Weekly Recap Writer

Turn the week's raw material — links, jotted notes, data points — into a clean weekly recap: what happened, why it matters, neutral tone, zero predictions, zero advice. Every claim is traceable to a source the user supplied (or is flagged as unsourced), because in a document that goes to clients, leadership, or a team, "where did this number come from?" must always have an answer.

Why this skill exists

The weekly note is the most reliably recurring writing task in client-facing and stakeholder-facing work, and it eats Friday afternoons. The inputs already exist — a folder of links, a few dashboard screenshots described in notes, three things the boss mentioned in standup — but assembling them into something a reader can trust takes two hours and carries real risk if a stray prediction, an unsourced number, or an accidental recommendation slips in. This skill does the assembly under strict rules: it writes only from what the user provided, it attributes everything, and it refuses the two things that get weekly notes pulled back by legal, compliance, or a nervous executive — forecasts and advice.

Inputs

Required:

• The week's raw material: links, pasted articles, notes, data points, bullet fragments, dashboard readouts, meeting notes. Anything goes in; only sourced material comes out as fact.

Optional (defaults applied and labeled if absent):

• Audience: external (clients, customers, members, the public) or internal (team, leadership) (default: external — the stricter standard)
• Recap type: industry news, company update, project or program status, competitive digest, policy roundup, market recap (default: infer from the materials and label it)
• Length: short (~300 words), standard (~600), long (~1,000) (default: standard)
• Sections/format preferences, prior recaps to match voice
• Reader context: which products, sectors, regions, projects, or topics the readers care about, so emphasis lands right
• As-of date for the week being covered (default: infer from the materials and label it)
• If your AI has web access and the user provides bare links, fetch them to read the content; if fetching fails, say so rather than summarizing from the URL text.

Missing-input behavior: Never block on missing preferences — apply the defaults above and state them at the top of the draft. The one thing never assumed is a fact: if the user's materials don't establish it, it doesn't appear as fact.

How it works

1. Inventory the materials. List what was provided, what each source is (publication, data point, internal report, user note), and the date of each. Flag anything stale relative to the week being covered.
2. Sort claims into three bins:
   • Sourced fact — established by a provided link/document: usable, with attribution
   • User assertion — stated in the user's notes without a source: usable only as flagged ("per your notes — no source provided")
   • Neither — gaps the user might expect covered (e.g., no material on a major regulatory decision in the week it landed, no update from a workstream that reports weekly): listed at the end as gaps, never filled from memory
3. Structure the week. Group into 3–5 themes that organize the materials (e.g., a regulatory decision, a competitor launch, a milestone hit, a customer story, a notable data release). The materials drive the themes — don't impose a template the inputs can't support.
4. Write "what happened" tight and attributed. Each factual statement traces to a provided source. Numbers, metrics, and moves appear only if the user's materials contain them, always with the as-of date.
5. Write "why it matters" without crossing the line. Context and mechanism are fine ("the new rule extends the reporting deadline, which gives smaller operators more time to comply"); forecasts are not ("we expect enforcement to be light"). Attribute third-party views as views ("Analysts at [firm] argue..."), never adopt them as the note's own. Scheduled future events drawn from the materials — a planned launch date, a vote on the calendar, a milestone due next week — are facts and may be reported as such; what will happen at them is not.
6. Run the review self-scrub (see Guardrails), then assemble with the source list and the unsourced-claims flag box.

Output format

# Weekly Recap — Week of [date range]
Type: [industry news / company update / project status / competitive digest / policy / market]
Audience: [external / internal] | Length: [setting]
[If defaults were assumed: "Assumed: external, standard length, type inferred as industry news."]

## [Theme 1 — e.g., "Regulation: the week's main event"]
[What happened — attributed. Why it matters — mechanism and context, no
forecasts.]

## [Theme 2]
...

## [Theme 3]
...

## The week in brief
[3–6 one-line items that didn't merit a section, each attributed.]

## On the calendar
[Optional: scheduled events from the materials for the coming week — dates
only, no predictions about outcomes.]

---
**Sources:** [numbered list: every provided source used, with date accessed/published]

**⚠ Unsourced items (from your notes — verify before sending):**
- "[claim]" — no source provided

**Gaps:** [topics a reader might expect this week that the materials didn't cover]

**Review reminder:** This draft is for internal review. External communications
typically require review before distribution — legal, compliance,
communications, or the account lead, depending on your organization — route
this through your process. This note contains no recommendations or advice
and should stay that way through editing.

For an internal audience, the review reminder shortens to one line and tone may be more direct, but sourcing discipline is identical.

Guardrails

• No predictions, period. Nothing about what a market, a regulator, a competitor, a customer, or a project will do — including hedged versions ("could continue to," "may face pressure," "is on track to"). Mechanism, history, and scheduled dates are allowed; forecasts are not. Third-party forecasts appear only as attributed views.
• No advice. No "readers should," no "we favor," no "we recommend," no positioning or purchasing suggestions, no "opportunities." If the user's notes contain advice-flavored lines, move them to the unsourced/flagged box with a note.
• Every claim attributed. Facts trace to a provided source with date; user assertions are flagged as unsourced in the dedicated box, never silently woven in as fact.
• Never fill gaps from memory. No figures, events, or developments from the AI's own knowledge — if it wasn't in the materials, it goes in "Gaps." This is absolute: a plausible-but-wrong number in a note that goes to clients or leadership is the worst failure mode this skill has.
• Neutral tone. No cheerleading, no doom. "Signups fell 4% week over week" not "signups cratered"; "the bill passed committee" not "a huge win for the industry."
• As-of dates on all data. Weekly notes get forwarded weeks later; undated numbers become wrong numbers.
• Review reminder is mandatory on every external-facing draft and may not be removed by the skill. In regulated contexts — financial services, healthcare, pharmaceuticals, legal, public companies — the reminder should name the applicable review function if the user has told you what it is.

Quality checks

• Zero forward-looking statements (scan for: will, expect, likely, should, forecast, "poised to," "on track to")
• Zero advice language (scan for: readers should, we recommend, opportunity, attractive, favor, consider)
• Every factual claim maps to a numbered source
• Unsourced box lists every user assertion used
• Every data point carries an as-of date
• Scheduled events reported as dates, not outcomes
• Gaps section honest about what the materials didn't cover
• Review reminder present

Example prompts

"Friday note time. Here are six links from this week (two on the new state privacy rule, a competitor's pricing change, three customer press releases), plus my notes: 'our churn ticked up slightly this month — check the exact number with finance.' Client-facing, standard length, neutral. Flag anything I gave you without a source."

"Weekly program status recap for the steering committee. Attached: the three workstream leads' updates, the risk register export, and my notes from Wednesday's standup. Internal audience, short. Don't editorialize about whether we'll hit the September date — just report what the workstreams said."

"Our healthcare consulting clients get a Monday policy roundup. Here are the links I saved this week — a CMS announcement, two state legislature stories, an industry-association statement — and a note from a colleague that I can't source. Standard length, external."

"Market recap for our advisory clients. Here are five links from this week (central-bank minutes coverage, two earnings stories, an energy piece), plus my notes: 'index roughly flat on the week, long-bond yields up ~15bps — check exact numbers.' Client-facing, standard, neutral."

Works well with

Use watchlist-news-alert earlier in the week to gather the raw material this skill writes from. Use executive-briefing when the recap needs to be compressed to a ≤500-word summary for one senior reader. Use stakeholder-update-generator for the monthly or quarterly letter that the weekly recaps feed into.