计划执行与检查节点
executing-plans
按既定实施计划完成工作,保持过程检查和反馈。
copywriting【效率与协作】Writes and rewrites the words on marketing pages: homepage, landing page, pricing, feature, product and about pages, including headlines, button text, section copy, title tag and meta description. Starts from the reader's own language and the proof the business can show, edits copy where it lives in the repository, and flags every claim that needs ev…
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Page copy has one job: let a particular visitor work out, in the few seconds they give it, what this is, whether it is for them, and what to do next. Usability research has found for decades that most visitors scan a page instead of reading it, and that concise, scannable, plainly worded text is measurably easier to use than promotional prose. So the work is mostly decisions made before any sentence is written: who is reading, what they already know, what can be proved.
The skill returns copy placed where it will live (the component, the Markdown file, the translation file), a short brief that explains the choices, alternatives for the headline and button, and a list of claims the user must confirm before publishing.
Locate the page and any strings it pulls in before asking questions:
grep -rnE "<h1|<h2|title:|description:" src/app src/pages content 2>/dev/null | head -30
ls locales messages i18n 2>/dev/null
Read what is there, including neighbouring pages, so the new text matches the product's existing names for things. Then ask for what is missing:
Copy sounds generic when it is written from the product's vocabulary. Ask for, or read in the repository and shared folders, any of these: support tickets, sales-call notes, reviews of this product and of its competitors, cancellation reasons, forum threads where the audience complains about the problem.
Pull four things out of that material and quote them exactly:
If no such material exists, say so and label the draft as a hypothesis to check against the first five customer conversations.
Seven lines, agreed with the user before drafting:
Reader who, in what role, at what size of company or stage of life
Situation what just happened that made them look
Wants the outcome in their words
Offer the product and the mechanism that produces that outcome
Proof what can be shown, with its source
Hesitation the one or two doubts that stop the click
Action the single thing this page asks for
Compress it into one sentence that the whole page defends, shaped like a Jobs to Be Done job story (situation, reader, outcome) with the mechanism added: "When next week's rota is still unfinished on Sunday night, a restaurant manager wants it done and seen by every employee, and Rotaleaf does that by sending a drag-and-drop rota to each phone." A page that cannot be summarised this way is trying to do two jobs; split it.
| The visitor arrived by | They already know | Open with |
|---|---|---|
| Searching the category ("staff scheduling app") | They need this kind of product | The category named plainly, then what sets this one apart |
| Searching the problem ("staff keep missing shifts") | The pain, not the product type | The problem in their words, then the outcome |
| Clicking an ad | Only the ad's promise | The same promise, in nearly the same words, as the headline |
| An email or a referral | The brand, maybe the product | The specific news or offer |
| Comparing shortlisted vendors | The alternatives | The concrete differences and the proof |
| Visitor's question | Section | What goes in it |
|---|---|---|
| What is this and is it for me? | Headline, subheadline, primary button | Outcome or category, who it is for, the next step |
| Can I believe it? | Proof, straight after the opening block | A number with its unit and date, logos with permission, one named quote |
| What do I actually get? | Three to five benefit blocks | Heading states the result; body states how the product delivers it |
| How hard is it? | Steps | What the first ten minutes look like, numbered |
| What does it cost? | Price or pointer to pricing | The number, the unit, what is included, what is not |
| What if it goes wrong? | Objections, as questions | The real hesitations from the brief, answered without hedging |
| What now? | Closing block | The offer restated in one line, the same button |
Page types change the emphasis, not the logic. A homepage serves several kinds of visitor, so its headline is the broadest true statement and its sections route people onward. A landing page has one audience and one action: remove navigation and every sentence that does not serve that action. A pricing page answers "which plan is mine" (name the customer each plan fits) and "what will I really pay". A feature page starts from the task the feature completes. An about page explains why the company's history makes it a safe choice and still ends with a next step.
Advertising law in the US requires that claims be truthful and that the advertiser hold evidence before publishing them; that covers what a sentence implies as well as what it states. In practice:
[CONFIRM: what is needed] in the draft.Give each page a unique title element and meta description. Google states no length limit for either and truncates to fit the device, so put the distinguishing words first and the brand last; as a rule of thumb, about 60 characters for a title and 150 for a description usually survive. The description is one or two sentences a person would want to click, not a keyword list. Google may show different text, so the visible headline has to work alone.
# stock phrases that carry no information
grep -nEio "seamless(ly)?|cutting-edge|world-class|revolutionary|next-generation|best-in-class|robust|leverag[a-z]*|unlock[a-z]*|supercharg[a-z]*|game-chang[a-z]*|empower[a-z]*|innovative|solutions?" content/home.md
# sentences longer than 25 words
grep -vE "^(#|[[:space:]]*$)" content/home.md | grep -oE "[^.!?]+[.!?]" | awk 'NF > 25 { print NF " words:" $0 }'
Then read the headings alone, read the page aloud once, and confirm each line of the brief appears on the page.
[CONFIRM] list: every claim awaiting evidence.Request: "Write the landing page for our ad group 'restaurant staff scheduling app'. Rotaleaf, $3 per active employee a month, 30-day trial with no card. Three customer reviews attached."
The agent quotes the reviews ("I was rebuilding the rota in Excel every Sunday night"; "swaps happen in the group chat and I find out when someone doesn't show") and agrees this brief:
Reader Owner or manager of an independent restaurant, 8 to 40 staff, builds the rota personally
Situation Sunday night, redoing next week's rota in a spreadsheet
Wants A rota finished quickly that staff really see, and no surprise gaps
Offer Rotaleaf: drag-and-drop rota sent to staff by text link, optional app, swaps the manager approves
Proof 212 restaurants (company figure, June 2026); quote from Inés Calloway, Brasa & Vine
Hesitation "My staff won't install another app"; "I have no time to set this up"
Action Start the 30-day trial
Title tag: Restaurant staff scheduling app | Rotaleaf
Description: Build next week's rota from last week's, send it to every phone and approve shift swaps yourself. $3 per active employee. 30-day trial, no card.
# Restaurant staff scheduling without the Sunday-night spreadsheet
Build next week's rota, send it to every phone, and approve shift swaps before they turn into no-shows.
[Start 30-day trial] No card needed. $3 per active employee after that.
Used by 212 restaurants (June 2026).
## Swaps you approve, not swaps you discover
Staff request a swap in the app. You get one notification, tap yes or no, and the rota updates for everyone.
## Next week's rota from last week's
Copy any week, drag the changes, publish. [CONFIRM: median time to publish a rota, from product analytics]
## "My staff won't install another app"
They do not have to. Everyone gets the rota by text message with a link; the app is optional.
> "I get my Sunday evenings back." Inés Calloway, owner, Brasa & Vine [CONFIRM: written permission]
Alternatives offered: "Stop rebuilding the rota every Sunday night" (leads with the pain; prefer for the problem-worded ad group) and "The rota your staff actually read" (leads with the result for managers whose trouble is no-shows).
Request: "Our homepage hero feels empty. File is src/components/Hero.tsx. Driftwatch watches Postgres schemas."
The check from step 9 finds five stock phrases in two sentences. The user confirms the facts: comparison runs every 60 seconds, alerts go to Slack, one database is free and needs no card. The agent edits the component and leaves the markup and class names alone:
@@ -3,7 +3,9 @@
<section className="hero">
- <h1>The next-generation platform to supercharge your database workflows</h1>
+ <h1>Know within a minute when production stops matching your migrations</h1>
<p className="sub">
- Driftwatch empowers engineering teams to seamlessly unlock visibility into their schema.
+ Driftwatch compares your live Postgres schema with the migrations in your repository
+ every 60 seconds and posts the exact difference to Slack.
</p>
- <a className="btn" href="/signup">Get started</a>
+ <a className="btn" href="/signup">Watch my first database</a>
+ <p className="note">Free for one database. No card needed.</p>
</section>
It reports why: the old headline could sit on any software site; the new one names the event the reader fears and the time to detection, the subheadline gives the mechanism, and the button describes the first step. It adds a title tag ("Postgres schema drift monitoring | Driftwatch", 45 characters) and notes one item to confirm: that "within a minute" holds for the largest customer schema.