Learn Agent Skills by Example with DeepLearning.AI's Course Files

Reading about skills is slower than seeing the same prompt run with and without them. The official artifacts repository for DeepLearning.AI's "Agent Skills with Anthropic" short course exists for exactly that: per its README, it contains summaries, notes, prompts, code files, and resources for the course. The repo sits at 1,380 stars on the SkillMap leaderboard.

Why This Skill Matters

The course materials page says the course shows skills in action across four surfaces: Claude.ai/Claude Desktop, the Claude API, Claude Code, and the Claude Agent SDK. The repo is organized by lesson — directories L1-partI through L7, with notes folders for the later lessons — so you can follow along episode by episode. And because skills are an open standard, the materials page notes you can try them in Claude.ai/Claude Desktop on the Free Plan, or with any skills-compatible application listed at agentskills.io.

Installation

This is a study repository, not an installable skill — the README documents no install command. Clone it:

git clone https://github.com/https-deeplearning-ai/sc-agent-skills-files.git
cd sc-agent-skills-files

Real Workflow: Compare Lesson 1 With and Without Skills

Lesson 1 runs a weekly marketing-campaign analysis, and the repo ships both halves of the comparison.

1. Open L1-partI/prompts_initial_conversation.md. This is the without-skills version: a long, hand-specified prompt. You define the funnel metrics yourself (CTR, CVR), paste historical benchmarks per channel, spell out the efficiency formulas (ROAS, CPA, net profit with shipping and product-cost assumptions), and describe the output table format.
2. Open L1-partI/prompts_with_skills.md. Same task, different shape: attach the weekly campaign data, state the constraint ("I can shift up to $10k between channels, and limit increases to 15% per channel"), then ask for an Excel report with an Executive Summary, Funnel Analysis, Efficiency Analysis, and Raw Data sheet, color-coded green and red. The expertise — funnel formulas, benchmarks, spreadsheet structure — moves from your prompt into the skill.
3. Run both against the bundled data: L1-partI/campaign_data_week1.csv is the week-one dataset, budget_reallocation_rules.md holds the reallocation best practices the conversation references — channel classification rules and a decision matrix — and output_report_example/ shows an example of the output.
4. Check the with-skills file's note: during filming, the Excel file the crew obtained carried values without formulas; the file shows the one-line prompt addition that asks for formulas and cross-sheet references instead.

The later lessons keep the same L<number> directory naming — lesson 4 (L4) carries a custom_skills/ folder and reading material, and the lesson 6 notes (L6_notes) include a clear.py script alongside the prompts.

Tips

• The course materials page notes each video has a "View Materials" link below it — use it to jump from a lesson moment to the matching file in the repo.
• Lesson 1's without-skills prompt is a useful template even outside the course: it shows how much specification a bare prompt needs when no skill carries the methodology.
• The skill/ directory inside L1-partI holds the lesson's skill definition — reading its structure next to the two prompts is the fastest way to see what a skill actually encapsulates.

When Not to Use This

The repo documents a specific course; it is not a skill library to install from, and the README does not present it as one. The course materials page points to agentskills.io for the open standard itself, not to a production skill catalog — if you want ready-to-install skills rather than teaching artifacts, browse the SkillMap leaderboard instead.

See the leaderboard for more skills.