Claude Code Promts Skills

Claude Code Prompts

Production-ready prompt library for Claude AI and coding agents, built on the APEI cycle: Analyze → Plan → Execute → Iterate.

Repository default is English-only and remains enforced to keep quality, tone, and contribution standards consistent across all prompts.

---

Start Here

0. In Claude Code, run /find-prompt <your task> — it routes any task to the exact prompt(s) to load. Or open REPOSITORY-MAP.md for manual navigation.
1. Use Agent System as the core prompt.
2. Add specialist prompts by tier — 1 for a single-domain task (the default), 2 only when the task spans two genuinely independent domains, Multi-Agent Orchestration when units of work need isolation. See Composition Tiers.
3. Validate outputs against explicit success criteria.

Six skills ship in .claude/skills/ — they load automatically when you run Claude Code from this repo:

Skill	Use when	File
/find-prompt	Routing any task to the right prompt(s) — the main entry point	View
deterministic-checks	Before committing / opening a PR — zero-model-cost scan for conflict markers, live-looking credentials, debug leftovers, untracked TODOs, oversized files	View
changelog-from-commits	Cutting a release — Conventional Commits parsed into a categorized changelog	View
doc-link-audit	Before publishing docs — internal links, heading anchors, and orphan pages, GitHub-accurate slug matching	View
skill-audit	Reviewing a skills directory — frontmatter validity, description quality (SDO), body size, cross-scope name collisions	View
capability-audit	Evaluating a GitHub-hosted skill, plugin, MCP server, or agent toolkit before installation — read-only repository evidence without cloning	View

These are operational skills rather than catalog-only examples: five include executable scripts, while /find-prompt is the routing layer. See Repository Structure below. This repo is also an installable plugin (.claude-plugin/plugin.json):

claude --plugin-dir /path/to/Claude-Code-Promts     # try it for one session
claude plugin validate /path/to/Claude-Code-Promts/.claude-plugin/plugin.json
claude plugin validate /path/to/Claude-Code-Promts/.claude-plugin/marketplace.json

The repository also ships a marketplace manifest, so it can be installed without remembering a local --plugin-dir path:

/plugin marketplace add Rtur2003/Claude-Code-Promts-Skills
/plugin install claude-code-prompts@claude-code-prompts

Marketplace installation is intentionally left as a user action because it changes Claude Code's persistent plugin state.

Installing as a plugin additionally wires two deterministic safety hooks (block destructive commands, block writing live-looking credentials) — see Claude Code Plugins.

Top 5 Outcome Scenarios

Scenario	Prompt Setup
General autonomous coding	Agent System
Debugging production issues	Agent System + Debugging & Troubleshooting
Security-sensitive changes	Agent System + Security Audit
Architecture decisions	Agent System + Architecture Patterns
Complex multi-agent execution	Agent System + Multi-Agent Orchestration

---

One Decision Tree + Token Budget

Use the single-source selector here: Prompt Selector Guide.

Rule: start minimal (Tier 1 — one specialist), move up a tier only if the task's checklist visibly can't be covered by what's loaded.

---

Modern Claude Code Stack

Coverage of the current Claude Code / Claude ecosystem (models, Skills, Plugins, subagents, MCP, hooks, dynamic workflows, the Agent SDK). Read REPOSITORY-MAP.md to jump straight to what you need.

Topic	Prompt / Guide
Pick a model and effort level	Model Selection Guide
What a current Claude Code build can do	Native Features Guide
Write / debug Agent Skills	Agent Skills
Connect an external system via MCP	MCP Integration
Bundle and distribute a plugin	Claude Code Plugins
Native subagents + dynamic workflows	Multi-Agent Orchestration
Automate on lifecycle events	Hooks & Automation
Project .claude/ config and settings	Claude Code Workflow
Thinking depth, effort, plan mode	Thinking & Planning
Build an agent programmatically	Agent SDK Guide
The sources behind this library + cross-tool AGENTS.md + adjacent agents	Reference Resources
Evaluate skills, plugins, MCP servers, and agent toolkits	Capability Ecosystem Discovery
Research consequential questions beyond coding	Evidence-Driven Research

---

Active Agent Portfolio (Outcome-First)

claude-agent-system-prompt.md is the operational source of truth. Full table with token counts and the task router: prompts/english/agents/INDEX.md.

Prompt	Use when	Do not use when	File
Agent System ⭐	Any autonomous task	You only need a tiny cheat sheet	View
Quick Reference	Token budget is extremely tight	You need deep specialist logic	View
Agent Skills ⭐	Writing or debugging a Claude skill	No skill authoring involved	View
MCP Integration ⭐	Connecting Claude to a DB/browser/API	No external system involved	View
Claude Code Plugins	Packaging skills/agents/hooks to share	Nothing to distribute	View
Multi-Agent Orchestration ⭐	Parallel agents, workflows, audits, review	Single-agent linear task	View
Hooks & Automation	Enforcing something on a lifecycle event	Advisory guidance is enough	View
Code Review	Reviewing a PR/change set	Writing new feature code from scratch	View
Security Audit	Threat/risk exposure is possible	Task has no security relevance	View
Refactoring	Improving maintainability safely	Incident response under active outage	View
Testing	Building or fixing test strategy	You only need non-test docs	View
Documentation	Producing technical docs	You need runtime diagnosis	View
Performance	Latency/throughput/cost bottlenecks	Problem is primarily correctness	View
Git & VCS	Workflow, branching, commit hygiene	App logic decisions	View
Accessibility Audit	WCAG and accessibility compliance	Backend-only infrastructure change	View
Migration & Upgrade	Framework/runtime/DB migrations	Greenfield implementation	View
Monitoring & Observability	Logs, metrics, traces, alerts	Pure UI copy/content task	View
Debugging & Troubleshooting	Root-cause and incident diagnostics	You are designing architecture from zero	View
Claude Code Thinking & Planning ⭐	Effort levels and planning depth	Non-Claude environments only	View
Claude Code Workflow	CLAUDE.md, rules, settings, permissions	You only need algorithm design	View
Technology Stack ⭐	Selecting tools/libraries	Stack is fixed and approved	View
Architecture Patterns	System design and trade-offs	Small local bug fix	View
Full-Stack Development	End-to-end app delivery	Single-layer scoped work	View
AI & LLM Integration	RAG/agents/model integration	No AI component exists	View
API Design & GraphQL	API contract/schema design	UI-only styling task	View
Cloud & Infrastructure	IaC/K8s/multi-region/cost	Local script-only changes	View
Data Engineering	Pipelines/streaming/data quality	CRUD app without data platform scope	View
Compliance & Governance	Regulated/security governance scope	Prototype with no compliance requirements	View
Monorepo & Complex Projects	Multi-package cross-cutting work	Small standalone repository	View
Error Handling & Resilience	Fault tolerance and degradation	Static content editing only	View
Developer Experience & Tooling	Linting/hooks/onboarding/DX	Feature logic changes only	View
Database Design & Optimization	Schema/index/query tuning	No persistent data layer exists	View
UI/UX & Design Systems	Design tokens/components/theming	API/backend-only task	View
Evidence-Driven Research	Source-verified research, comparisons, and consequential recommendations	A primary source already answers a simple fact	View

---

Common Combinations

Start with Agent System, then add specialists per the Composition Tiers — most rows below are Tier 1 (one specialist); rows with two specialists or Multi-Agent Orchestration are Tier 2/3 and need a stated reason for the extra load. These pairings recur:

Task	Setup
Add a feature to a web app	Agent System + Full-Stack Development (or Web Development project type)
Design or change an API	Agent System + API Design & GraphQL
Diagnose a production incident	Agent System + Debugging & Troubleshooting
Security-sensitive change	Agent System + Security Audit
Architecture decision	Agent System + Architecture Patterns
Pick the stack for a new project	Agent System + Technology Stack
Large migration or repo-wide refactor	Agent System + Migration & Upgrade + Multi-Agent Orchestration
Add test coverage to legacy code	Agent System + Testing
Schema or query performance	Agent System + Database Design & Optimization
Set up Claude Code for a repo	Claude Code Workflow (+ Agent Skills / Hooks & Automation / MCP Integration as needed)
Build an AI feature	Agent System + AI & LLM Integration + Model Selection Guide
Monorepo coordination	Agent System + Monorepo & Complex Projects

---

Archived (Merged) Prompts

These were removed from the active catalog to reduce overlap:

• Error Analysis
• Project Workflow
• Integration Guardian
• Claude Code Token Optimization
• Prompt Chaining

See Archive Index.

---

Quality Gate (Repository Standard)

Every active prompt must include:

• ## Role
• ## Protocol / Core Loop
• ## Phases
• ## Remember (final section)

And must follow No Vague Advice:

• each recommendation must end with a concrete decision, tool, or validation step.

Use the measurable review checklist: Prompt Review Checklist

---

90-Day Improvement Cycle

• Monthly: usage, token, and link hygiene review
• Quarterly: keep/merge/archive portfolio review and active catalog cleanup

Guide: Portfolio Maintenance Guide

---

Repository Structure

prompts/
└── english/
    ├── INDEX.md          # Global router: task -> file
    ├── agents/           # Active agent prompts + Claude Code coverage
    │   ├── INDEX.md
    │   └── archive/      # Archived prompts removed from active catalog
    ├── base/             # Foundation prompt
    ├── project-types/    # Domain-specific prompts
    ├── examples/         # Real-world usage examples
    └── workflows/        # Model selection, native features, Agent SDK,
        └── INDEX.md      #   APEI, setup, selection, troubleshooting, maintenance

Resources

• REPOSITORY-MAP.md — 30-second orientation, jump straight to a file
• QUICK-START.md
• USAGE.md
• CONTRIBUTING.md
• CHANGELOG.md
• llms.txt — full LLM router index
• Agent Index
• Prompt Index
• Workflows Index
• Prompt Selector Guide
• Prompt Review Checklist
• Evals — routing-accuracy regression tests, run in CI on every PR
• CI workflow — markdownlint, link audit, skill audit, deterministic-checks, plugin validation, routing eval