Repository navigation
chore(prepare-pr): hold the skill under its 68000-byte ceiling - #14404
Conversation
…ill under its byte ceiling
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
First Principles Review (Fable 5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing to check. What this change shipsInventory (1 item) — 1 justifiedIntent: get the
[FIRST-PRINCIPLES-REVIEWED] 03919d7 |
Design Review (Fable 5) — 🟡 CONCERNSDesign-level review of Design-Verdict: CONCERNS Stated harm no longer exists at this base: #14398 (the base commit) already trimmed SKILL.md to 67854 bytes, so the ceiling test is green without this PR. Watch
[DESIGN-REVIEWED] 03919d7 |
Opus 5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
|
Simple change to unlock pr-readiness characters count cap. Will follow another PR to get more changes done. |
Problem / Motivation
Goal: Keep the
prepare-prskill under its 68000-byte ceiling so every open PR's Backend Tests stay green.The
prepare-prSKILL.mdgrew to 68015 bytes on main. The testtest_slim_skills_do_not_regrow_duplicate_guidance_or_wire_model_ids[prepare-pr-68000]asserts the file stays at or under 68000 bytes, so it now fails on main and every open PR inherits that red through its merge-ref.Why it matters
A red on main's byte-ceiling test blocks the Backend Tests shard on every open PR, regardless of what the PR changes. Contributors cannot go green until main is back under the ceiling.
Not a goal
What changed (motivation → approach → change)
Symptom: the file is 15 bytes over its 68000 ceiling. Root cause: it sat right at the edge for several revisions and #14181 pushed it over. The skill's own top note says it should carry only what the loop executes — rationale and justification belong in
references/rationale.md.Change: removed the
why that strengthcolumn from the "Two checks, two strengths" table. That column held only justification prose (an unnamed change is how a stray edit rides along,with the ledger complete, a cap cuts only restated facts; the diff is the evidence) — it explained why the two checks pull in opposite directions, which is reader-facing rationale, not an instruction the loop follows. The operational columns (check, what it measures, on breach — including exit 20 and exit 21) are unchanged.Result: 68015 → 67854 bytes, 146 under the ceiling with margin for a future one-line edit.
Backwards compatibility
Compatible: only reader-facing prose was removed; every instruction, exit code, and threshold the loop reads is unchanged.
Tests
No test added — the existing
test_slim_skills_do_not_regrow_duplicate_guidance_or_wire_model_ids[prepare-pr-68000]case is what this restores to green. Verified locally:3 passed, 17 deselected.Manual verification
N/A — the byte-ceiling assertion is the verification, and it passes.
Pattern harvest
Not generalizable: a single skill file drifting past a per-file byte ceiling is a one-off content-size overage, cleared by trimming justification prose the file itself says belongs in a rationale doc.