screen-locker/todo_fix_pylint_10_prompt.md

Fix TODO — make pylint a perfect 10.00 gate

This file is a ready-to-use prompt. Paste it to Claude, or open this repo and say "do todo_fix_pylint_10_prompt". It is self-contained: everything needed to execute is below. Generated 2026-08-15 from a measured run.

Goal

pylint must score 10.00/10 on this repo, enforced by the pre-commit hook at --fail-under=10 — not the current --fail-under=8.0, which lets real findings through as long as enough clean files dilute them.

Why the current bar is wrong: pylint's hook scores the staged batch as one number. A commit touching many test files fails while the identical code committed alongside a few clean source files passes. The score is a property of which files you staged, not of the code. A 10.00 bar removes that entirely — zero findings is zero findings regardless of batching.

Measured starting point (2026-08-15)

Whole repo (screen_locker/ + scripts/): 8.76/10.

Findings by category, highest first:

count	check	fixable?
473	protected-access (W0212)	no — test-inherent
369	unused-argument (W0613)	no — test-inherent
245	missing-function-docstring (C0116)	yes
238	suppressed-message (I0020)	disappears with W0212/W0613
73	duplicate-code (R0801)	no — test-inherent
66	use-implicit-booleaness-not-comparison-to-zero (C1805)	yes
27	import-outside-toplevel (C0415)	yes, mostly
25	missing-class-docstring (C0115)	yes
25	locally-disabled (I0011)	disappears with W0212/W0613
24	use-implicit-booleaness-not-comparison (C1803)	yes
10	useless-suppression (I0021)	yes — delete the dead disables
~40	long tail (see "Long tail" below)	yes

The decision already taken

Three checks are disabled for test paths only, because they are not defects — they are how pytest works, and fixing them would mean rewriting ~840 call sites and making production APIs public to suit tests:

• protected-access (W0212) — tests drive private mixin methods directly (locker._verify_phone_workout()). This is deliberate: those methods are the unit under test. Ruff already exempts the same thing in tests via SLF001 in pyproject.toml's per-file-ignores, so pylint disagreeing with ruff is just noise.
• unused-argument (W0613) — pytest fixtures such as mock_tk and mock_sys_exit must be requested in the signature to take effect, but are never referenced in the body. Consuming them with a dummy assert to satisfy a linter is strictly worse code.
• duplicate-code (R0801) — near-identical with patch(...) preambles across sibling test modules. Deduplicating them into helpers hurts: a test should read top-to-bottom without chasing a shared fixture.

Everything else gets fixed for real. No # pylint: disable= lines, no per-line suppressions — this repo bans # noqa / # type: ignore repo-wide and the same spirit applies here.

Steps

1. Scope the three disables to tests only. They must NOT apply to screen_locker/*.py or scripts/*.py. pylint has no per-path disable in pyproject.toml, so use one of:

   • a screen_locker/tests/.pylintrc that inherits and adds the disables (pylint picks the nearest rcfile per directory), or
   • two pre-commit hooks (pylint-src with exclude: ^screen_locker/tests/, pylint-tests with files: ^screen_locker/tests/), each --fail-under=10.

   Verify the scoping works: introduce a deliberate protected-access in a source file and confirm it still fails.

2. Fix missing-function-docstring / missing-class-docstring (270). Docstrings must say what the test pins down, not restate its name. Good: """An unsigned entry is rejected once a key is available.""" Bad: """Test rejected when key available.""" A generator that drafts from the test name lives at /tmp/.../scratchpad/add_docstrings.py in the session that wrote this file; it is refactor tooling, not a deliverable — rewrite it if it is gone. Every generated line still needs reading: the draft is a starting point.

3. Fix the booleanness checks (94). assert x == [] → assert not x, assert n == 0 → assert not n, assert s == "" → assert not s. Careful: only where the value really is a sequence/int/str. is False / is True assertions on tri-state returns must NOT be collapsed.

4. Fix import-outside-toplevel (27). Hoist each to module scope unless it exists to break an import cycle — where it does, that is a real constraint; restructure the modules so the cycle is gone, or leave it and note why.

5. Delete useless-suppression (10) — dead # pylint: disable= comments that no longer suppress anything.

6. Long tail (~40). too-many-instance-attributes, too-many-arguments, too-many-return-statements, unspecified-encoding, line-too-long, comparison-with-callable, superfluous-parens, pointless-string-statement, reimported, consider-using-with, unreachable. Fix each on its merits.

   • unspecified-encoding (3) is a genuine latent bug — add encoding="utf-8".
   • comparison-with-callable (3) is often a real bug (comparing a function instead of calling it). Read each one carefully.
   • unreachable (4) is sys.exit(0) followed by return in _startup_checks.py and elsewhere. Those returns are load-bearing under test: conftest.py patches sys.exit to a no-op, so without the return execution continues into the next branch. Do not delete them blindly — restructure so the intent is expressed without dead code, or leave them and record the reason.
   • too-many-ancestors (1) is ScreenLocker's 19 mixins. Inherent to the design; raise max-parents in config rather than collapsing mixins.

7. Flip the gate. Set --fail-under=10 in .pre-commit-config.yaml (both hooks if you split them) and update the hook name: to say 10.

Done condition

• pylint --rcfile=pyproject.toml screen_locker/ scripts/ prints rated at 10.00/10.
• pre-commit run --all-files passes.
• A deliberately-introduced pylint finding in a source file makes git commit fail (proves the gate is live and correctly scoped).
• python -m pytest still green at fail_under = 100 branch coverage.
• No # pylint: disable= added anywhere as part of this work.

Verify

Run the suite, then actually run the locker entry point — python -m screen_locker.screen_lock --status — because this repo deploys from the working tree, so a broken import is a live outage. Hoisting imports (step 4) is exactly the change that can cause one.

Do not

• Do not lower --fail-under below 10 once it is raised.
• Do not add per-line # pylint: disable= comments to reach the number.
• Do not disable W0212/W0613/R0801 globally — tests only. A source file calling another class's private attribute is a real finding.
• Do not touch refactor_claude_todo.md's 250-line work; that is a separate, in-progress task with its own prompt file.