[tzhouam]

Purpose

Owner rule request: a PR's code diff excluding tests should not exceed ~1k lines.

REV-3b in knowledge/general/review/rules.md states it, and says the part that is easy to get wrong -- where the number comes from. The host supplies it; the reviewer reports it. A model counting the diff itself would be counting a patch that is truncated for exactly the large files the budget exists for. The enforced counter for the bot's Direct lane is omni-reviewbot#57.

Outside the budget: tests, documentation, lock files and pure renames. Deletions never count. So the rule cannot be satisfied by writing fewer tests, and a cleanup is not punished for its size.

Over budget, the review asks for a split plan or a stated exemption (vendored sync, generated code) on the largest contributing file. A PR body that already gives one is recorded instead. Size stays non-blocking and never raises another finding's severity.

Coverage matches #148: DIRECT_REVIEW_CHECKLIST for Direct consumers, strict-review-checklist.md for Strict (now 7,714 chars against the 8,000 cap raised in #148 -- 286 characters of headroom left, so the next addition to that page needs a trim or another bump).

The contradiction this had to resolve

agent.assess_pr_quality's prompt said in as many words that a large diff is never sufficient by itself -- it would have argued against this rule every time it fired. It now records volume above the budget as a concerns reason in its own right, while the existing code path that requires two grounded reasons keeps size alone from reaching needs_rework. A test pins both halves so they cannot drift apart again.

That assessor never sees rules.md, so its prompt now carries the exclusions and the already-justified exemption itself. Without that, a docs-only PR could collect a size concern for a budget it cannot exceed. (Both points came from the commit review gate; the second caught a real error in my first wording, which said a deletion-heavy PR cannot exceed the budget -- 1,500 added lines are over budget beside any number of removed ones.)

SPEC freshness

engine/steps/review.md said size may never be a needs-rework reason on its own -- still true, and still enforced in code -- but it now also records that volume can stand alone as a concerns reason, and why the exclusions travel in the prompt rather than by reference to the knowledge tree.

Test Plan

PYTHONPATH=$PWD/src python -m pytest -q
python knowledge/tools/check_knowledge_tree.py
python knowledge/tools/check_wiki_lint.py

New coverage: test_quality_prompt_keeps_size_a_concerns_reason_only asserts the budget, the exclusion list, the deletion distinction, the exemption, and that size alone never reaches needs_rework.

Note: PYTHONPATH=$PWD/src is required here -- the shared venv has infermatrix_copilot installed editable against another checkout and this repo's pytest config sets no pythonpath, so a bare pytest silently tests the other tree.

Test Result

Full suite on this branch: 1 failure, test_phase1_cluster.py::test_drift_guard_unimportable_checked_class_fails_guard, identical to origin/main at e40258d3b and to #148.

Knowledge validators: check_knowledge_tree.py 0 errors / 22 pre-existing size reminders; check_wiki_lint.py 0 errors, 0 warnings across 387 pages.

Stacked on #148 (shared checklist page, shared DIRECT_REVIEW_CHECKLIST, and it consumes the 8k injection cap raised there). Merge #148 first.

Not covered: no live review was run, so how often the ask fires on real PRs is unmeasured. The 1,000-line number lives in the rule text on this side and in a code constant on the reviewbot side; there is no single source of truth binding them yet.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JCk5xJGRyydouK7tuhs8kk

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.