The recap that gets a reply
- What we heard, in their words
- Where the problem actually sits
- What we would need to confirm
- One next step, not three
A free Claude skill. It reads the transcript, writes the recap or the full scope, and marks every claim it cannot trace back to something that was actually said.
Discovery summary
Forty five inbound leads a month, two closed. The handoff between the form and the first call is unowned.
We just send them a calendar link and hope. Honestly nobody owns it after the form.J. Reyes, 06:40
| Workstream | Where it came from | Weeks | Owner |
|---|---|---|---|
| Pre call qualification | They asked | 1 to 2 | Us |
| First response inside an hour | We propose | 2 to 3 | Shared |
| Pipeline reporting | We are guessing | 4 | Us |
Open with you
Who is going to read it, and how far it goes. The skill asks you both, every time. A client facing document and an internal one are not the same document with sections hidden, so it writes them separately.
Most tools in this category fill the gaps. They read a call where nobody mentioned a budget and produce a document with a budget in it, because a document with a hole in it looks unfinished. This one leaves the hole and labels it.
No number, date, name, job title or quote that is not in the source. Every claim traces back to a line in the transcript, and the internal version prints the ledger.
Every item is tagged. They asked for it, you are proposing it, or you are guessing. A guess is never written as committed work in something a client reads.
It reports what they said about budget and what the work implies about effort. The number is yours. If nobody said one, the section says so.
Anything it does not know comes out as an open question. A call summary with no open questions is either a transcript or a fiction.
Ten questions on the first run. Your company name, your website, your defaults, who documents go out from, and the one most people skip: what you do not sell. That last answer is what keeps a scope from drifting into work you would have to refuse.
Services, the words you use for them, how you deliver, how long your engagements run, and what you can prove. Your vocabulary ends up in the document instead of ours.
Every line comes with where it came from. You correct it before anything is saved. Nothing the site does not support goes in.
Upload a brand guide, a page whose style you want matched, or just hex codes and font names. Documents come out in your colours and type instead of the packaged ones.
If you have an editorial or content style guide, it writes to yours rather than its own.
A transcript, an export from whatever records your calls, your own notes, or a filled questionnaire. It does not need a recording.
Once for what was said. Once for what was missing. The second read is where the open questions come from.
Client facing or internal. Summary, or summary plus scope. Ask for both and you get two files.
Claims traced, arithmetic re-run, every gap either closed or visible, and a check that no internal material ended up in the client version.
An HTML document that opens anywhere, prints clean, and loads nothing from the internet. Email it, print it, or save it as a PDF.
There is a Present button in the corner of every document. It turns the thing you are reading into a slideshow, built from its own sections, so there is no deck to keep in step with the document. Change one and the other changes.
Give it the next transcript and the earlier document. It marks what got confirmed, what changed, and what is still open, rather than starting again.
A Claude skill is a folder. The download is that folder, zipped.
Most people
Needs a Pro, Max, Team or Enterprise plan with code execution turned on. On a Mac, Safari expands downloads by default. If you end up with a folder instead of a zip, turn that setting off and download again.
If you work in a terminal
~/.claude/skills/.claude/skills/ inside one project.Python 3 is used for the build step. Nothing else is installed.