The doc-digest skill: find the part that costs you money

A new client sends you a 14-page contract. You skim it, nothing looks scary, you sign. Eleven months later you find out it renews itself for another full year unless you gave 90 days' notice, and that window shut last month. The clause was one sentence on page nine, written in the same tone as everything else.

The doc-digest skill reads the whole thing so that doesn't happen. It pulls six things: what the document is, the decisions inside it, what it wants from you, every number and date copied exactly, the parts that look like traps, and what a document like this should say but doesn't. Then it gives you five to eight questions in the words you could send back.

It's free and public, one of the 55 skills in our freeskills repo. It works on contracts, reports, research papers, and email threads that have gone 40 messages deep.

Skills live in a folder that Claude Code reads on startup. You copy this one in and it becomes a slash command. Run these three lines:

Then start a new Claude Code session so it picks up the folder. That's it. No API keys, no config, nothing to sign up for.

Type /doc-digest and hand it the document. A file path works for text, Markdown, PDF, Word, CSV, and HTML. Pasting the text straight into the chat works too, and a URL works if your setup has web access. It asks at most one question, and only when the answer would change what it pulls out: are you about to sign this, reply to it, decide something, or just understand it? Signing pulls out terms and traps. Replying pulls out asks and deadlines.

Then it reads. Long PDFs get read in chunks so nothing drops out on page 30. Anything important comes back with the line quoted and the page noted, and anything it worked out rather than read is marked as inferred.

A good first result: a digest saved to ~/digests/ with the document name and today's date, plus the single most important line from the whole document pasted into the reply so you see it without opening anything. The trap list is where the value sits. Each one gets the quoted line and one plain sentence on why it matters. If a clause is normal and fine, it says so instead of dramatizing it.

Save the contract somewhere you can point at, then run this. Swap the path for your own file.

Threads hide the decision somewhere in the middle. Paste the whole thing and ask for what was actually agreed.

For research, the useful question is whether the conclusion is bigger than the evidence. Ask for that directly.

If you'd rather say this out loud than read it, here's what the skill does and why it's worth having.

1. 1Hook: there's a brand-new free Claude skill that takes you from a 14-page contract you were going to skim to the three lines in it that cost you money.
2. 2Mechanism: you point it at the document and tell it whether you're signing, replying, or deciding. It reads the whole thing in one pass and pulls out the decisions, everything the document wants from you, the numbers and dates copied exactly, and the parts that look like traps, auto-renewals, one-sided rights, who owns the work at the end. You get back what it says, what you should do about it, and the questions to ask before you sign.
3. 3Transformation: you go into the conversation knowing the auto-renewal date, the payment terms, and the three questions worth asking. You stop agreeing to things you haven't read.
4. 4Close: it costs nothing, and it's one of 55 in the repo. If you want the path that gets you writing the contracts instead of reading them, that's Claude Code Club, $9 a month. Link below.

Find the longest document currently sitting in your inbox unread. Install the skill and run it on that one tonight. The scary part is almost never written in a scary way, which is exactly why you want something reading every line.