What You Will Learn

Claude has several features that can look similar when you first encounter them.
The easiest way to understand the difference is to think about the role each one
plays in your workflow.

Project = context   |   Skill = method   |   Artifact = output

Part 1

Create Your First Claude Project

A Project gives you a dedicated workspace for a specific body of work.
For an evaluator, this could be a programme evaluation, country portfolio,
research project, donor-funded intervention, or monitoring assignment.

Step 1 — Open Claude

Sign in to Claude and go to the main Claude workspace.

Step 2 — Open Projects

From the left side of the Claude interface, select Projects.

Step 3 — Create a Project

Click + New Project.

For this tutorial, use:

Education Programme Evaluation 2026

You can use any name that reflects your own evaluation work.

Step 4 — Add Your Project Knowledge

On the Project page, locate the Project knowledge area.
Add the documents Claude should use as background for your evaluation.

For example:

• Theory of Change
• Logframe
• Evaluation Terms of Reference
• M&E Plan
• Indicator definitions
• Survey instruments
• Programme reports

Step 5 — Add Project Instructions

Project instructions tell Claude how you want it to work within this evaluation.

For example:

You are supporting an evaluation team working on an education programme.
Use the documents in Project knowledge as the primary context.
When analysing evidence:
- distinguish evidence from interpretation;
- identify assumptions and limitations;
- do not invent data;
- flag missing information;
- use M&E terminology accurately;
- ask for clarification when important information is missing.
Present recommendations in practical language suitable for programme managers and evaluation stakeholders.

Try It

Ask Claude:

Part 2

Create Your First Artifact

An Artifact is substantial, standalone content Claude creates in a dedicated
workspace. Depending on the task, an Artifact can be a document, piece of code,
HTML page, diagram, visualization, interactive component, or another substantial
output.

Step 1 — Create an Evaluation Matrix

Ask Claude:

Create an evaluation matrix for the Education Programme Evaluation 2026.
Include:
1. Evaluation question
2. Indicator
3. Data source
4. Data collection method
5. Analysis method
6. Key assumptions
7. Limitations
8. Responsible person
Use the Project knowledge as the basis.
Flag any areas where the available information is insufficient.

Step 2 — Review the Artifact

Do not treat the first output as the final evaluation product.
Review it as you would review work produced by a junior analyst.

• Are the evaluation questions correct?
• Are the indicators measurable?
• Are the proposed data sources realistic?
• Are assumptions clearly separated from evidence?
• Has Claude introduced information that was not supported by the source material?
• Are there methodological weaknesses that need human judgement?

Part 3

Understand Claude Skills

A Skill teaches Claude a repeatable procedure or specialized workflow.
Skills can contain instructions, scripts, examples, and other resources.
Claude loads relevant Skills when they are needed.

For M&E professionals, useful Skills could include:

• Indicator quality assurance
• Survey data-quality checks
• Qualitative coding procedures
• Evaluation report quality checks
• Donor reporting workflows
• Recommendation quality assurance

Option A — Build a Skill Through Chat

This approach works well when the important knowledge is mainly in your head:
methodological rules, decision logic, exceptions, quality standards, and
professional judgement.

You can work through a task with Claude, correct the results, refine the process,
and then turn the repeatable workflow into a Skill.

If you create a custom Skill manually, Claude’s current Skill structure uses a
skill.md file with required metadata such as a name and description.
More complex Skills can also contain additional resources and executable code.

---
name: M&E Indicator QA
description: Review M&E indicators for clarity, measurability, data source, frequency and common quality problems.
---
# Purpose
Review indicators systematically and identify weaknesses.
# Checks
- Clarity
- Measurability
- Unit of measurement
- Data source
- Frequency
- Responsibility
- Disaggregation
- Feasibility

Custom Skills can then be packaged as a ZIP file and uploaded through
Customize → Skills → + → Create skill → Upload a skill.

Option B — Record a Skill

Claude also provides a way to create a Skill by recording yourself performing
a task. This is particularly useful when a workflow is easier to demonstrate
than to explain in writing.

When is recording useful for M&E?

Imagine a monthly monitoring workflow where you:

1. Open a monitoring spreadsheet.
2. Review specific worksheets.
3. Check for missing indicator values.
4. Look for unusual values.
5. Review data-quality flags.
6. Prepare a QA summary.

If you would normally tell a colleague, “Let me show you how I do this,”
recording may be a good approach.

How to Record a Skill

In Claude for Mac, open Cowork.

You can start recording from the + button in the composer
and select Record a skill. You can also go to
Customize → Skills → Add → Record your screen.

Before recording, close anything you do not want captured. Do not display
passwords, API keys, confidential documents, private conversations, or other
sensitive information.

While recording, perform the task as you normally would. If there are decisions
that Claude cannot understand simply by watching the screen, explain them aloud.

A recording can run for approximately 10 minutes. When you finish, Claude reviews
the recording and proposes a Skill. Review the proposed Skill before saving it.

Improve the Skill: Run, Learn, Update

The first version of a Skill will not necessarily handle every situation.
The most useful Skills are improved through testing.

1. Run the Skill.
   Test it with a real but appropriate example.
2. Identify what went wrong.
   Look for mistakes, exceptions, missing rules, and corrections you had to provide.
3. Review the lessons.
   Ask Claude to identify what it learned from the run and what feedback you provided.
4. Update the Skill.
   Add only the lessons that you have reviewed and confirmed.
5. Run it again.
   Test the updated version with another example.

For example, after testing an Indicator QA Skill, you might ask:

List the important lessons from this run.
Also list the feedback and corrections I gave you during the task.
Do not update the Skill yet. I want to review the lessons first.

This last sentence is important. Review the proposed lessons before allowing
them to become permanent Skill instructions. A single example does not always
justify a new rule.

Once you have reviewed the lessons, you can ask Claude to incorporate the valid
changes into the Skill.

When Should You Record a Skill?

A useful rule of thumb is simple: if you would say,
“Let me show you,” recording may be appropriate.
If you would say, “Let me explain the rules,”
building the Skill through conversation is usually better.

Part 4

Put Everything Together

The three main concepts work together rather than competing with each other.

Project gives Claude the context. Skill gives Claude the method. Artifact gives you the output.

Your 15-Minute M&E Practice Exercise

1. Minutes 1–4:
   Create a Project called “Programme Evaluation Practice”.
2. Minutes 5–7:
   Add a sample Theory of Change, logframe, evaluation plan, or other appropriate material.
3. Minutes 8–10:
   Ask Claude to create an evaluation matrix as an Artifact.
4. Minutes 11–13:
   Ask Claude to identify three quality problems in the matrix.
5. Minutes 14–15:
   Identify one repetitive M&E workflow that could eventually become a Skill.

If Record a Skill is available on your Claude account, choose a small,
repetitive workflow and use non-sensitive sample data for your first recording.

Quick Knowledge Check

Keep the Human in the Loop

Claude Skills can make M&E work more consistent and efficient, but they
do not replace professional judgement.

• Review important evaluation outputs.
• Verify claims against the original evidence.
• Check calculations and indicators.
• Review assumptions and methodological decisions.
• Check whether recommendations are actually supported by evidence.
• Do not expose confidential information unnecessarily.
• Use appropriate human review for high-stakes decisions.