Find leads
Two tools from the headstart connector, search_x_posts and read_x_profile, plus your own
web search and fetch. A lead is a decision maker at a company that shows a signal and fits
the brief; the person may never have posted, and the signal may be someone else's post. One
run returns ten leads unless the user names a count. Three to five companies examined per
lead is normal.
Six steps, in order. Read the step's file before the step; each step ends on a done-when you
tick in the transcript before the next starts. Open references/signals.md, the catalog of
signal types, when Brief or Sweep names a type, and references/reads.md, the pages that read
and the pages that do not, before the first web fetch.
1. Brief: read references/brief.md
On claude.ai the brief lives in the Project instructions between chats: read it there before
asking, and end the run by returning it extended for the user to paste back. Done when fields
2 (Buys), 3 (Company shape) and 4 (Signals) are answered and the other five are answered or
none.
2. Sweep: read references/sweep.md
Collect signals of the types field 4 (Signals) names. Done when every signal carries its type, its date, its URL, the raw poster and the words that make it a signal.
3. Resolve: read references/resolve.md
Turn each signal into the company behind it. Done when every signal carries a domain or a drop reason.
4. Fit: read references/fit.md
Give each company a verdict against field 5 (Exclusions). Done when every company carries a verdict: in, or out with the rule that fired, and the URLs behind it.
Loop. When Fit leaves fewer companies in than the run asks for, return to Sweep with new angles or another signal type from field 4 (Signals); parked companies carry over.
5. Name: read references/name.md
Find the person in the role field 2 (Buys) names. Done when a person with a public URL is
named, or the company is parked in field 8 (Parked) as fits, person not found.
6. Why: read references/why.md
Write each lead as the block below. Done when every lead is a complete block and every fact
in it carries a URL or the word inferred.
The lead block
1. Name, Role at Company (domain)
Person: public URL (X profile, LinkedIn, or the company site).
Company: size as read, stage.
Signal: type, date, URL.
Fit: one line; the six rules checked, none fired.
Why now: one or two sentences.
Opener context: the fact the customer leads with.
Warm path: the shared investor, batch, employer or community, or none.
Evidence: a URL per claim above, or `inferred`.
The report
Open with one line: how many leads, and the signal types the sweep covered. Then the numbered
blocks. Then the brief, whole, with every company shown added to field 7 (Suppression) and
every fits, person not found added to field 8 (Parked). Close with the spend line:
Spend: N searches, N profile reads, N web fetches; N companies examined for N leads.
When fewer than the count clear Fit and Name after the loop, say how many did and which signal types the sweep did not reach. The list holds leads only; the gap is said, never filled.