Consolidate VALID_CATEGORIES; delete dead prompt_loader (TECHNICAL-REVIEW §4.1, §4.2)

[mmcky]

Summary

Addresses §4.1 and §4.2 of TECHNICAL-REVIEW.md:

• §4.1 — Delete style_checker/prompt_loader.py. It was dead production code: only the test suite imported it. The runtime path (reviewer.create_single_rule_prompt) loads prompts directly from disk.
• §4.2 — Consolidate the four (!) sources of truth for the 8-category list into one tuple in style_checker/categories.py.

What changed

Before: the category list was defined in four places that could drift silently —

• cli.ALL_CATEGORIES (list)
• GitHubHandler.VALID_CATEGORIES (set, class attribute)
• PromptLoader.VALID_CATEGORIES (list, on the dead class)
• reviewer.RULE_EVALUATION_ORDER keys (dict)
• plus a duplicate hardcoded list inside review_lecture_smart

Test coverage only verified that two of those four agreed.

After: one tuple, in one file:

# style_checker/categories.py
VALID_CATEGORIES = (
    "writing", "math", "code", "jax",
    "figures", "references", "links", "admonitions",
)

RULE_EVALUATION_ORDER still owns the per-category rule ordering, but a new test (test_rule_evaluation_order_keys_match_valid_categories) asserts its keys are exactly VALID_CATEGORIES in the same order — drift fails CI loudly on the next edit.

Files

Change	File	Notes
NEW	style_checker/categories.py	22 lines — the single source
DELETE	style_checker/prompt_loader.py	~230 lines, unused at runtime
DELETE	tests/test_prompt_loader.py	Tested the deleted class
modified	style_checker/cli.py	Import VALID_CATEGORIES; drop ALL_CATEGORIES alias
modified	style_checker/github_handler.py	Drop class-attr; module-level import
modified	style_checker/reviewer.py	Drop hardcoded list in review_lecture_smart
modified	tests/test_cli.py	Import from canonical location
modified	tests/test_parsing.py	Drop fixture's class-attr setup
modified	tests/test_reviewer.py	Import from canonical + add drift-detection test

Net diff: +66 / −360 across 9 files.

Test plan

• Unit tests pass: uv run pytest tests/ → 85 passed
• All modules import cleanly
• Every call site of VALID_CATEGORIES resolves to the canonical tuple
• New test_rule_evaluation_order_keys_match_valid_categories catches drift between the two definitions
• Smoke test via test-action-style-guide issue TASK: Evaluate individual style-guide rules to improve #1 (post-merge)

Why now

Surfaced during the deep technical review (PR #15 was the first batch — P0/P1 correctness bugs). This is the next-priority architectural cleanup item: low-risk, mostly mechanical, and removes 4× the surface area for the "I forgot to update one of the lists" class of bug.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR removes unused prompt-loading infrastructure and centralizes the style-checker category list into a single canonical location, reducing the chance of category “drift” across the CLI, GitHub Action, and reviewer logic.

Changes:

• Added style_checker/categories.py to define VALID_CATEGORIES as the single source of truth.
• Deleted dead PromptLoader code and its dedicated test module.
• Updated CLI, GitHub handler, reviewer, and tests to import and use VALID_CATEGORIES, plus added a drift-detection test to keep RULE_EVALUATION_ORDER aligned.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
style_checker/categories.py	Introduces canonical VALID_CATEGORIES tuple and documents intended invariants.
style_checker/prompt_loader.py	Deletes unused prompt loader implementation.
style_checker/cli.py	Switches category validation/defaulting to VALID_CATEGORIES.
style_checker/github_handler.py	Removes class-level category set; validates against imported VALID_CATEGORIES.
style_checker/reviewer.py	Replaces hardcoded default category list with VALID_CATEGORIES.
tests/test_cli.py	Updates tests to reference VALID_CATEGORIES instead of ALL_CATEGORIES.
tests/test_parsing.py	Simplifies fixture setup now that category validation is module-level.
tests/test_reviewer.py	Replaces PromptLoader.VALID_CATEGORIES usage and adds drift-detection test for RULE_EVALUATION_ORDER keys/order.
tests/test_prompt_loader.py	Deletes tests tied to removed PromptLoader.

[mmcky]

Expanded scope to address Copilot's feedback at the root cause. New commit 01c50cf pulls in TECHNICAL-REVIEW §1.4 (roadmap §4.6) — consolidating the 8 byte-identical prompt files into a single shared prompts/prompt.md.

Why expand scope

Both Copilot comments pointed at the prompt-file-per-category invariant:

• Comment 1: "docstring should mention {name}-prompt.md is required"
• Comment 2: "we lost the test that verified each {name}-prompt.md exists"

Rather than document/test an invariant that the roadmap already flagged as work-to-remove, this commit removes the invariant entirely — one shared prompt.md, no per-category prompt files at all. The category arg on create_single_rule_prompt is kept (unused) so future callers don't need to change if a category-specific prompt is ever reintroduced.

What's in 01c50cf

Change	Notes
style_checker/prompts/prompt.md	New canonical single shared prompt
style_checker/prompts/{8 categories}-prompt.md	Deleted (were byte-identical)
style_checker/reviewer.py	create_single_rule_prompt now loads the single file
tests/test_reviewer.py	New TestPromptFile class with test_prompt_file_exists + test_prompt_file_has_version_header
6 doc updates	architecture, contributing, testing, roadmap, extended-thinking, tests/README — all stale references to the 8-file structure and to prompt_loader.py are removed

Test status

• 87 tests pass (was 85; +2 new prompt-file tests)
• Verified create_single_rule_prompt('writing', ...) == create_single_rule_prompt('math', ...) — consolidation produces identical prompts as intended
• CI green on the previous commit; new commit waiting for CI

Diff stats

This commit: +87 / −346 across 16 files (mostly deletions of duplicate prompt content + doc cleanup).
Full PR (both commits): +153 / −706 across 17 files (~5× more code removed than added).