Checked against Anthropic's own documentation, because this changes often and most articles about it are out of date.
Pro is $17 a month billed annually, or $20 monthly. Max starts at $100 a month.
Claude pricing · Projects · Memory · Skills
Paid plan, about 20 minutes to set up once. A scheduled task fires on Sunday evening, reads your week out of your calendar, notes and files, and has the review waiting before Monday starts. The value is not the summary. It is the pattern you would never notice yourself, like the third week running that Thursday got eaten by meetings.
How to set it up: open Claude Cowork, connect Google Calendar and wherever your notes live, then type /schedule and set it to weekly on Sunday evening. It runs in the cloud, so your laptop can be shut. On the free plan you run the same prompt yourself on a Sunday and get the same review, you just have to remember.
Read my calendar, notes and edited files from the last seven
days. Write my weekly review in four parts.
1. WHAT ACTUALLY MOVED. Finished things only. A meeting about
a thing is not the thing.
2. WHAT SLIPPED. Anything I planned on Monday and never
touched. Say what got in the way, from the evidence.
3. THE PATTERN. Compare this week to the last three. Tell me
one thing that keeps happening that I probably have not
noticed. Be specific and slightly blunt.
4. NEXT WEEK, RANKED. Three things, in the order that matters,
not the order I wrote them down.
Do not congratulate me. Do not pad. If it was a bad week, the
useful version of this says so and tells me why.The honest gate: the quality depends entirely on whether your week actually leaves a trail. If your work does not touch a calendar, a doc or a task list, there is nothing to read and you will get a vague review. Fix the trail first, then automate the reading of it.
Scheduled tasks in Cowork · What Cowork is
Free plan, about 15 minutes, no setup. Upload the PDF and tell it what you think you agreed to. It reads what the document actually says instead. Auto-renewal buried in clause 8, a 90-day notice period nobody mentioned, an IP assignment that quietly covers work outside the project. This is the highest-value thing on the list and it costs nothing.
How to set it up: make a project called Contracts, drop the PDF in, and ask. Claude reads up to 20 files per chat and PDFs up to 1,000 pages, though only the first 100 pages get visual analysis, which matters if your contract is a scan rather than real text.
Attached is a contract I am considering signing. Here is what
I believe I agreed to, in my own words: [describe it plainly,
including the price, the term, and how I get out].
Now tell me where the document disagrees with me.
Go clause by clause and give me:
1. THE TERM LENGTH and whether it auto-renews. Quote the
clause and give the number.
2. THE NOTICE PERIOD to get out, and the exact date I would
need to act by, counting back from the renewal date.
3. WHAT CAN GO UP, and by how much. Price rise clauses,
indexation, "reasonable" adjustments.
4. WHAT I AM GIVING AWAY that I might not realise. IP,
exclusivity, data, the right to be named as a client.
5. WHAT HAPPENS IF IT GOES WRONG. Liability caps, who pays
legal costs, which country's courts.
Rules: quote the exact wording and the clause number for
every point. If the contract is silent on something, say SILENT
rather than assuming a normal default. At the end, give me the
three clauses I should put in front of a lawyer, ranked by how
much they could cost me.The honest gate: this is preparation, not legal advice, and the distinction matters. What it genuinely does is turn "I cannot afford a lawyer to read all forty pages" into "here are the three clauses worth paying a lawyer to look at", which is a completely different bill. A scanned contract with no text layer will extract as nothing, so check it is a real PDF and not a photo.
Claude projects · File and PDF limits
Any plan, about 20 minutes, no setup. Most people say their number out loud for the first time in the actual conversation, which is why it comes out as a question. This fixes that. Claude plays the other side and pushes back the way they really will, three times, until holding the number stops feeling rude.
It works for far more than salary: client rates, supplier terms, a rent renewal, a scope change someone is trying to slip in for free.
I need to rehearse a negotiation. Here is the situation:
[what I want, what I currently have, what I know about their
position and their constraints, and what my actual walk-away
point is].
You are going to play them, not me. Rules for you:
- Be realistic, not cartoonishly difficult. Use the pushbacks
a reasonable person in their position would actually use.
- Push back at least three times before you move at all.
- Use the awkward ones: silence, "that is above our band",
"let us revisit at review", "this is genuinely our ceiling".
- Do not let me get away with justifying my number twice in
one answer. That is the tell that I am negotiating myself
down, and you should point it out when I do it.
After three rounds, stop and give me:
1. Where I weakened, with the exact sentence I said.
2. The one line I should have used instead.
3. My counter, written out, ready to say.The honest gate: it will be a bit too agreeable unless you explicitly tell it to be difficult, which is why the prompt says so twice. And it does not know their real constraints, only what you tell it, so the rehearsal is exactly as good as your briefing.
Free plan, five minutes per decision. The point is not writing decisions down. You have probably tried that. The point is that this one surfaces the old decision when the new one starts to look like it, without you having to remember to go and check. You chose the cheaper contractor in March and it went badly. In August, when you are about to do it again, it says so.
How to set it up: make a project called Decisions and add a new entry each time you make a real one. Memory works on the free plan, so Claude carries the context between chats.
Log this decision, then check it against my history.
THE DECISION: [what I am choosing]
THE OPTIONS I REJECTED: [and why]
WHAT I AM ASSUMING: [the beliefs this depends on]
WHAT WOULD PROVE ME WRONG: [the signal to watch for]
REVIEW DATE: [when to come back to this]
Now do three things:
1. Search what I have logged before. Have I made a decision
with this shape already? If so, show me it and how it
turned out.
2. Name the assumption above that is most likely to be wrong,
and say why.
3. Write the one sentence I would want to read in six months
if this goes badly.
Be direct. I am not logging this to feel organised, I am
logging it so you can tell me when I am repeating myself.The honest gate: it is only as good as your honesty in the entry. If you write the flattering version of why you chose something, it will pattern-match on the flattering version, and the whole thing quietly becomes useless.
Free plan with code execution on, about 30 minutes for the first one. If you have explained the same process to Claude three times, stop explaining it. Describe it once, save it as a skill, and after that it is one command that runs the same way every time. It is also the closest thing to a written operating manual you will ever actually produce, which means you can hand the work to someone else.
How to set it up: turn on code execution in settings, then create a custom skill. The best first one is whatever you did most recently that felt repetitive and slightly annoying.
I want to turn a process I repeat into a skill. Interview me
first, then write it. Do not write anything until you have
asked me the questions.
The process: [name it in one line]
Ask me about:
- The trigger. What has to happen for me to start this?
- The inputs. What do I need in front of me?
- The steps, in my order, including the ones I do without
thinking.
- What "done well" looks like, specifically enough to check.
- The mistakes. What has gone wrong before, and what I now
always check because of it.
- The judgement calls. Where would you need to stop and ask
me rather than guess?
Then write the skill so that a competent stranger could run it
and get my result. Where my answer was vague, say so and ask
again rather than inventing a step.The honest gate: the interview is the work. A skill written from a one-line description is just a prompt with a nicer name. The reason it is worth thirty minutes is that answering "what has gone wrong before" out loud is usually the first time anyone has written your actual process down.
What skills are · Creating a custom skill