Build Claude plugins and skills

Guide plugin creation from idea to deployed GitHub repository

Taking a plugin idea from raw concept through discovery, design, implementation, and GitHub push in a structured 23-step process.

• · Plugin vision and problem statement
• · Target audience and success criteria
• · Input/output specifications
• · Open source library research findings
• · Platform target (claude.ai|Claude Code|both)

• · Classified plugin type (skill, agent, command, or embedded knowledge)
• · All necessary plugin files (SKILL.md, scripts/, configuration)
• · GitHub repository push with proper structure
• · Copy-paste shell script (claude.ai) or automatic git push (Claude Code)

• · GitHub CLI (gh) for Claude Code environment
• · Bash and file system for code generation

• · User completes all 7 discovery questions
• · Open source research conducted on relevant libraries
• · Affinity clustering applied to tool selection

• · Jumping to structure without reading source repos: detectable and restarts discovery
• · Tier 3 tool (full app) treated as callable library: fails and retracts
• · Knowledge-only content (Tier 2) placed in separate file: refactored into prompt body

• · Reads all relevant open source repos before architecture decision
• · Three-tier tool classification (Callable Library, Extractable Knowledge, Pattern Only)
• · Detects platform (Claude Code vs claude.ai) and uses correct push method
• · Affinity clustering prevents tool misclassification