Set Up a Reusable Claude Project for Your Practice
Build a reusable Claude Project loaded with your practice's voice, templates, and default rules so every matter starts pre-configured instead of from a blank chat, and package repeated tasks as Skills.
When to use this
Many lawyers re-explain who they are, their jurisdiction, and their drafting style at the start of every Claude chat, which wastes time and produces inconsistent output. A Claude Project lets you save those instructions plus anonymized templates and checklists once, so every new chat in that Project already knows your context. When a task keeps repeating, a Skill holds the procedure itself. Together they make the other workflows on this site more consistent.
If you use Claude for more than the occasional one-off question, you have probably noticed how much time you spend re-establishing context. Every fresh chat starts cold: you re-explain your practice area and jurisdiction, that you want a measured tone, that you never want invented facts. A Claude Project fixes that by saving your instructions and knowledge files in one place that every chat in the Project inherits.
Think of this as the meta-workflow. Building a well-configured Project once pays off on every matter afterward, because your voice, your standard positions, and your safety rules are already loaded. It also makes output more consistent, which matters when several chats, or several lawyers on a team or in a legal department, should sound like the same practice.
Projects and Skills do different jobs. Project instructions hold the context for one workspace: who you are, your jurisdiction, your house style. A Skill holds a repeatable procedure that produces a specific output (a deposition summary, a contract triage) and can carry its own templates; Claude loads it when it's relevant. This workflow sets up the Project first and ends with turning your most-repeated task into a Skill. The prompts below are text you paste into the configuration, not questions you ask in a chat. You stay responsible for everything Claude produces, and your duty of confidentiality applies to what you load in.
The Workflow
Decide what to load
Before you open Claude, make a short list of the context you find yourself repeating. The goal is to capture the reusable parts of your practice, not any single client's matter. Good candidates: anonymized writing samples that show your voice, your standard contract positions and fallbacks, a short style guide, the practice areas you handle, and your jurisdiction. Keep this list to things that are true across matters.
PromptMy reusable Project should include: - Who I am: [ROLE] at [FIRM / LEGAL DEPARTMENT], practicing [PRACTICE AREAS] - Jurisdiction(s): [STATE / COURTS] - My writing voice: [e.g., plain, measured, client-friendly; avoids legalese] - Standard contract positions: [e.g., mutual indemnity preferred; cap at fees paid] - Documents to attach as knowledge: [anonymized engagement letter], [demand letter template], [intake checklist], [clause bank] - Default rules I always want followed: [list here] - Tasks I repeat most often (candidates for a Skill later): [e.g., deposition summaries, NDA triage]
What you get: A clear, short inventory of reusable practice context, a list of the anonymized files you will attach, and a note of your most-repeated tasks. Nothing client-specific should be on this list.
Create the Project and write custom instructions
In Claude, create a new Project and give it a clear name like "Drafting: [Practice Group]". Then write the Project instructions: who you are, your jurisdiction, the tone you want, and your default rules. The safety rules matter most. Tell Claude to flag every citation as UNVERIFIED until you confirm it, never to invent facts, to quote the source when it relies on a document, and to ask before assuming. These instructions apply to every chat you start in the Project.
PromptYou are assisting [ROLE], a [PRACTICE AREA] attorney at [FIRM / LEGAL DEPARTMENT] licensed in [JURISDICTION]. Apply these rules to every conversation in this Project: Voice: Write in [plain, measured, client-friendly] English. Avoid legalese unless I ask for it. Safety rules: - Flag EVERY case citation, statute, or rule reference as "UNVERIFIED" until I confirm it. Never present a citation as verified. Use [APPLICABLE RULE] when you are not certain of a rule number. - Never invent facts, dates, dollar amounts, or party names. If you need one, leave a [BRACKET] for me to fill. - When you rely on a document I've provided, quote the passage and give its page or section. - If a request is ambiguous, ask me before assuming. Tell me when you are uncertain. - You are a drafting aid, not the lawyer. I review and am responsible for all output. Defaults: [e.g., American spelling; cite to [JURISDICTION] authority; no deadline assumptions].
What you get: A saved Project whose instructions encode your identity, jurisdiction, tone, and safety defaults so you never retype them. The unverified-citation rule should show up in Claude's behavior on the next task.
Add knowledge files
Upload the anonymized documents from your inventory as Project knowledge: a clean engagement letter, a demand or motion template, an intake checklist, your clause bank or standard positions. These persist across every chat in the Project, so Claude can match your structure and language. Strip client names, matter numbers, and any privileged specifics before uploading; a knowledge file is reusable practice material, not a place for live matter detail.
PromptKnowledge file outline: "Standard Positions (Anonymized)" 1. Engagement terms: scope language, fee structure options, termination clause 2. Contract fallbacks: indemnity [preferred / acceptable / walk-away]; liability cap [preferred / acceptable]; governing law [JURISDICTION] 3. Demand letter skeleton: facts, liability, damages, demand, deadline 4. Intake checklist: conflicts, jurisdiction, limitations period [attorney to confirm], key documents 5. House style notes: defined-term conventions, heading format, signature block (Replace all party names with [CLIENT], [OPPOSING PARTY]; remove matter numbers.)
What you get: A handful of anonymized reference files attached to the Project. Future drafts will follow your templates and standard positions without you pasting them each time.
Test it on a real task and refine
Open a new chat inside the Project and run a real task, such as a first draft of a demand letter or a clause comparison, using non-confidential or anonymized facts. Watch two things: does it match your voice and templates, and does it honor your safety rules (citations flagged UNVERIFIED, sources quoted, no invented facts, questions when unsure)? Where it misses, tighten an instruction, then re-run to confirm the fix.
Prompt<document type="[e.g., incoming redline / intake notes]" note="anonymized"> [PASTE AN ANONYMIZED TEST DOCUMENT, IF THE TASK NEEDS ONE] </document> Draft a [demand letter / engagement letter / clause comparison] for [SHORT, NON-CONFIDENTIAL SCENARIO; say whose side I'm on and what the goal is]. Use my standard structure and voice. Quote the document where you rely on it, flag any citation as UNVERIFIED, leave [BRACKETS] for facts you don't have, and list anything you were unsure about at the end.
What you get: A draft that sounds like your practice and follows your templates. Any gaps reveal an instruction to sharpen, e.g., "never use the word 'utilize'" or "always include a response deadline."
Maintain it, and turn repeated tasks into Skills
A Project is only as good as it is current. When you make a drafting decision you'd want repeated (a new standard clause, a tone tweak, a position change), fold it back into the instructions or a knowledge file, and revisit the Project periodically to remove stale templates. When a task keeps coming back with the same long prompt, package it as a Skill: a folder with instructions, a description that tells Claude when to use it, and any templates. Skills uploaded to your claude.ai account are also available in Cowork, and on Team and Enterprise plans they can be shared with colleagues. Keep shared Projects and Skills to anonymized practice material, and route confidential matter work to access-controlled Projects.
PromptHelp me turn my most-repeated task into a Claude Skill. The task is: [e.g., summarizing deposition transcripts for the litigation file]. Using the instructions and templates in this Project, draft: 1. A one-sentence description that tells Claude when to use the Skill (e.g., "Use when summarizing a deposition transcript for a litigation file"). 2. Step-by-step instructions for the procedure, including where the input document goes (<document> tags) and the exact output format. 3. The safety rules that must carry over (UNVERIFIED citations, no invented facts, quote sources, flag gaps). 4. A list of any template files the Skill should include. Flag anything in my current Project instructions that belongs in the Skill instead, or vice versa.
What you get: A living Project that improves with use, plus a draft Skill for your most-repeated task that you can review, upload, and (on a Team or Enterprise plan) share so colleagues run the same procedure.
Example Output
Illustrative example — names, facts, and figures are fictional.
PROJECT: Drafting: Employment Practice Group, [Firm A] (fictional) WHO YOU ARE ASSISTING You support attorneys in [Firm A]'s employment practice group, licensed in [STATE]. We represent employees and employers in wage, discrimination, and severance matters. VOICE - Write in plain, measured English a client can read without a dictionary. - Short paragraphs. No throat-clearing. Avoid "utilize," "herein," "aforementioned." - Confident but never overstated; we do not guarantee outcomes. SAFETY RULES (ALWAYS) - Flag EVERY case, statute, or rule citation as UNVERIFIED until a lawyer confirms it. - Never invent facts, dates, dollar figures, or names. Leave [BRACKETS] instead. - Quote the source document, with page or section, when you rely on it. - Ask before assuming when a request is ambiguous. - You are a drafting aid. A licensed attorney reviews and is responsible for all output. JURISDICTION & DEFAULTS - Default authority: [STATE] statutes and [STATE] / [FEDERAL CIRCUIT] courts. - American spelling. No deadline assumptions unless I provide them. HOUSE STYLE - Demand letters: facts, liability, damages, demand, response deadline. - Define a party once in bold, then use the short form. - Always include a signature block with bar number placeholder [BAR NO.]. WHAT NOT TO DO - Do not include client-identifying details from prior matters. - Do not offer a legal conclusion as settled; frame options and risks. SKILL CANDIDATES: deposition summaries; severance-agreement review.
Tips
- •Lead your instructions with the safety rules, not your bio. The unverified-citation and no-invented-facts rules are the highest-value lines in the whole Project.
- •Name Projects so you can tell them apart at a glance, e.g. "Drafting (shared)" versus "[Matter name] (matter)." The name is your first reminder of what may safely go inside.
- •Keep knowledge files small and clean. A tight, anonymized clause bank works better than a large dump of old documents.
- •If a prompt keeps getting long and keeps getting reused, it's a Skill waiting to be written. Put the context in the Project and the procedure in the Skill.
- •Treat the Project as version one. The value comes from the small refinements you make after each task, not from getting it perfect on setup day.
A note on confidentiality
What goes into this Project is your practice's own material: templates, clause banks, standard positions, and writing samples that started life as client documents. Before uploading, strip client names, matter numbers, deal figures, and any privileged facts from every sample, because Project knowledge persists across every chat and, on Team or Enterprise plans, Projects and Skills can be shared with colleagues. Keep confidential matter work in separate, access-controlled Projects. Run client work on a commercial plan (Team or Enterprise) whose terms exclude your content from model training. On Pro or Max, disable model training in Privacy settings; a commercial plan's contract terms are the stronger answer for client work. Your duties under Model Rule 1.6 and ABA Formal Opinion 512 apply to everything you upload. Every output from the Project is a draft for attorney review: verify every citation and fact before use.
Frequently Asked Questions
What is the difference between a Claude Project and a normal chat?
A normal chat starts with no memory of your practice. A Project wraps your chats in saved instructions and knowledge files, so every chat you start inside it already knows your role, jurisdiction, voice, and rules, and you stop re-explaining the basics every time.
Should I use a Project or a Skill?
Both, for different things. Project instructions hold context for one workspace: who you are, your jurisdiction, house style. A Skill holds a repeatable procedure that produces a specific output, can carry templates, and loads when Claude judges it relevant. If a prompt keeps getting long and keeps getting reused, turn it into a Skill. The site's Skills post covers how to build one.
Is it safe to put my templates and standard positions into a Project?
Yes, as long as they are anonymized practice materials: clean templates, checklists, and clause banks with client names and matter numbers removed. The line to watch is privileged client specifics. Keep those out of any shared template Project and use a separate, access-controlled Project for confidential matter work.
Do I still have to check Claude's citations and facts if my Project tells it to flag them?
Yes. The unverified-citation rule is a safety net that reminds you what to check; it does not do the checking. You remain responsible for verifying every citation, fact, and legal conclusion before anything leaves your office.
Prompts used in this workflow
Get New Workflows Like This Every Week
Join the free Claude for Lawyers newsletter — practical workflows, prompts, and guides for everyday practice.