Agent Skills in a Day : Build Reusable AI Capabilities

Understand the skills architecture and how the cloud loads only the requested skill and its needed markdown and references into the context, while the extract script runs locally.

Load the architecture step by step, matching the pdf processing skill to user requests, trigger the skill, and apply progressive disclosure via front matter and lightweight metadata.

Leverage progressive disclosure to engineer skills with startup metadata of 100 tokens and roughly 5k tokens for skill markdown bodies, using guide references, domain mapping, and conditional scripts patterns.

Learn progressive disclosure for structuring large skills, compare guide-with-references versus domain-based organization, and follow golden rules like under 500 lines, one-level markdown, and descriptive file naming for efficient cloud-based access.

Master skill anatomy with YAML front matter rules and naming constraints. Author skills as markdown files, then install and enable them; use skills via APIs to create PowerPoint and Excel.

Explore skill anatomy by mastering front matter in a yaml block, including name and description rules, third-person triggers, and precise activation phrases for reliable skill deployment.

Learn to set the right degree of freedom for cloud skills, balancing high, medium, and low freedom with examples like code reviews, templates, and exact commands to prevent risky deviation.

Explore three core skill body patterns: template, example, and workflow, showing how to structure skills markdown, provide fixed headers, and use multi-step outputs for clear progress.

Explore the three loading levels for skills: startup YAML front matter loads ~100 tokens per skill; triggered loading brings in domain expertise markdown; on-demand loads fetch referenced files or scripts.

Plan the skill before coding by defining its domain, scope, and what it should and should not do, then select a pattern and degree of freedom for Cloud.

Define and implement a reusable code review skill by creating a skills directory and skills.md, outlining structured reviews with issues, severity, location, and actionable suggestions.

Master the plan-validate-execute design pattern, a production orchestration approach for verifiable intermediate outputs. Plan identifies inputs and scripts; validate checks structure and columns; execute writes a deterministic output after validation.

Explore how skills differ across cloud AI, cloud code, and cloud API, with distinct scopes, network needs, and pre-installed packaging for personal, project, and enterprise deployments.

Build a production-quality, multi-file analytics skill with a Python CSV validator, organized folders, and reference files for finance and sales, using a plan-validate-execute workflow.

Define a CSV data analysis skill workflow by creating a markdown with front matter, domain mapping, validation, and structured outputs using pandas to summarize, visualize, and report.

Validate the skill end-to-end by analyzing a bad CSV, adding missing columns, validating, then analyzing a valid finance CSV to generate insights, summary, and data quality notes.

Create an Excel summary from finance data by running a Python script after analysis, generating an executive summary, charts, and data sections for reuse across the team.

Run end-to-end tests with Playwright in a TypeScript setup, validate a dev server on port 3000, and generate a JSON test report.

Run an end-to-end playwright test suite with skill, introduce a negative condition by changing the port, and review the JSON report detailing the pass, fail, file, line, and error for diagnosis.