# LinkedIn poster — a starter prompt for reacting to tools/news

**Why this variant exists:** the oracle→interview→draft pipeline (`01`–`06`) is built for
posts that come from *your own* experience — it needs a transcript because the claims are
first-party. Reacting to someone else's news or tool is a different shape: the input already
exists (an article, a release, a screenshot), so the job is verification and framing, not
interview. Same discipline, different pipeline: **news → verify → draft → board → publish.**

## The flow

1. **News** — pull in a candidate: a tool launch, an industry report, something someone
   posted. Any source works (an RSS feed, a newsletter inbox, a saved search) as long as you
   can trace every claim back to where it came from.
2. **Verify** — before drafting a single word, check every factual claim against its primary
   source. Don't repeat a stat because a newsletter repeated it; find the original report or
   the tool's own page. If you can't verify a number, don't use it as fact — frame it as "a
   report claims" or drop it.
3. **Draft, outcome-first** — lead with what changes for the reader, not with the news itself
   ("a new model shipped" is not a hook; "this makes X ten minutes instead of two hours" is,
   if you can back it). Credit the source lightly by name or handle — never claim someone
   else's finding as your own first-party result. Run the same voice guide and banned-word
   checks as `03-draft.md` / `04-validate.md`.
4. **Board** — put every draft on a simple review board before it ships: a running list (a
   spreadsheet, a Kanban, a single doc) that shows the draft, its source links, and a
   "needs verification" flag if anything's still unconfirmed. **Nothing publishes until a
   human looks at the board and says go.** This is the one hard gate in the whole poster
   flow — automate everything up to here, not this.
5. **Publish** — post it, on whatever schedule fits your account. Keep the source link or
   attribution visible if the post reacts to someone else's work.

## Guardrails

- Never present someone else's finding as your own outcome. If you didn't measure it, it's
  "a report found," "someone put it well," or a light credit to the real handle/outlet — not
  your own first-party number.
- A missing or guessed attribution is worse than no attribution. No verifiable source? Say
  "someone" or drop the claim.
- One human always reviews before it posts. This flow is fast enough to tempt full
  automation — don't. The verification step is exactly where a wrong or outdated claim slips
  through if no one double-checks it.

## What you need to build this yourself

- A source (anything that surfaces candidate news/tools on a schedule).
- A verification habit (or a script that checks a claim against a URL before it's allowed to
  ship) — even a manual "did I check the primary source" checkbox is enough to start.
- A board (even a shared doc works) that a human actually looks at before anything posts.