[Introduction to Claude Code E26] Creating Claude Code practical skills: PR review. test creation. document creation SKILL.md example

Let's apply the SKILL.md pattern learned in class 25 to three actual skills. Starting from an empty directory, we build PR review, test creation, and document creation skills step by step, and look at the differences from built-ins, description optimization, Gotchas, and how the validation loop goes in. 📌 Contents covered - Check built-in: /review, /security-review, /simplify, /init, /batch — Use built-in if the task is the same, custom only when our project convention is required - PR review skill (review-pr): context: fork + agent: Explore + Isolate large PR with dynamic context (!gh pr diff) Review in subagent, only result table in main - Test generation skill (generate-tests): $ARGUMENTS + argument-hint, Given-When-Then convention, Validation loop (repeated 5 times until passing), Vitest + __tests__ location convention - Document generation skill (generate-docs): Check user before bulk overwriting with Plan-validate-execute, save tokens by separating templates with assets/api-template.md, search for exported symbols with Glob/Grep - Apply pattern at 25: description trigger phrase, Gotchas are unique to this project Non-obvious facts (soft delete, different names for the same ID, /health vs /ready), output format templates, Procedures-not-declarations, Defaults-not-menus - Three skills in one workflow: tasks → /generate-tests → /generate-docs → /commit → /pr → /review-pr (one coherent unit principle) - Debugging tips: When automatic calls are not made, when called too often, only 5,000 tokens with auto-compaction When there are left, ! When a command is not executed 📌 Environment information - Latest version of Claude Code (supports Skills, slash commands, allowed-tools) - .claude/skills/ Project scope (team sharing) - gh CLI (for PR review skills), Vitest (for test creation skills) 📌 Chapter 0:00 Apply the patterns learned last time to the actual three skills 0:34 Look at the built-ins and start customizing 1:43 PR review: context fork + agent Explore + dynamic context injection 3:57 PR review: Gotchas section and output table template 5:47 Create tests: $ARGUMENTS and Validation loop 7:51 Create tests: Given-When-Then and Vitest conventions 9:47 Create documents: separate Plan-validate-execute and assets/ templates 13:40 Three skills in one workflow 14:32 Debugging tips 📌 Reference links - Blog: https://neosarchizo.github.io/posts/claude-code-26-practical-skills - Extend Claude with skills: https://code.claude.com/docs/en/skills - Commands reference: https://code.claude.com/docs/en/commands - Subagents: https://code.claude.com/docs/en/sub-agents - Agent Skills Specification: https://agentskills.io/specification - Anthropic skills repository: https://github.com/anthropics/skills - webapp-testing skill: https://github.com/anthropics/skills/tree/main/skills/webapp-testing - doc-coauthoring skill: https://github.com/anthropics/skills/tree/main/skills/doc-coauthoring - skill-creator skill: https://github.com/anthropics/skills/tree/main/skills/skill-creator #ClaudeCode #PracticalSkill #SKILLmd