Claude Projects & Skills

Reference Guide for Enterprise Compliance Teams

Projects vs. Skills — When to Use Each

Projects

What: A persistent workspace with uploaded documents (knowledge) and custom instructions.

When: You have a set of documents and rules Claude should always reference for a body of work.

Example: Your "Compliance" project contains the AI Policy, risk matrix, and compliance manual. Every conversation in this project can reference them.

Sharing: Share with specific people (view or edit) or make visible to the whole organization.

Skills

What: A portable procedure anyone can run — like a template that tells Claude how to do a specific task.

When: You have a repeatable process you want others to follow consistently.

Example: The "Compliance Checker" skill — any employee runs it, gets a structured compliance report.

Sharing: Provision org-wide (all employees get it automatically) or share with specific people.

They work together: A skill running inside a project can reference that project's knowledge files. Your compliance skill + compliance project = a fully informed review tool.

Part 1 — Creating a Project

1 Open Claude Desktop and click the Projects tab in the left sidebar.

2 Click "New Project" and give it a name (e.g., "Compliance").

3 Add project instructions. Click the project settings gear icon, then "Instructions." These are custom instructions Claude follows in every conversation within this project.

Example instructions for a compliance project:

You are assisting the CCO of an SEC-registered RIA.

Rules:
- AI is a drafting tool only — never a decision-maker
- Cite specific policy sections and rule numbers
- Flag anything that needs human review
- Format output as structured reports

Reference the uploaded AI Policy, risk matrix, and
compliance manual for all reviews.

4 Upload knowledge files. In the project settings, click "Knowledge" and upload your documents:

AI Policy (PDF)
Risk matrix (Excel — Claude reads .xlsx)
Compliance manual sections
Email review lexicon
Any templates or reference docs

5 Test it. Start a new conversation in the project and ask: "Review our AI Policy against the latest SEC AI exam priorities. What gaps should we address?"

Part 2 — Creating a Skill

Skill structure

A skill is a folder (delivered as a ZIP file) containing:

compliance-checker/
  SKILL.md          ← required: metadata + instructions
  references/       ← optional: supporting documents
    ai-policy-rules.md
  scripts/          ← optional: Python/Bash/JS scripts

The SKILL.md file

This is the core of the skill. It has two parts:

---
name: compliance-checker
description: >-
  Review documents or AI usage against the firm's
  AI Policy and compliance requirements. Generates
  a structured report for CCO review.
---

# Compliance Checker

[Instructions Claude follows when this skill activates]
[When to use, what to check, how to format output]

The description field is critical. Claude reads it to decide when to automatically activate the skill. Write it as if you're telling Claude "use this when someone asks about..."

1 Create your skill folder with the SKILL.md file and any reference files.

2 ZIP the folder. On Windows: right-click the folder → "Compress to ZIP file."

3 Upload to Claude. Go to Customize → Skills → Add Skill and upload the ZIP.

4 Test it. In any conversation, say: "Check this document for compliance issues." Claude should automatically load the skill.

Part 3 — Sharing with Your Organization

Sharing a Project

Action	How
Share with a specific person	Project settings → Members → add by name or email → choose "Can view" or "Can edit"
Make visible to everyone	Project settings → Visibility → switch to "Public" (all org members can see it in the Team tab)

Provisioning a Skill Org-Wide

Requires Organization Owner role. Only Organization Owners (not regular admins or members) can provision skills for the entire organization. If you don't have this role, ask your Claude Enterprise administrator to provision it.

1 Go to Organization Settings → Skills.

2 Click Upload Skill and select your ZIP file.

3 Choose the provisioning scope: all users, or specific groups.

4 The skill is now automatically enabled for all selected users. They can toggle it on/off in their own Customize → Skills.

Part 4 — Maintaining and Updating Skills

Versioning strategy

Claude does not have built-in version history for skills. To maintain versions:

Keep a folder of dated ZIPs (e.g., compliance-checker-2026-09-25.zip, compliance-checker-2026-12-01.zip)
When updating: edit the SKILL.md, re-ZIP, and re-upload to Organization Settings
All users get the update automatically the next time they use the skill
To roll back: upload an older ZIP — it replaces the current version

When to update

New regulation or SEC guidance issued
AI Policy changes
Risk matrix categories change
You discover a gap in what the skill checks
After each compliance review cycle (quarterly recommended)

Tracking who ran the skill

Claude does not natively track which users ran a specific skill. Workarounds:

The skill instructions tell Claude to include the user's name and date in every report
Ask employees to save reports to a shared project folder ("Compliance Submissions")
Use the Enterprise Compliance API for broader activity logs

Part 5 — Scheduled Tasks (Automation)

Claude can run tasks automatically on a recurring schedule. This is ideal for daily compliance scans, weekly report generation, or any repeating workflow.

1 In Cowork, click "Scheduled" in the left sidebar.

2 Click "New task".

3 Choose how to set it up:

"Create with Claude" — describe what you want and how often; Claude configures the rest (recommended for first-time setup)
"Set up manually" — fill in name, instructions, frequency, approval mode, model, and working folder

4 Available frequencies: hourly, daily, weekly, weekdays, or manual (on-demand only).

5 Review and click "Schedule".

Key facts

Tasks run in the cloud — your computer can be off
Each run creates its own session with results you review
Tasks use your connected connectors (M365, Slack, etc.) and installed skills
Tasks cannot access local computer folders — cloud only

Requirement: Scheduled tasks need "Run Cowork in the cloud" to be enabled. If your organization has this turned off, talk to your admin about enabling it for your account or compliance group specifically — Enterprise plans support per-user and per-group controls so you don't have to enable it for everyone.

Email delivery

If your M365 connector has write tools enabled, you can configure a scheduled task to draft an email with the results. The draft appears in your Outlook Drafts folder for you to review and send — keeping human-in-the-loop for all outbound communications.

Part 6 — Project Knowledge Limits

Limit	Value
Max file size (project knowledge)	30 MB per file
Number of files	Unlimited (within context window)
Supported document types	PDF, DOCX, XLSX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON
Supported image types	JPEG, PNG, GIF, WebP
PDF page limit	1,000 pages
XLSX files	Require "Code execution" to be enabled in org settings

On paid plans, Claude automatically enables RAG (Retrieval Augmented Generation) when your project knowledge approaches the context window limit, expanding effective capacity by about 10x while maintaining quality.

Quick Reference — Key Differences

Feature	Project	Skill
Always loaded	Yes (in that project's chats)	No (loads on demand when triggered)
Carries documents	Yes (knowledge files, up to 30 MB each)	Yes (references folder in the ZIP)
Shareable	Yes (view/edit permissions)	Yes (read-only for recipients)
Org-wide distribution	Public visibility (anyone can find it)	Owner provisions via Org settings, or share with organization
Editable by others	If given "Can edit" permission	No — owner only; recipients get updates automatically
Auto-updates for recipients	Changes visible immediately	Recipients get the new version at next use
Works in Desktop app	Yes (Windows and Mac)	Yes (Windows and Mac)
Can use connectors	Yes (M365, etc. are account-level)	Yes (runs in context with active connectors)
Can automate on schedule	No (projects are passive knowledge)	Yes (via Cowork scheduled tasks)