cost time in the last session

• Test fixtures and coverage. When splitting a test file, shared helpers must go where the coverage config already excludes them (in poker-stakes that is src/test/**, matched by path). Putting them beside the source makes the coverage tool count them as production code and the 100% gate fails. Move the file; do not add a coverage.exclude entry.
• end-of-file-fixer aborts commits. Files built with sed/cat often lack a trailing newline. The hook fixes it and then fails the commit ("files were modified by this hook"). Just git add again and re-commit.
• Formatters silently revert edits. A PostToolUse formatter stripped an added import twice. After editing, grep the file to confirm the change survived before running anything.
• Contracts. Staging ≥4 code files requires a fresh docs/superpowers/contracts/*.json in that same commit, on top of the per-commit docs/superpowers/evidence/*.json. Required keys: title, objective, acceptance_criteria, out_of_scope, verifier. Validate with python3 meta/scripts/validate_contract.py <file>.
• invariants.test.ts in poker-stakes flakes on its 5s timeout under parallel coverage load. It passes standalone in ~2.6s. Not a regression — do not "fix" it by weakening the test.
• Long commands: background anything over ~60s rather than blocking.

phone_focus_mode — highest blast radius, do it LAST

Two silent failure modes:

1. deploy.sh has a hardcoded flat push list (search for adb_cmd push, around lines 374–383) with no lib/ subdirectory. Every new split lib must be added there in the same commit, or the phone sources a missing file at runtime — gate green, tests green, device broken.
2. config.sh (556 lines) is sourced by 11 scripts. Keep config.sh as the entry point that re-sources its own parts, so all 11 callers keep working untouched.

Static check after each split: grep deploy.sh's push list and every source/. line in phone_focus_mode/*.sh and confirm each new lib appears.

Suggested order

1. The 48 near-misses (251–300) — mechanical, biggest count drop.
2. python_pkg (50) — pytest-coverage gates each changed package on commit.
3. linux_configuration (84) — largest; check systemd unit paths still resolve.
4. kcd2_dice_solver, focus_owner, billsplit, the rest.
5. phone_focus_mode (13) — with the guard above.
6. Gate wiring + CI workflow — last.

Done condition

cd ~/testsAndMisc && bash ~/utils/scripts/check_file_length.sh --all   # exit 0
pre-commit run --all-files                                             # passes

Plus: a staged 251-line file makes git commit fail.