AI Guides › Playbooks
By Nigel Guy · 7 min read
Most people use Claude as a clever stranger. Each chat starts cold, you re-explain who you are and what good looks like, and then you conclude the tool is "fine for emails". The fault is not the model. It is that nothing you worked out last week was saved anywhere Claude can use it.
A note on provenance: this guide was commissioned from a social-media post about "five smart systems" built with Claude. We could not verify what that author built, so we have not repeated it. What follows is five systems you can build yourself from Claude features we checked against Anthropic's own help pages on 2026-10-04.
The rule: a system is anything that saves a decision so you never make it again. Build one rung at a time, and only climb when the rung below has paid for itself twice.
| Rung | System | Built from | What it saves you |
|---|---|---|---|
| 1 | The Standing Brief | A Project with instructions and files | Re-explaining yourself |
| 2 | The Recipe | A custom Skill | Re-specifying a repeated task |
| 3 | The Weekly Pull | Connectors to your own apps | Copying data into chats |
| 4 | The Decision Log | A Project file you append to | Re-arguing settled questions |
| 5 | The Working Page | An artifact (small tool or dashboard) | Rebuilding the same view by hand |
Create a Project and put three things in it: project instructions (role, audience, tone, what to refuse), reference files (your style guide, price list, past good work), and nothing else. Anthropic's help page says projects are available on all plans, with Free limited to five. On paid plans, projects use retrieval automatically when you approach the context limit, so large knowledge bases still work.
Check the brief by asking a cold question and seeing whether the answer sounds like your business without prompting.
You are helping me set up the standing instructions for a Claude Project.
Context: I am [YOUR_ROLE] at [YOUR_BUSINESS]. My audience is [AUDIENCE]. My writing tone is [TONE_WORDS]. Things I never want: [BANNED_HABITS].
Goal: write project instructions of under 300 words that make every future chat in this project behave like a well-briefed colleague.
Steps:
1. Ask me up to five questions about anything above that is vague. Wait for my answers.
2. Draft the instructions in this order: role, audience, tone, default output format, hard constraints, what to do when information is missing.
3. Add the line "If an input is missing, ask for it rather than guessing."
4. Before showing me, check that nothing contradicts itself and that every rule is testable.
Output: the instructions in a code block, then a three-line list of files I should upload to support them.
Fill in the role, business, audience, tone and habits.
When you catch yourself pasting the same long prompt a third time, turn it into a Skill. Anthropic describes skills as folders of instructions, scripts and resources Claude loads when relevant; the only required file is SKILL.md, and a simple skill needs no code. Skills are on Free through Enterprise plans, appear under Customize, then Skills, and need code execution enabled.
Keep each recipe to one job. Anthropic's own guidance is that several focused skills compose better than one large one.
You are an expert at writing Claude Skills (SKILL.md files).
Task: turn the repeated job below into a skill.
Job: [JOB_IN_ONE_SENTENCE]
Input I usually give: [TYPICAL_INPUT]
What a good result looks like: [GOOD_RESULT]
One example of a past good result: [PASTE_EXAMPLE]
Steps:
1. If the job, input or example is unclear, ask me before writing.
2. Write a SKILL.md with a name, a description saying exactly when it should trigger, numbered steps, the output format, and a short checklist Claude runs before replying.
3. Add two things the skill must never do.
Output: the complete SKILL.md in one code block, then two test inputs I can use to try it.
Fill in the job, input, result and example.
Connectors let Claude read from apps you already use. The help page gives the path as Customize, then Connectors, then the plus button, and says connectors respect your existing permissions: if you cannot open a file, Claude cannot reach it through the connector. Free accounts are limited to one custom connector.
Start read-only. Ask Claude to summarise, not to send or edit.
You are my weekly operations analyst.
Context: I run [BUSINESS_TYPE]. Use only the connected sources I name: [CONNECTED_APPS]. Time window: [DATE_RANGE].
Goal: a one-page weekly review I can read in three minutes.
Steps:
1. List what you found in each source, with the item name so I can check it.
2. Group into: needs a reply, needs a decision, waiting on others, can be ignored.
3. Propose the three actions most worth my time, with a one-line reason each.
Constraints: read only; do not send, edit or delete anything. If a source returns nothing or you cannot access it, say so plainly rather than filling the gap.
Self-check: confirm every item in your summary came from a source you can name.
Fill in the business, apps and dates.
Keep one text file in your Standing Brief called decisions. After any meaningful choice, have Claude write a dated entry: the decision, the reason, what would change your mind. Upload the updated file to the project. Future chats then start from your settled positions instead of reopening them.
You are the keeper of my decision log.
Context: [DECISION_JUST_MADE], made on [DATE], because [MY_REASONING]. Alternatives I rejected: [OPTIONS].
Write one log entry with these fields: date, decision, reason, rejected options, what evidence would make me revisit it, and a review date. Maximum 120 words. If any field is missing, ask me instead of inventing it.
Then check the entry against the existing log pasted below and flag any contradiction.
[PASTE_EXISTING_LOG]
Fill in the decision, date, reasoning, options and existing log.
An artifact is, per Anthropic, something Claude makes that you could put in front of someone: a document, dashboard or small interactive tool, opened beside the chat, editable and shareable. Artifacts are on all plans. Use one when you keep asking for the same view, such as a pricing calculator or a pipeline tracker.
You are a careful front-end builder making a small single-page tool.
Purpose: [WHAT_THE_TOOL_DECIDES_OR_SHOWS]. Users: [WHO_USES_IT]. Inputs: [FIELDS_AND_UNITS]. Output: [RESULT_FORMAT].
Steps: confirm my inputs and the formula, then build it, then list three test cases with expected answers so I can check the arithmetic.
Constraints: no invented figures; any assumption goes in a visible notes line; works on a phone screen. Ask if a formula is missing.
Fill in the purpose, users, inputs and output.
At time of writing, Anthropic lists Claude Pro at $20 a month ($17 on annual billing), about £15 a month at time of writing, and Max from $100 a month, about £75; check the £ price at checkout, as it varies with exchange rate and tax. Rungs 1, 2, 3 and 5 are listed on the Free plan too, with limits (five projects, one custom connector). Heavier use will hit usage caps sooner.
Say you run a small bakery's online shop. Rung 1: a Project holding your voice notes and allergen policy. Rung 2: a "supplier chaser" recipe. Rung 3: Friday's read-only summary of orders and emails. Rung 4: a log entry, "stopped selling gift boxes in December: packing time". Rung 5: a wholesale quote calculator. Each took one sitting, and none needed code.