{"id":50430558,"url":"https://github.com/unkownpr/prompt-engineering-skill","last_synced_at":"2026-05-31T14:01:42.626Z","repository":{"id":359330596,"uuid":"1245464354","full_name":"unkownpr/prompt-engineering-skill","owner":"unkownpr","description":"Agent Skill for prompt engineering, RAG, CoT, hallucination reduction, and self-improving LLM loops","archived":false,"fork":false,"pushed_at":"2026-05-21T11:09:02.000Z","size":56,"stargazers_count":2,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-21T19:27:36.661Z","etag":null,"topics":["agent-skill","agent-skills","claude-code","llm","prompt-engineering","rag","skills-sh"],"latest_commit_sha":null,"homepage":"https://ssilistre.dev","language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/unkownpr.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-05-21T08:43:30.000Z","updated_at":"2026-05-21T11:09:08.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/unkownpr/prompt-engineering-skill","commit_stats":null,"previous_names":["unkownpr/prompt-engineering-skill"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/unkownpr/prompt-engineering-skill","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unkownpr%2Fprompt-engineering-skill","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unkownpr%2Fprompt-engineering-skill/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unkownpr%2Fprompt-engineering-skill/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unkownpr%2Fprompt-engineering-skill/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/unkownpr","download_url":"https://codeload.github.com/unkownpr/prompt-engineering-skill/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unkownpr%2Fprompt-engineering-skill/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":33733754,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-05-31T02:00:06.040Z","response_time":95,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent-skill","agent-skills","claude-code","llm","prompt-engineering","rag","skills-sh"],"created_at":"2026-05-31T14:01:39.425Z","updated_at":"2026-05-31T14:01:42.620Z","avatar_url":"https://github.com/unkownpr.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\n\u003cimg src=\"https://media2.giphy.com/media/v1.Y2lkPTc5MGI3NjExbnNkbmZzYjFnYWR5N3p6Z2p4bWNlbnA1NnBxenBnZGVibWQ2OTY5biZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/e7CD41ekE5ZUSgTnkn/giphy.gif\" alt=\"Prompt Engineering Skill\" width=\"480\" /\u003e\n\n# 🧠 Prompt Engineering Skill\n\n### An Agent Skill for getting consistent, accurate, and reproducible output from LLMs\n\n[🇹🇷 Türkçe](./README.tr.md) \u0026nbsp;·\u0026nbsp; **🇬🇧 English**\n\n[![skills.sh](https://skills.sh/b/unkownpr/prompt-engineering-skill)](https://skills.sh/unkownpr/prompt-engineering-skill)\n[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)\n[![Agent Skills Spec](https://img.shields.io/badge/spec-agentskills.io-7c3aed)](https://agentskills.io)\n[![Claude Code](https://img.shields.io/badge/Claude%20Code-compatible-ea580c)](https://code.claude.com)\n\n**Hallucination ↓ \u0026nbsp;·\u0026nbsp; Consistency ↑ \u0026nbsp;·\u0026nbsp; Token efficiency ↑**\n\nChain-of-Thought, RAG, COSTAR, and self-improving loop techniques bundled into a single skill.\nCompatible with Claude Code, Cursor, Codex, Gemini CLI, and 50+ other agents.\n\n```bash\nnpx skills add unkownpr/prompt-engineering-skill\n```\n\n\u003c/div\u003e\n\n---\n\n## 🎯 What This Skill Does\n\nMakes prompt engineering systematic in AI projects. When the agent detects one of these situations, the skill auto-activates and treats your request like an experienced prompt engineer would:\n\n| Situation | How the Skill Helps |\n|-----------|---------------------|\n| You're getting broken or incomplete output from an LLM | Finds the root cause of the hallucination and suggests preventive techniques |\n| You're writing a new prompt from scratch | Picks the right scaffold (COSTAR or Chain-of-Thought) and delivers a ready skeleton |\n| You're building a RAG system | Concrete suggestions for retrieval quality, chunking, re-ranking, and citation patterns |\n| You're designing an agent or autonomous workflow | Produces multi-step prompt chains, role assignment, and guardrails |\n| You're comparing multiple models | Model-specific optimization for Claude, GPT, Qwen, Kimi, and MiniMax |\n| You don't know if your prompt actually works | Sets up a golden dataset and evaluation framework so quality becomes measurable |\n\n---\n\n## ✨ How It Produces Quality Prompts\n\nThe difference here is that the skill doesn't blindly apply a single template. The moment it receives your request, it **synthesizes information from multiple reference files**:\n\n1. **Classifies the task.** It looks at what you're asking for: code generation, classification, summarization, agent workflow? The task type drives template selection.\n2. **Picks the right framework.** Pulls the most suitable one from COSTAR, RTF, CRISPE, or ICE — defined in `references/frameworks.md`. It doesn't just copy the doc — it adapts the framework to your context.\n3. **Adds the reasoning technique.** If the task requires multi-step thinking, it pulls Chain-of-Thought, Tree-of-Thought, or self-consistency patterns from `references/cot-techniques.md`. For single-step tasks, this step is skipped.\n4. **Applies anti-hallucination rules.** From `references/anti-hallucination.md`: forced \"I don't know\" language under uncertainty, mandatory source attribution, citation patterns, and grounding rules.\n5. **Optimizes for the target model.** Claude → XML tags. GPT → markdown headers. Qwen/Kimi → CJK-friendly patterns. Details in `references/model-selection.md`.\n6. **Integrates RAG or retrieval if present.** If the system uses RAG, it pulls source citation format, chunk references, and fallback behavior from `references/rag-tips.md`.\n7. **Makes the output evaluable.** Attaches 3-5 input-output example pairs, a golden dataset suggestion, and an LLM-as-judge rubric. The \"does it work?\" question turns from a guess into a measurement.\n\nThe prompt you receive isn't the output of one template — it's a **synthesis of 4-6 different techniques selected for your task type**. Every suggestion comes with the underlying reference file, so you can drill deeper when needed.\n\n---\n\n## 🪶 How Token Efficiency Is Achieved\n\nThe skill is written around the **progressive disclosure** principle. Not every piece of information loads into context every time:\n\n1. **Discovery phase** — On startup, the agent only reads the skill's `name` and `description`. Total cost: ~250 tokens.\n2. **Activation phase** — When a trigger word appears in conversation (e.g. \"hallucination\", \"RAG\", \"COSTAR\"), the full `SKILL.md` loads.\n3. **Reference phase** — Files under `references/` and `templates/` are read only as the current task demands. They don't all enter context at once.\n\nBeyond that, the skill applies these practices internally:\n\n- **Split reference files** — Instead of one giant doc, 9 topic-scoped references. The agent loads only what it needs.\n- **Ready-made templates** — Frequently used COSTAR, CoT, and RAG scaffolds live under `templates/`. No tokens spent regenerating them.\n- **Sharp trigger words** — The keywords in `description` prevent the agent from loading the skill unnecessarily.\n\nResult: even though the skill contains 10+ files and thousands of lines, the load it adds to a typical conversation stays very low.\n\n---\n\n## ⚡ Quick Install\n\n**Install globally to all supported agents:**\n\n```bash\nnpx skills add unkownpr/prompt-engineering-skill -g\n```\n\n**Install to Claude Code only:**\n\n```bash\nnpx skills add unkownpr/prompt-engineering-skill -g -a claude-code\n```\n\n**List the contents first:**\n\n```bash\nnpx skills add unkownpr/prompt-engineering-skill --list\n```\n\n\u003e [!TIP]\n\u003e The `-g` flag installs globally (`~/.claude/skills/`). Without the flag, install is project-scoped (`.claude/skills/`).\n\n---\n\n## 🔥 Trigger Words\n\nThe skill activates when the agent picks up one of these keywords:\n\n\u003ctable\u003e\n\u003ctr\u003e\n\u003ctd\u003e\u003cb\u003eWriting \u0026 Improving\u003c/b\u003e\u003c/td\u003e\n\u003ctd\u003e\n\n`write a prompt` · `improve prompt` · `system prompt` · `agent prompt`\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\u003cb\u003eQuality \u0026 Errors\u003c/b\u003e\u003c/td\u003e\n\u003ctd\u003e\n\n`hallucination` · `LLM output quality` · `prompt injection` · `format error`\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\u003cb\u003eTechniques\u003c/b\u003e\u003c/td\u003e\n\u003ctd\u003e\n\n`Chain of Thought` · `CoT` · `RAG` · `COSTAR` · `few-shot` · `role assignment`\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\u003cb\u003eEvaluation\u003c/b\u003e\u003c/td\u003e\n\u003ctd\u003e\n\n`prompt evaluation` · `golden dataset` · `LLM-as-judge` · `regression test`\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003e\u003cb\u003eModel Selection\u003c/b\u003e\u003c/td\u003e\n\u003ctd\u003e\n\n`Claude prompt` · `GPT prompt` · `Qwen prompt` · `Kimi prompt` · `MiniMax prompt`\n\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n## 📦 Skill Contents\n\n```\nprompt-engineering-skill/\n├── README.md\n└── skills/\n    └── prompt-engineering/\n        ├── SKILL.md                     # Main skill instructions + frontmatter\n        ├── references/\n        │   ├── agentic-loops.md         # Self-improving loops, ReAct, reflection\n        │   ├── anti-hallucination.md    # Hallucination prevention techniques\n        │   ├── cot-techniques.md        # Chain-of-Thought variants\n        │   ├── evaluation.md            # Prompt evaluation + golden dataset\n        │   ├── frameworks.md            # COSTAR, RTF, CRISPE, ICE\n        │   ├── mempalace-integration.md # Memory palace + long-context patterns\n        │   ├── model-selection.md       # Claude / GPT / Qwen / Kimi / MiniMax comparison\n        │   ├── prompt-chaining.md       # Multi-step prompt chaining\n        │   └── rag-tips.md              # Retrieval, chunking, re-ranking\n        └── templates/\n            ├── costar-template.md       # COSTAR boilerplate\n            ├── cot-template.md          # Chain-of-Thought boilerplate\n            └── rag-template.md          # RAG system prompt boilerplate\n```\n\n---\n\n## 🧩 Supported Agents\n\nThe skill is **compatible with 50+ agents** because it follows the Agent Skills spec:\n\nClaude Code · Cursor · Codex · Gemini CLI · OpenCode · GitHub Copilot · Amp · Goose · Roo Code · Continue · Kiro · Crush · Windsurf · Factory · Junie · and [more](https://agentskills.io/clients)\n\n---\n\n## 💡 Example\n\nAfter install, just talk to the agent normally. The skill triggers itself.\n\n**Your request:**\n\n\u003e \"Write a prompt that classifies e-commerce support emails into complaint, refund_request, shipping_query, or general_question. It will run on Claude. It must not misclassify.\"\n\n**The prompt the skill produces (abbreviated):**\n\n````markdown\nYou are an email triage assistant for an e-commerce company.\nClassify the incoming customer email into exactly one of these categories:\n\n\u003ccategories\u003e\n- complaint        — product quality, service, or experience issues\n- refund_request   — explicit refund, return, or money-back asks\n- shipping_query   — delivery status, ETA, tracking, address change\n- general_question — anything that doesn't match above\n\u003c/categories\u003e\n\n\u003crules\u003e\n1. If the email touches multiple topics, pick the PRIMARY intent —\n   the action the customer is asking for.\n2. If you cannot confidently classify, output \"general_question\".\n   Do NOT invent a new category.\n3. Output ONLY the category slug. No explanation, no greeting,\n   no markdown.\n\u003c/rules\u003e\n\n\u003cexamples\u003e\n\u003cemail\u003eOrder #4521 still hasn't arrived, been 2 weeks.\u003c/email\u003e\n\u003ccategory\u003eshipping_query\u003c/category\u003e\n\n\u003cemail\u003eThe headphones broke after 3 days. I want my money back.\u003c/email\u003e\n\u003ccategory\u003erefund_request\u003c/category\u003e\n\n\u003cemail\u003eDo you ship to Turkey?\u003c/email\u003e\n\u003ccategory\u003egeneral_question\u003c/category\u003e\n\u003c/examples\u003e\n\n\u003cemail\u003e\n{{user_email}}\n\u003c/email\u003e\n````\n\n**The skill also ships with:**\n- A 10-example golden dataset (2-3 per category)\n- An LLM-as-judge rubric (3 criteria: correct category, no unwanted output, edge case handling)\n- An edge case list (empty email, multi-intent, sarcastic tone, foreign language)\n- A regression test plan to run before production rollout\n\n---\n\n## 📄 License \u0026 Contact\n\nMIT \u0026copy; [ssilistre](https://ssilistre.dev) \u0026nbsp;·\u0026nbsp; 🐙 [@unkownpr](https://github.com/unkownpr)\n\nFeedback and feature requests welcome at [Issues](https://github.com/unkownpr/prompt-engineering-skill/issues). PRs welcome.\n\n---\n\n\u003cdiv align=\"center\"\u003e\n\n### 💎 Sponsor\n\n\u003ca href=\"https://yapayzekapi.store\" target=\"_blank\"\u003e\n  \u003cimg src=\"https://i.hizliresim.com/snpojv7.png\" alt=\"yapayzekapi.store — One API Key, All AI Models\" width=\"640\" /\u003e\n\u003c/a\u003e\n\n**[yapayzekapi.store](https://yapayzekapi.store) — One API Key, All AI Models**\n\nSmart routing and Chinese model families with a single API key.\nUse directly in Cursor, VS Code, and Cline.\n\n**[💰 View Prices →](https://yapayzekapi.store)**\n\n\u003c/div\u003e\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Funkownpr%2Fprompt-engineering-skill","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Funkownpr%2Fprompt-engineering-skill","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Funkownpr%2Fprompt-engineering-skill/lists"}