Package a Repeat Legal Task as a Claude Skill (Build, Test, Share)
Write your playbook down once as a Claude Skill, test it against past matters with known good outputs, and share it so every summary or review follows the same standard.
When to use this
Use this when you keep pasting the same long prompt for a task with a stable format, such as a deposition summary, an NDA triage against your positions, or a client status letter. A Skill holds the instructions, the template, and the review rules in one folder that Claude loads when the task calls for it. Build it once, test it on matters where you already know what good looks like, and share it with colleagues.
Most lawyers who use Claude regularly end up with a handful of prompts they paste over and over: the deposition-summary format, the NDA positions, the closing checklist. Each paste is a chance to grab an older version, skip the verification rules, or forget the template. A Claude Skill fixes that by putting the playbook, the output template, and the instructions in one folder with a required SKILL.md file. Claude reads the skill's description and loads it when a request matches.
This workflow walks through the whole cycle: choose a task that is stable enough to encode, have Claude draft the SKILL.md from your own materials, assemble and upload the folder under Customize, then Skills, test it against three past matters where you already know the right answer, and share it once it holds up (sharing with colleagues is a Team and Enterprise feature). The most important line is the description, which is limited to 200 characters and tells Claude when to use the skill. Write it as a trigger condition, not a label.
A skill does not change who is responsible for the work. Every output it produces is still a draft for attorney review, and the skill itself should say so in its instructions. Build in the habits you want enforced: quote the source for every statement, mark what could not be confirmed, and stop to ask when information is missing rather than filling the gap.
The Workflow
Pick one task and gather your materials
Choose a task you do often whose output format rarely changes. Good candidates: your deposition-summary format, your NDA playbook, a demand-letter checklist. Collect three things: the written playbook or standards, one example output you consider good (redacted or rebuilt with fictional facts), and the checklist you use when reviewing that work product. If your playbook lives only in your head, have Claude interview you to get it on paper first.
PromptI want to turn one of my repeat legal tasks into a Claude Skill. The task is [TASK, e.g., "summarizing deposition transcripts for our litigation files"]. My practice is [PRACTICE AREA] and I work on the [PLAINTIFF/DEFENSE/BUYER/SELLER/IN-HOUSE] side. Below are my current standards for this task, an example output I consider good (facts are fictional or redacted), and my review checklist. <document name="standards"> [PASTE YOUR PLAYBOOK, FORMAT RULES, OR STANDING INSTRUCTIONS] </document> <document name="example_output"> [PASTE ONE GOOD EXAMPLE OUTPUT] </document> <document name="review_checklist"> [PASTE THE CHECKLIST YOU USE TO REVIEW THIS WORK] </document> Before drafting anything, list (1) the rules you can identify in these materials, each with a short quote from where it appears, (2) places where the example output does something the standards do not explain, and (3) up to 8 questions about gaps or contradictions. Do not fill gaps with your own assumptions.
What you get: A numbered list of rules with source quotes, a short list of unexplained patterns in the example, and targeted questions. Answer the questions before moving on; the answers become part of the skill.
Have Claude draft the SKILL.md
Ask Claude to turn your materials and answers into a SKILL.md. The file opens with two required fields: a name (64 characters max) and a description (200 characters max) that says what the skill does and when to use it. The body holds the instructions, the output format, and your verification rules. Keep example outputs in supporting files, built on fictional facts, rather than inside the instructions.
PromptUsing my materials and my answers to your questions, draft a SKILL.md for this task. Requirements: - Start with a name (64 characters max) that includes a version date, e.g. "[TASK NAME] (Firm Standard, v[YYYY-MM])". - Then a description (200 characters max) written as a trigger condition: "Use when asked to [ACTIONS] for [MATTER TYPE]. Produces [OUTPUT]." Count the characters and show the count. - Then the instructions as a numbered list, in the order the work should happen. - Include these rules: quote or pin-cite the source document for every factual statement; if information is missing or a document appears incomplete, say so and ask rather than assume; close every output with an "Open Items for Attorney Review" list; label every output "DRAFT FOR ATTORNEY REVIEW". - Refer to the output template by filename (e.g., template.md) instead of repeating it inline. - Keep it to this one workflow. Do not add features I did not describe. After the draft, list any instruction you were unsure how to phrase and why.
What you get: A complete SKILL.md with name and description character counts shown, a numbered instruction list, and a short list of judgment calls for you to confirm.
Build the folder and upload it
Create one folder named for the skill. Put SKILL.md in it along with the supporting files: the output template, any fallback-position table, and a fictional example output. Ask Claude to draft the template file from your good example if you do not have one. Zip the folder so the skill folder is the root of the zip, then upload it in Claude under Customize, then Skills, and enable it.
PromptDraft the supporting files for this skill folder, one at a time, each in its own code block with the filename as a heading: 1. template.md: the output skeleton based on my example output, with [BRACKETED PLACEHOLDERS] where matter-specific content goes, a header block that includes "DRAFT FOR ATTORNEY REVIEW", and the Open Items section at the end. 2. [OPTIONAL: positions.md: a table of each clause or issue, our preferred position, our fallback, and the point where it must go to a partner. Use only positions stated in my materials; mark any row I need to complete as [TO CONFIRM].] 3. example-output.md: a short sample output built entirely on fictional parties and facts that follows template.md exactly. Then give me the final folder listing (folder name, then each file) so I can check the structure before zipping.
What you get: A template file, an optional positions table limited to what you supplied, a fictional example, and a folder listing to check against before you zip and upload.
Test it on three past matters
Run the skill on three past matters where you already have a good final output, using redacted or rebuilt fictional versions. First confirm the skill activates on its own from a normal request; if it does not, sharpen the description before changing anything else. Then compare each result against the known-good output.
PromptI am testing a draft Claude Skill. Below is the output the skill produced for a past matter and the final version our attorney approved for the same matter (both redacted). <document name="skill_output"> [PASTE THE SKILL'S OUTPUT] </document> <document name="approved_output"> [PASTE THE APPROVED FINAL VERSION] </document> Compare them and return a table with columns: Issue | Where it appears (quote from each version) | Type (missing content, wrong format, unsupported statement, tone, other) | Suggested change to the SKILL.md instructions. Then list anything in the skill output that is not supported by the source materials. Do not rewrite the output itself; I want changes to the instructions so the skill gets it right next time.
What you get: A drift table tied to specific instruction changes. Apply the changes, rerun on all three matters, and stop when the remaining differences are matters of judgment rather than format or accuracy.
Share it and record a version note
Once the skill is stable, share it. On Team and Enterprise plans you can share it with specific colleagues so it appears in their Shared with you section, and an organization owner can provision it for everyone so it appears in the Organization section. On other plans, colleagues upload the same zipped folder to their own accounts. Add a version note at the top of SKILL.md (date, what changed, who approved it) so you can tell which version produced a given draft.
PromptDraft a short version note to place at the top of this SKILL.md body, plus a one-paragraph message to colleagues introducing the skill. Version note format: Version [YYYY-MM-DD] | Approved by [NAME] | Changes: [BULLETS FROM OUR TESTING] The colleague message should say what the skill does, the kind of request that triggers it, what it will not do, that every output is a draft for attorney review with sources checked, and who to contact with problems. Keep it under 150 words and do not overstate what the skill does.
What you get: A version block and a short rollout message you can send with the share.
Maintain it when the playbook changes
Put a review date on your calendar and revisit the skill whenever your playbook, your templates, or [APPLICABLE RULE] changes. Update the version note each time. When a colleague reports a bad output, add the case to your test set before you change the instructions, so the fix is tested against the problem.
Example Output
Illustrative example — names, facts, and figures are fictional.
--- name: NDA Triage (Hollis & Vane Standard, v2026-10) description: Use when asked to review, triage, or redline an NDA or confidentiality agreement for a client. Produces an issues table against our positions and a cover note. --- Version 2026-10-01 | Approved by M. Hollis | Changes: added non-solicit rule; tightened residuals fallback Instructions 1. Read the whole agreement first. If pages, exhibits, or defined terms appear to be missing, list them and stop. 2. Identify whose side we are on from the request. If unclear, ask. 3. For each issue in positions.md, find the clause, quote it, and classify it: Acceptable / Push back / Escalate to partner. 4. Use the fallback in positions.md only. If a clause raises an issue not covered there, mark it "NOT IN PLAYBOOK" and do not propose language. 5. Produce the output in template.md: header, issues table, short cover note. 6. Close with Open Items for Attorney Review: missing exhibits, ambiguous definitions, anything you could not confirm. 7. Label every output DRAFT FOR ATTORNEY REVIEW. Folder: nda-triage/ SKILL.md template.md positions.md example-output.md (fictional: Brightwell Robotics / Norrland Freight) Test log (fictional matters) Matter A: activated from "triage this NDA"; matched approved output except residuals clause. Fixed in v2026-10. Matter B: activated; flagged missing Exhibit A correctly. Matter C: did not activate from "look over this confidentiality agreement". Description updated to include "confidentiality agreement".
Tips
- •Spend most of your effort on the description. It is the only part Claude reads when deciding whether to load the skill, so name the actions and documents that should trigger it.
- •Keep each skill to one workflow. An NDA skill and a deposition-summary skill should be two folders, not one skill with branches.
- •Build example outputs on fictional parties and facts. The skill folder goes to everyone it is shared with, so it should never contain real client material.
- •Put the verification rules in the skill itself (quote sources, list open items, label it a draft) so they apply even when a colleague forgets to ask.
- •Keep your three test matters and rerun them after every change. A fix for one problem can quietly break the format elsewhere.
- •If you rely on a long prompt in a Claude Project today, that prompt is a good first skill. See the Project workflow for what stays in Project instructions.
A note on confidentiality
A skill folder is shared material: never put real client names, facts, or documents in SKILL.md, templates, or example files. Use fictional or fully redacted examples, and run tests on redacted versions of past matters. When the skill runs on live client documents, use a Team or Enterprise plan whose commercial terms do not use your content for training, or confirm your consumer-plan privacy settings. Every output the skill produces is a draft for attorney review, with every citation and quoted source checked against the original.
Frequently Asked Questions
Which Claude plans support Skills?
Skills are available on Free, Pro, Max, Team, and Enterprise plans. Sharing a skill with specific colleagues, and provisioning it organization-wide through the Organization section, are Team and Enterprise features; check the Customize menu in your own account for what your plan shows.
What goes in the SKILL.md file?
Two required fields at the top, a name of up to 64 characters and a description of up to 200 characters that says what the skill does and when to use it, followed by the instructions in plain markdown. Templates, position tables, and example outputs go in separate files in the same folder.
My skill does not activate. What should I change?
Start with the description. Rewrite it as a trigger condition that names the actions and documents involved, such as "Use when asked to review, triage, or redline an NDA or confidentiality agreement." Test again with the phrasing your colleagues actually use before you change the instructions.
Does a skill make the output reliable enough to send without review?
No. A skill makes the format and the process consistent, but the content is still a draft. The responsible attorney reviews every output, checks quoted sources and citations against the originals, and resolves the open items the skill lists.
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.