Repository navigation
docs(docs-hygiene): cut write-for-agents' 'Ask for outcomes, not thinking' paragraph, measured no-loss #12359
Closed as not planned
Description
This was generated by AI.
Parent
Sliced from #11823 (finding 5; see #12272's "Remaining on #11823").
Problem
The #6916 sweep marked a cut to the "Ask for outcomes, not thinking" paragraph of plugins/docs-hygiene/skills/write-for-agents/SKILL.md (from :139 on main, 2026-11-02). The paragraph restates, per model, whether a "think carefully" line helps. Neither #11823 nor #12272 records why the sweep chose this cut; read sweep records 6916-w3 and 6916-w5 before cutting. A cut to an existing skill ships only with eval evidence of no loss (plugins/playbooks/skills/skill-authoring/reference/skill-criteria.md, "Rewriting an existing skill"). Re-read the line against current main first.
Done when
- An eval case exercises the cut text, with a negative control (the passage deleted) that drops clearly below the before text.
- Before and after are measured on VALID runs. The cut ships only on no loss; a loss is recorded here.
Activity
github-actions commentedon Nov 2, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Small
Complexity: Medium
Uncertainty: Knowable
Flags: none
Blocked by: none
Understood: this asks to cut the "Ask for outcomes, not thinking" paragraph from the write-for-agents skill in docs-hygiene. The cut ships only if an eval case with a negative control shows no loss on valid runs. If the eval shows a loss, record it here instead of cutting. The sweep records (6916-w3, 6916-w5) and the current line on main should be read first. No likely duplicates found among open issues.
Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Nov 2, 2026
Not admitted to the issue lane: it carries host-run. Admission takes the issue once each of those is met, when intake triages it again (
.github/actions/resolve-config/admission-rules.json).melodic-automation-lanes commentedon Nov 2, 2026
The latest triage verdict holds nothing: admission moved the issue to
status: confirmed; its comment says why.kyle-sexton commentedon Nov 2, 2026
Claimed by lane 1 (melo-lap-001, hub assignment 2026-11-02 16:5xZ). Plan: read sweep records 6916-w3/w5 and current main first; eval case plus deleted-passage negative control, before/after on VALID runs; cut ships only on no loss.
test(docs-hygiene): name the skill in the three outcomes-not-thinking…
test(docs-hygiene): replace the reasoning-echo case after the smoke pass
docs(docs-hygiene): keep write-for-agents' outcomes paragraph as meas…
kyle-sexton commentedon Nov 2, 2026
Lane 1 result: kept with reason. No cut ships, and the paragraph keeps its current text.
Measured 2026-11-02 18:06-18:08Z on Opus 5.5 (Claude Code 2.1.316). Three new deterministic-grader cases ran in three versions (before, a trimmed after, and a negative control with the paragraph deleted), 3 runs each. All three versions passed
run-validity.py(VALID).case drop. That is under the preregistered 0.15 detection threshold, so this suite cannot measure the paragraph. Underrewrite-a-plugin.mdthe before text stays, and so the done-when's "control drops clearly below" is not met.reasoning_extractionsafeguard refusal at the first turn (3 wordings tried).6916-w3does not mention write-for-agents. The per-finding6916-w5record lived in a worktree that has since been reaped, so why the sweep chose this cut is UNVERIFIED.Re-measure trigger: a model release for which the bare model keeps think lines, or a case the bare model fails. Branch
docs/12359-wfa-cut-outcomes-paragraph(head1eb8666ba6) holds the cases and the run record; no PR, since it changes no skill text.