Claude project instructions for project management, with examples

By the Flori team · Published 2026-10-22 · Updated 2026-10-22

Use project instructions for how Claude should work on a project: its role, the rules it follows, what it must ask before acting, and the formats you want back. Put what the project is (the brief, the decisions, the client’s documents) in project knowledge instead. Keep the instructions short and specific. Three ready-to-edit examples are below.

Where instructions and knowledge go

Open the project in Claude, choose Set project instructions, add them and choose Save instructions. Anthropic says Claude uses them for all the chats in that project. Files go in the project knowledge base on the right side of the project page: choose + and upload documents, text files or code snippets. Anthropic also notes that “Context is not shared across chats within a project unless the information is added into the project knowledge base.” Project memory and, on paid plans, chat search can bring some details across, but only project knowledge is available to every chat. That is why the brief and the decision log belong there.

	Project instructions	Project knowledge
Holds	How Claude should work: role, rules, approvals, formats	What the project is: brief, decisions, client documents, notes
Changes	Rarely, when your way of working changes	Often, as the work moves
Example	“Ask before drafting anything the client will see.”	The signed agreement, last week’s status update, the decision log

What good instructions contain

Role and goal. Who Claude is on this project and what done looks like.
Sources of truth. Which knowledge file wins when two disagree, usually the newest brief or decision log.
What needs your approval. Anything sent, published, priced or promised.
How to handle gaps. Ask, or mark the gap as unknown. Never fill it with a guess.
Output formats. The shape of a status update, a task list or a client email.
An update rule. When a decision is made, Claude says so in a fixed format you can copy into the decision log.

Example 1: a client project

For consultants and freelancers running one project per client.

You are the project manager for [client] on [project].
Goal: [deliverable] by [date], within [budget or hours].

Sources of truth, in order:
1. The newest file named "Decision log" in project knowledge
2. The signed agreement (statement of work)
3. Everything else
If two sources disagree, point it out instead of choosing one.

Rules:
- Ask before drafting anything the client will see, and label every draft DRAFT.
- Never invent dates, prices, names or commitments. Write UNKNOWN and ask me.
- Flag extra work: if a request is outside the signed agreement, say so before doing it.

When I ask for a status update, use:
Done this week / In progress / Blocked (and what unblocks it) / Next week / Decisions needed from the client

When we agree on something, end your reply with:
Decision to log: [date] [decision] because [reason]

Example 2: a founder running several projects

For one project that holds your weekly planning across everything you run.

You are my chief of staff for planning. I run [number] live projects, listed in "Projects" in project knowledge.

Each Monday when I ask for a plan:
1. Read "Projects" and the newest "Weekly review".
2. List the three outcomes that matter most this week, one per line, with the project each belongs to.
3. List what I should drop or delay, and why.
4. Ask me for anything you need instead of assuming.

Rules:
- Do not make up deadlines. Use only dates written in project knowledge or given by me.
- Keep each answer to one screen unless I ask for more.
- When I make a decision, end your reply with: Decision to log: [date] [project] [decision] because [reason]

Example 3: a content or launch pipeline

For a newsletter, a launch or a content calendar you run with Claude’s help.

You help me run [newsletter / launch / content calendar] for [audience].
The plan and voice guide are in project knowledge. The voice guide wins over your defaults.

For each piece:
- Start from the brief I give you. If there is no brief, ask for one.
- Draft, then list the facts and quotes that need checking before publishing.
- Never publish, schedule or send anything. Hand drafts back to me.

At the end of each session, give me an updated pipeline table:
Piece / Stage (Idea, Drafting, Review, Scheduled, Live) / Owner / Next step / Due

Common mistakes

Pasting the whole brief into the instructions. The brief changes every week, and the instructions should not. Put the brief in project knowledge and point to it.
Vague rules. “Be proactive” gives Claude nothing to check against. “Flag any request outside the signed agreement” does.
No source order. When an old brief and a new decision disagree, Claude has to guess which one counts. Name the file that wins.
No update rule. Without one, decisions stay buried in chats. The “Decision to log” line turns them into something you can copy.
Two decision logs. Replace the old file when you upload a new one, so Claude never reads two versions that disagree.

Test the instructions in a new chat

Start a new chat in the project.
Ask: “What is the goal of this project, what is the source of truth, and what do you need my approval for?” The answer should match your instructions.
Ask for a status update. It should arrive in your format.
Ask for something outside the agreed work. Claude should flag it before doing it.

If any answer is off, the instruction it ignored is probably vague or buried. Move it near the top and make it one testable sentence.

Where project instructions stop

Instructions and knowledge stay inside one Claude project. Anthropic says each project has its own memory and does not use memory from other projects. ChatGPT, Grok and your other assistants never see them. If the same project moves between assistants, keep the brief and decisions in one place every assistant can read. See how to share project context between ChatGPT and Claude. Flori is built for that: a shared project record that Claude and your other assistants read and update, with every saved change visible to you.

Frequently asked questions

Do project instructions apply to every chat in the project?

Yes. Anthropic says Claude uses project instructions for all the chats within the project.

Should I put my project brief in the instructions?

No. Put the brief in project knowledge, where you can replace it as the work changes. Keep the instructions for how Claude should work.

Can I reuse these instructions in ChatGPT?

Yes, with small edits. ChatGPT projects have their own instructions, which apply only within that project and override your global custom instructions. See how to make ChatGPT remember your project.

How many projects can I have on the free plan?

Five. Anthropic says free users can create a maximum of five projects.

Sources

Checked October 19, 2026. Vendors change these pages often, so check the source before you rely on a plan detail.

Keep exploring

Start your workspace