universal-creator

Shared resources for universal-creator generator skills: prompting technique registry (techniques.json), the technique-selector subagent, the 18 worked prompt-technique examples (examples/), and the reusable agent trio (agents/) — validation-reviewer, artifact-router, prompt-strategist — plus the shared memory-guardrails snippet. Installed automatically when any dependent generator skill is installed. DO NOT USE directly — consumed by agent-generator, skill-generator, hook-generator, and prompt-generator.

Designs and generates Claude agent definitions: custom sub-agents (.agent.md), bounded-role system prompts, tool-use policies, handoff criteria, and delegation boundaries. Use when the user wants to create a specialized Claude agent, define a multi-stage agentic workflow, set tool restrictions per agent role, design a sub-agent that returns a single result and terminates, scope an agent to a specific domain or task type, or prevent an agent from overstepping its role. Generates .agent.md files, system prompt stubs, tool-allow/deny matrices, and role boundaries. DO NOT USE for Claude Code lifecycle hooks (use hook-generator); for writing prompts for one-shot tasks (use prompt-generator); for writing always-on workspace instructions (use instruction-generator).

Designs and generates Claude Code hook configurations for all 30+ lifecycle events including PreToolUse, PostToolUse, PermissionRequest, UserPromptSubmit, Stop, SessionStart, SubagentStop, PreCompact, PostCompact, FileChanged, ConfigChange, TeammateIdle, TaskCreated, TaskCompleted, and more. Use when the user asks to automate Claude Code behavior, enforce rules deterministically, block or approve tool calls, inject context after compaction, format code after edits, notify on idle, audit configuration changes, gate sub-agent execution, or add any lifecycle automation. Generates ready-to-paste JSON hook blocks, shell command scripts, prompt-based hook stubs, and agent-based hook stubs. DO NOT USE for general Claude API prompt engineering (use prompt-generator skill); for creating Claude agents or sub-agents (use agent-generator skill); for writing VS Code instructions (use instruction-generator skill).

Creates and refines workspace and file-level instruction files: copilot-instructions.md, AGENTS.md, and *.instructions.md. Use when the user wants to save coding preferences, enforce team conventions, apply rules to specific file patterns, configure always-on agent behavior, document style guides for Claude to follow automatically, or scope instructions to a particular language, framework, or workflow area. Generates ready-to-use instruction markdown with correct applyTo patterns, concise rule sets, and optional extended variants for fragile workflows. DO NOT USE for one-shot prompts (use prompt-generator); for custom agent role definitions (use agent-generator); for Claude Code lifecycle hooks (use hook-generator).

Designs and generates effective prompts for Claude and other LLMs using all 18 prompting techniques: zero-shot, few-shot, chain-of-thought, prompt chaining, ReAct, meta-prompting, self-consistency, generate-knowledge, tree-of-thoughts, RAG, ART, APE, active-prompt, DSP, PAL, reflexion, multimodal-CoT, and graph-prompting (full catalog: docs/techniques.md; worked examples live in the shared dependency at skills/shared/examples/). Use when the user wants to write a Claude system prompt, create a .prompt.md file, improve an existing prompt, choose between prompting techniques, structure multi-step reasoning, or apply context-budget-aware prompt engineering. Generates ready-to-use prompt files with technique justification and eval scenarios. DO NOT USE for Claude Code lifecycle hooks (use hook-generator); for creating agent role definitions (use agent-generator); for workspace instructions (use instruction-generator).

Designs and generates new skill directories for the universal-creator framework: SKILL.md definitions, workflow steps, domain conventions, generate_X_stub.py scaffold scripts, validate_X_output.py validators, agent definitions, templates, and evals. Use when the user wants to add a new skill to universal-creator, package a new artifact type as a reusable generator, create a skill from scratch following framework conventions, define a new skill for prompts/configs/documents or any domain-specific output, or extend the skill library with a new generator. Also use when someone says "make me a skill that generates X" or "I want a skill-creator style skill for Y" or "scaffold a new skill". DO NOT USE for creating Claude Code lifecycle hooks (use hook-generator); for writing workspace instruction files (use instruction-generator); for writing one-shot task prompts (use prompt-generator); for designing sub-agents with tool policies (use agent-generator).