Skip to main content

change-brief

Write a Change Brief, a compact 6 Ws summary (What, Why, Who, Where, When, How) with the change's Impact and one insight worth remembering, after completing a substantial change or when the user asks what changed, for a PR description, or for the impact of a change. Use it for code, config, infra, data, documents, policies, or any edited work product. "Substantial" means 2+ files, ~25+ lines, or any config, infra, data, security, or user-facing change.

来源信息

仓库
pipeshub-ai/sharp
最近来源活动
2026年10月15日 14:15
检测到的 SKILL.md 语言
英语
星标
4
分支
0

安装方式

默认使用会先检查来源的 Prompt;你也可以切换为直接命令,或下载本地副本。

检查来源文件

决定是否安装前,请先阅读 SKILL.md,以及 SkillsMP 当前展示的配套文件。

正在显示 SKILL.md

SKILL.md
来源说明 · 只读预览
name
change-brief
description
Write a Change Brief, a compact 6 Ws summary (What, Why, Who, Where, When, How) with the change's Impact and one insight worth remembering, after completing a substantial change or when the user asks what changed, for a PR description, or for the impact of a change. Use it for code, config, infra, data, documents, policies, or any edited work product. "Substantial" means 2+ files, ~25+ lines, or any config, infra, data, security, or user-facing change.
when_to_use
After finishing multi-file or risky edits; "summarize the change", "what's the impact", "write the PR description", "what did you do and why".
# Change Brief Give **just the right amount**: the human should understand the change, its risk, and one rule for the next similar situation, in under 30 seconds. Use sharp's detail level (from session context; default **brief**). If the user asks for "more", "details", or a PR description, step up one level. **The Change Brief replaces your usual end-of-task summary. Don't write both.** At most one short sentence may come before it (for example, "Fixed; tests pass 2/2."). Everything else belongs inside the brief, or is left out and given if the user asks for more. ## brief (default): 4 lines, all 6 Ws compressed ``` ### Change Brief **What & why:** <what changed, in user terms> because <root cause, not the symptom>; <how, in a few words if non-obvious>. **Who · where · when:** <who's affected or should know> · <components touched> · <takes effect on merge/deploy/flag/now> **Impact:** <low|medium|HIGH> blast radius · <two-way: rollback = …|ONE-WAY: …> · Verified: <what you ran> · Not verified: <gap or "—"> **Next time:** when <situation you'd recognise> in <context>, <what to do>, because <why>, unless <when it doesn't apply>. ``` ## standard: about 12 lines ``` ### Change Brief **What:** … **Why:** … **Who:** … **Where:** … **When:** … **How:** … (one line each) **Impact:** blast radius · reversibility · behaviour change · risk to watch · verified / not verified (one line each) **Next time:** when …, …, because … ``` ## full Standard, plus rollback steps, alternatives considered and why they were rejected, the list of changed files, and a trade-off line for any conflict resolved. ## Rules (all levels) - **Proportional.** Mechanical changes (renames, formatting, typos, moving files) get ONE line ("Renamed X → Y in 3 files; no references left."), not a brief. - **Conflicts must be named.** If requirements or stakeholders conflicted, "What & why" names both sides and the principle that decided it ("PM wanted X, finance needed Y; one shared calculation satisfies both"). - **Say each thing once.** Never restate the opening sentence inside the brief, or the brief inside "Next time". - **Who = who's affected or should know**, never the requester's name or email. - **Risk first.** A HIGH blast radius or a ONE-WAY door goes in bold on the Impact line. - **Honest verification.** Say "Verified" only for what you actually ran. Untested paths go under "Not verified". This line matters most for the human's judgment. - **Next time** is a rule for a *future, similar situation*, not a fact to memorize or a restatement of the diff. Describe the situation by the signals someone would notice ("a total is off by a cent", "a migration touches a column other services read"), so it's recognisable next time. Good: "Next time: when money math gives stray decimals, convert to integer cents and round once, because binary floats can't represent 0.1." Bad: "Changed money.js to use cents." - **Context matters.** The same action can be right in one context and wrong in another (half-up rounding suits checkout totals but would be wrong for scientific averages). Name the context the rule holds in, and its main exception, if there is one. - **Save it** (silently, one command) if the rule is genuinely reusable, so sharp can bring it back when a similar situation appears: ```bash node ~/.sharp/bin/sharp.mjs lesson add --topic "<area>" --when "<the situation, described by signals you would notice next time>" --context "<where this holds: environment, stakes, constraints>" --unless "<when it does NOT apply>" --principle "<what to do or think>" --why "<the underlying reason>" --source "<the task/file/PR it came from>" ``` - **Conflict resolved?** In brief mode, fold it into Next time ("decided by <principle>"). Standard and full get a separate Trade-off line. - **One learning element per reply.** The brief's "Next time" line *is* that element, so don't add a separate quiz, question or principle on top. If this topic already got a "Next time" earlier in the conversation, drop the line. - **Don't pad.** Drop any W that adds nothing, and never repeat what the user already saw in the conversation. - **Non-code work** (documents, analyses, plans): "where" means sections or audiences, and "verified" means the sources checked. - **Legal:** every citation, quote or statute either appears under Verified (checked against the primary source) or under Not verified, marked UNVERIFIED. Never leave it implicit. - **Learn mode:** leave Next time for the user to write. Ask "What's your rule for the next time this happens?" and refine it with them.
在 GitHub 查看