From 891abaae8aafb913d9e6b4b675b63651250468c7 Mon Sep 17 00:00:00 2001 From: alanhc Date: Thu, 3 Sep 2026 18:31:35 +0800 Subject: [PATCH 1/2] Name the reply that hands the choice back The Pull requests section bars pasted walkthroughs and status tables, and says a reply carries the correction, the measurement, or nothing. It does not name the sentence that closes a reply by offering to redo the change if the reviewer would prefer it, which PR#350 rejected on sight. That is the artifact the two earlier citations were already about, in a politer form. It asks the reviewer for an instruction instead of stating a position, and the tradeoff it puts up for a vote had been measured before the reply was written. --- .claude/skills/elfuse-conventions/SKILL.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/.claude/skills/elfuse-conventions/SKILL.md b/.claude/skills/elfuse-conventions/SKILL.md index 05b11a63..9a70bd72 100644 --- a/.claude/skills/elfuse-conventions/SKILL.md +++ b/.claude/skills/elfuse-conventions/SKILL.md @@ -380,6 +380,14 @@ addressed threads with "Resolve conversation"; a reply carries the correction, the measurement, or nothing. A concise what and why belongs in the commit body, not the thread. +A reply that hands the choice back to the reviewer is the same artifact in +a politer form. "If you would rather the source read uniformly one way or +the other, say so and I will move the remaining three" asks for an +instruction where a technical answer belongs ("Let's concentrate on real +discussions!", PR#350). A tradeoff already measured is settled in the +reply, with the measurement as the reason. Ask the reviewer a question only +where the code leaves one open. + The body is intent plus reproduction and commands: for a bug, a minimal reproduction with host macOS and SDK version, hardware, and `make check` status (PR#21, PR#41); for a performance claim, A/B benchmarks on a named From 1804ceb317c4320ed9811cd3757cb149414aab4c Mon Sep 17 00:00:00 2001 From: alanhc Date: Thu, 3 Sep 2026 19:56:08 +0800 Subject: [PATCH 2/2] Ban deferral and self-report in prose The register binds PR bodies and review replies and names the classes that mark machine-written prose. One was missing: text about the writer rather than about the work. PR#350 carried an instance. A paragraph closed a measured decision by offering to reverse it on request, and the maintainer quoted it back with "Let's concentrate on real discussions!". The same shape reaches review text as an apology for a correction and as an announcement of candor before a caveat that stands on its own. An offer to redo the work another way asks the reviewer to choose between two shapes the writer has already measured, and the decision plus the reason that settles it is what they needed instead. A reviewer who disagrees says so without being invited. --- .../skills/elfuse-conventions/references/prose-register.md | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/.claude/skills/elfuse-conventions/references/prose-register.md b/.claude/skills/elfuse-conventions/references/prose-register.md index 673fc18f..3e0f37ec 100644 --- a/.claude/skills/elfuse-conventions/references/prose-register.md +++ b/.claude/skills/elfuse-conventions/references/prose-register.md @@ -36,6 +36,11 @@ State the fact and stop: - Effort and flattery: "carefully reviewed", "comprehensive", "thoroughly tested", "Great catch", "You're absolutely right". Effort is not a finding; name what ran and what it reported. +- Deferral and self-report: an offer to redo the work another way ("say so + and I will", "happy to split this"), an apology for a correction, or an + announcement of candor before a caveat that stands on its own ("worth + flagging rather than hiding"). State the decision and the reason that + settles it; a reviewer who disagrees says so without being invited. - Formatting as emphasis in docs and PR text: bolded bullet-header runs where a paragraph belongs, decorative rules, emoji. - Machine artifacts, defects on sight: zero-width and bidi characters,