{"id":"439abbf4-b003-42f1-8e07-294496d8b0a5","shortId":"YGYNGa","kind":"skill","title":"prompt-engineering","tagline":"Universal prompt engineering techniques for any LLM. Use when crafting, optimizing, or reviewing prompts for AI models. Triggers on requests like \"improve this prompt\", \"write a system prompt\", \"optimize my instructions\", \"help me prompt engineer\", \"audit this prompt\", \"review my","description":"# Prompt Engineering\n\nUniversal techniques for crafting effective prompts across any LLM.\n\n## Core Principles\n\n### 1. Structure with XML Tags\n\nUse XML tags to create clear, parseable prompts:\n\n```xml\n<context>Background information here</context>\n<instructions>\n1. First step\n2. Second step\n</instructions>\n<examples>Sample inputs/outputs</examples>\n<output_format>Expected structure</output_format>\n```\n\n**Benefits:**\n- **Clarity**: Separates context, instructions, and examples\n- **Accuracy**: Prevents model from mixing up sections\n- **Flexibility**: Easy to modify individual parts\n- **Parseability**: Enables structured output extraction\n\n**Best practices:**\n- Use consistent tag names throughout (`<instructions>`, not sometimes `<steps>`)\n- Reference tags explicitly: \"Using the data in `<context>` tags...\"\n- Nest tags for hierarchy: `<examples><example id=\"1\">...</example></examples>`\n- Combine with other techniques: `<thinking>` for chain-of-thought, `<answer>` for final output\n\n### 2. Control Output Shape\n\nSpecify explicit constraints on length, format, and structure:\n\n```xml\n<output_spec>\n- Default: 3-6 sentences or ≤5 bullets\n- Simple yes/no questions: ≤2 sentences\n- Complex multi-step tasks:\n  - 1 short overview paragraph\n  - ≤5 bullets: What changed, Where, Risks, Next steps, Open questions\n- Use Markdown with headers, bullets, tables when helpful\n- Avoid long narrative paragraphs; prefer compact structure\n</output_spec>\n```\n\n### 3. Prevent Scope Drift\n\nExplicitly constrain what the model should NOT do:\n\n```xml\n<constraints>\n- Implement EXACTLY and ONLY what is requested\n- No extra features, components, or embellishments\n- If ambiguous, choose the simplest valid interpretation\n- Do NOT invent values, make assumptions, or add unrequested elements\n</constraints>\n```\n\n### 4. Handle Ambiguity Explicitly\n\nPrevent hallucinations and overconfidence:\n\n```xml\n<uncertainty_handling>\n- If the question is ambiguous:\n  - Ask 1-3 precise clarifying questions, OR\n  - Present 2-3 plausible interpretations with labeled assumptions\n- When facts may have changed: answer in general terms, state uncertainty\n- Never fabricate exact figures or references when uncertain\n- Prefer \"Based on the provided context...\" over absolute claims\n</uncertainty_handling>\n```\n\n### 5. Long-Context Grounding\n\nFor inputs >10k tokens, add re-grounding instructions:\n\n```xml\n<long_context_handling>\n- First, produce a short internal outline of key sections relevant to the request\n- Re-state user constraints explicitly before answering\n- Anchor claims to sections (\"In the 'Data Retention' section...\")\n- Quote or paraphrase fine details (dates, thresholds, clauses)\n</long_context_handling>\n```\n\n## Agentic Prompts\n\n### Tool Usage Rules\n\n```xml\n<tool_usage>\n- Prefer tools over internal knowledge for:\n  - Fresh or user-specific data (tickets, orders, configs)\n  - Specific IDs, URLs, or document references\n- Parallelize independent reads when possible\n- After write operations, restate: what changed, where, any validation performed\n</tool_usage>\n```\n\n### User Updates\n\n```xml\n<user_updates>\n- Send brief updates (1-2 sentences) only when:\n  - Starting a new major phase\n  - Discovering something that changes the plan\n- Avoid narrating routine operations\n- Each update must include a concrete outcome (\"Found X\", \"Updated Y\")\n- Do not expand scope beyond what was asked\n</user_updates>\n```\n\n### Self-Check for High-Risk Outputs\n\n```xml\n<self_check>\nBefore finalizing answers in sensitive contexts (legal, financial, safety):\n- Re-scan for unstated assumptions\n- Check for ungrounded numbers or claims\n- Soften overly strong language (\"always\", \"guaranteed\")\n- Explicitly state assumptions\n</self_check>\n```\n\n## Structured Extraction\n\nFor data extraction tasks, always provide a schema:\n\n```xml\n<extraction_spec>\nExtract data into this exact schema (no extra fields):\n{\n  \"field_name\": \"string\",\n  \"optional_field\": \"string | null\",\n  \"numeric_field\": \"number | null\"\n}\n- If a field is not present in source, set to null (don't guess)\n- Re-scan source for missed fields before returning\n</extraction_spec>\n```\n\n## Web Research Prompts\n\n```xml\n<research_guidelines>\n- Browse the web for: time-sensitive topics, recommendations, navigational queries, ambiguous terms\n- Include citations after paragraphs with web-derived claims\n- Use multiple sources for key claims; prioritize primary sources\n- Research until additional searching won't materially change the answer\n- Structure output with Markdown: headers, bullets, tables for comparisons\n</research_guidelines>\n```\n\n## Example: Before/After\n\n**Without structure:**\n```\nYou're a financial analyst. Generate a Q2 report for investors. Include Revenue, Margins, Cash Flow. Use this data: {{DATA}}. Make it professional and concise.\n```\n\n**With structure:**\n```xml\nYou're a financial analyst at AcmeCorp generating a Q2 report for investors.\n\n<context>\nAcmeCorp is a B2B SaaS company. Investors value transparency and actionable insights.\n</context>\n\n<data>\n{{DATA}}\n</data>\n\n<instructions>\n1. Include sections: Revenue Growth, Profit Margins, Cash Flow\n2. Highlight strengths and areas for improvement\n3. Use concise, professional tone\n</instructions>\n\n<output_format>\n- Use bullet points with metrics and YoY changes\n- Include \"Action:\" items for areas needing improvement\n- End with 2-3 bullet Outlook section\n</output_format>\n```\n\n## Prompt Migration Checklist\n\nWhen adapting prompts across models or versions:\n\n1. **Switch model, keep prompt identical** — isolate the variable\n2. **Pin reasoning/thinking depth** to match prior model's profile\n3. **Run evals** — if results are good, ship\n4. **If regressions, tune prompt** — adjust verbosity/format/scope constraints\n5. **Re-eval after each small change** — one change at a time\n\n## Quick Reference\n\n| Technique | Tag Pattern | Use Case |\n|-----------|-------------|----------|\n| Separate sections | `<context>`, `<instructions>`, `<data>` | Any complex prompt |\n| Control length | `<output_spec>` with word/bullet limits | Prevent verbosity |\n| Prevent drift | `<constraints>` with explicit \"do NOT\" | Feature creep |\n| Handle uncertainty | `<uncertainty_handling>` | Factual queries |\n| Chain of thought | `<thinking>`, `<answer>` | Reasoning tasks |\n| Extraction | `<schema>` with JSON structure | Data parsing |\n| Research | `<research_guidelines>` | Web-enabled agents |\n| Self-check | `<self_check>` | High-risk domains |\n| Tool usage | `<tool_usage_rules>` | Agentic systems |\n| Eagerness control | `<persistence>`, `<context_gathering>` | Agent autonomy |\n| Persona | `<role>` + behavioral constraints | Tone & style |\n\n## Prompting Techniques Catalog\n\nComprehensive catalog of prompting techniques. Full details, examples, and academic references in [references/prompting-techniques.md](references/prompting-techniques.md).\n\n| Technique | Use Case |\n|-----------|----------|\n| **Zero-Shot Prompting** | Direct task execution without examples; classification, translation, summarization |\n| **Few-Shot Prompting** | In-context learning via exemplars; format control, label calibration, style matching |\n| **Chain-of-Thought (CoT)** | Step-by-step reasoning; arithmetic, logic, commonsense reasoning tasks |\n| **Meta Prompting** | LLM as orchestrator delegating to specialized expert prompts; complex multi-domain tasks |\n| **Self-Consistency** | Sample multiple CoT paths, pick majority answer; boost accuracy on math & reasoning |\n| **Generated Knowledge** | Generate relevant knowledge first, then answer; commonsense & factual QA |\n| **Prompt Chaining** | Break complex tasks into sequential subtasks; document analysis, multi-step workflows |\n| **Tree of Thoughts (ToT)** | Explore multiple reasoning branches with lookahead/backtracking; planning, puzzles |\n| **RAG** | Retrieve external documents before generating; knowledge-intensive tasks, fresh data |\n| **ART (Auto Reasoning + Tools)** | Auto-select and orchestrate tools with CoT; tasks requiring calculation, search, APIs |\n| **APE (Auto Prompt Engineer)** | LLM generates and scores candidate prompts; prompt optimization at scale |\n| **Active-Prompt** | Identify uncertain examples, annotate selectively for CoT; adaptive few-shot |\n| **Directional Stimulus** | Add a hint/keyword to guide generation direction; summarization, dialogue |\n| **PAL (Program-Aided LM)** | Generate code instead of text for reasoning; math, data manipulation, symbolic tasks |\n| **ReAct** | Interleave reasoning traces with tool actions; search, QA, decision-making agents |\n| **Reflexion** | Agent self-reflects on failures with verbal feedback; iterative improvement, debugging |\n| **Multimodal CoT** | Two-stage: rationale generation then answer with text+image; visual reasoning tasks |\n| **Graph Prompting** | Structured graph-based prompts; node classification, relation extraction, graph tasks |\n\n### Prompting Fundamentals\n\nLLM settings, prompt elements, formatting, and practical examples — see [references/prompting-introduction.md](references/prompting-introduction.md). Covers:\n- **LLM Settings** — temperature, top-p, max length, stop sequences, frequency/presence penalties\n- **Prompt Elements** — instruction, context, input data, output indicator\n- **Design Tips** — start simple, be specific, avoid impreciseness, say what TO do (not what NOT to do)\n- **Task Examples** — summarization, extraction, QA, classification, conversation, code generation, reasoning\n\n### Risks & Misuses\n\nAdversarial attacks, factuality issues, and bias mitigation — see [references/prompting-risks.md](references/prompting-risks.md). Covers:\n- **Adversarial Prompting** — prompt injection, prompt leaking, jailbreaking (DAN, Waluigi Effect), defense tactics\n- **Factuality** — ground truth grounding, calibrated confidence, admit-ignorance patterns\n- **Biases** — exemplar distribution skew, exemplar ordering effects, balanced few-shot design\n\n## Prompt Audit / Review\n\nWhen asked to audit, review, or improve a prompt, follow this workflow. Full checklist with per-check references: [prompt-audit-checklist.md](references/prompt-audit-checklist.md).\n\n### Workflow\n\n1. **Read the prompt fully** — identify its purpose, target model, and deployment context (interactive chat, agentic system, batch pipeline, RAG-augmented)\n2. **Walk 8 dimensions** — check each, note issues with severity (Critical / Warning / Suggestion):\n\n| # | Dimension | What to Check |\n|---|-----------|---------------|\n| 1 | **Clarity & Specificity** | Task definition, success criteria, audience, output format, conflicting constraints |\n| 2 | **Structure & Formatting** | Section separation (XML tags), prompt smells (monolithic, mixed layers, negative bias) |\n| 3 | **Safety & Security** | Control/data separation, secrets in prompt, injection resilience, tool permissions |\n| 4 | **Hallucination & Factuality** | Role framing, grounding, citation-without-sources, uncertainty handling |\n| 5 | **Context Management** | Info placement (not buried in middle), context size, RAG doc count, re-grounding |\n| 6 | **Maintainability & Debt** | Hardcoded values, regenerated logic, model pinning, testability |\n| 7 | **Model-Specific Fit** | Model-specific params and gotchas (see Model-Specific Guides below) |\n| 8 | **Evaluation Readiness** | Eval criteria, adversarial test cases, schema enforcement, monitoring |\n\n3. **Produce a report** — issues table (dimension, check, severity, issue, fix) + rewritten prompt or targeted fix suggestions. Use the report template from the checklist reference.\n4. **For each issue**, cite the relevant reference file so the user can dive deeper.\n\n### Quick Decision: Which Dimensions to Prioritize\n\n- **User-facing chatbot** → prioritize Safety (#3), Hallucination (#4), Clarity (#1)\n- **Agentic system with tools** → prioritize Safety (#3), Context (#5), Maintainability (#6)\n- **Batch/pipeline** → prioritize Structure (#2), Evaluation (#8), Maintainability (#6)\n- **RAG-augmented** → prioritize Context (#5), Safety (#3), Hallucination (#4)\n\n## Common Mistakes & Anti-Patterns\n\nThree complementary layers — use the one matching your need:\n\n**Deep-dives by category** — root causes, mechanisms, prevention checklists (from \"The Architecture of Instruction\", 2026):\n\n| Mistake Category | Key Issues | Reference |\n|-----------------|------------|-----------|\n| **Hallucinations & Logic** | Ambiguity-induced confabulation, automation bias, overloaded prompts, logical failures in verification tasks, no role framing | [mistakes-hallucinations.md](references/mistakes-hallucinations.md) |\n| **Structural Fragility** | Formatting sensitivity (up to 76pp variance), reproducibility crisis, prompt smells catalog (6 anti-patterns), deliberation ladder | [mistakes-structure.md](references/mistakes-structure.md) |\n| **Context Rot** | \"Lost in the middle\" U-shaped attention, RAG over-retrieval, naive data loading, context engineering shift | [mistakes-context.md](references/mistakes-context.md) |\n| **Prompt Debt** | Token tax of regenerative code, debt taxonomy (prompt/hyperparameter/framework/cost), multi-agent solutions, automated repair | [mistakes-debt.md](references/mistakes-debt.md) |\n| **Security** | Direct/indirect injection, jailbreaking, system prompt leakage (OWASP LLM07:2025), RAG poisoning, multimodal injection, adversarial suffixes | [mistakes-security.md](references/mistakes-security.md) |\n\n**Quick reference** — 18-category taxonomy with MRPs, risk scores, case studies, action items: [failure-taxonomy.md](references/failure-taxonomy.md). Start here for an overview or to prioritize which categories to address first. Covers: control-plane vs data-plane model, heuristic risk scoring, real-world incidents (EchoLeak CVE-2025-32711, Mata v. Avianca, Samsung shadow AI).\n\n**How to measure & test** — eval metrics, CI gating, red-teaming, tooling: [evaluation-redteaming.md](references/evaluation-redteaming.md). Covers: TruthfulQA, FActScore, SelfCheckGPT, PromptBench, AILuminate, LLM-as-judge pitfalls, guardrail libraries, open research questions.\n\n## Model-Specific Guides\n\nEach model family has unique parameters, gotchas, and patterns. Consult the reference for your target model:\n\n- **[Claude Family](references/claude-family-prompting.md)** — Opus 4.7 / 4.6 / Sonnet 4.6 / 4.5 / Haiku 4.5: adaptive thinking (`effort` with new `xhigh` on 4.7), `task_budget` agentic-loop ceiling, legacy `thinking.budget_tokens` 400-error on 4.7, new tokenizer (~1.35× text, ~3× images), tool under-triggering on 4.7 (vs 4.6 over-triggering), more literal instruction-following, server-side compaction beta, Managed Agents memory beta, Cyber Verification gate, prefill deprecation, Structured Outputs, prompt caching, citations, context engineering, vision crop tool, migration paths 4.5 → 4.6 → 4.7\n- **[GPT-5 Family](references/gpt5-family-prompting.md)** — GPT-5 / 5.1 / 5.2 / 5.4 / 5.5: `reasoning_effort` (last-mile knob in 5.4/5.5), `text.verbosity`, named tools (`apply_patch`), agentic eagerness templates, completeness/verification contracts, compaction API, `phase` field, outcome-first prompts, personality vs collaboration style, retrieval budgets, mini/nano guidance, migration paths\n- **[Gemini 3 Family](references/gemini3-family-prompting.md)** — Gemini 2.5/3/3.1: temperature MUST be 1.0, `thinking_budget` vs `thinking_level`, constraint placement (end of prompt), persona priority, function calling, structured output, multimodal, image generation\n- **[GPT-5.2 Specifics](references/gpt5-prompting-guide.md)** — Compaction API code examples, web research agent prompt, full XML specification blocks","tags":["prompt","engineering","driven","development","codealive-ai","agent-safety","agent-skills","ai-coding","ai-driven-development","ai-safety","antigravity","bash"],"capabilities":["skill","source-codealive-ai","skill-prompt-engineering","topic-agent-safety","topic-agent-skills","topic-ai-coding","topic-ai-driven-development","topic-ai-safety","topic-antigravity","topic-bash","topic-claude-code","topic-codex-cli","topic-cursor","topic-developer-tools","topic-gemini-cli"],"categories":["ai-driven-development"],"synonyms":[],"warnings":[],"endpointUrl":"https://skills.sh/CodeAlive-AI/ai-driven-development/prompt-engineering","protocol":"skill","transport":"skills-sh","auth":{"type":"none","details":{"cli":"npx skills add CodeAlive-AI/ai-driven-development","source_repo":"https://github.com/CodeAlive-AI/ai-driven-development","install_from":"skills.sh"}},"qualityScore":"0.483","qualityRationale":"deterministic score 0.48 from registry signals: · indexed on github topic:agent-skills · 67 github stars · SKILL.md body (15,592 chars)","verified":false,"liveness":"unknown","lastLivenessCheck":null,"agentReviews":{"count":0,"score_avg":null,"cost_usd_avg":null,"success_rate":null,"latency_p50_ms":null,"narrative_summary":null,"summary_updated_at":null},"enrichmentModel":"deterministic:skill-github:v1","enrichmentVersion":1,"enrichedAt":"2026-05-18T18:57:07.245Z","embedding":null,"createdAt":"2026-05-04T06:56:23.611Z","updatedAt":"2026-05-18T18:57:07.245Z","lastSeenAt":"2026-05-18T18:57:07.245Z","tsv":"'-2':403 '-2025':1626 '-3':260,267,685 '-32711':1627 '-5':1768,1772 '-5.2':1845 '-6':157 '/3/3.1':1820 '/5.5':1785 '1':57,74,172,259,402,646,699,1216,1255,1416 '1.0':1824 '1.35':1718 '10k':308 '18':1582 '2':77,142,165,266,655,684,708,1238,1267,1431 '2.5':1819 '2025':1571 '2026':1475 '3':156,201,662,718,1281,1360,1412,1423,1443,1720,1815 '4':244,726,1293,1385,1414,1445 '4.5':1692,1694,1764 '4.6':1689,1691,1729,1765 '4.7':1688,1702,1715,1727,1766 '400':1712 '5':160,176,301,734,1305,1425,1441 '5.1':1773 '5.2':1774 '5.4':1775,1784 '5.5':1776 '6':1322,1427,1435,1514 '7':1332 '76pp':1507 '8':1240,1349,1433 'absolut':299 'academ':826 'accuraci':91,903 'acmecorp':626,633 'across':52,695 'action':643,676,1035,1591 'activ':988 'active-prompt':987 'adapt':693,997,1695 'add':241,310,1003 'addit':571 'address':1606 'adjust':731 'admit':1176 'admit-ignor':1175 'adversari':1146,1157,1354,1576 'agent':354,793,803,807,1041,1043,1231,1417,1556,1706,1744,1791,1854 'agentic-loop':1705 'ai':19,1633 'aid':1015 'ailumin':1653 'alway':475,486 'ambigu':228,246,257,549,1484 'ambiguity-induc':1483 'analysi':927 'analyst':596,624 'anchor':337 'annot':993 'answer':278,336,452,578,901,914,1063 'anti':1449,1516 'anti-pattern':1448,1515 'ape':973 'api':972,1797,1849 'appli':1789 'architectur':1472 'area':659,679 'arithmet':872 'art':956 'ask':258,440,1195 'assumpt':239,272,464,479 'attack':1147 'attent':1531 'audienc':1262 'audit':39,1192,1197 'augment':1237,1438 'auto':957,961,974 'auto-select':960 'autom':1487,1558 'autonomi':808 'avianca':1630 'avoid':194,418,1123 'b2b':636 'background':71 'balanc':1186 'base':293,1075 'batch':1233 'batch/pipeline':1428 'before/after':589 'behavior':810 'benefit':84 'best':109 'beta':1742,1746 'beyond':437 'bias':1151,1179,1280,1488 'block':1859 'boost':902 'branch':939 'break':920 'brief':400 'brows':538 'budget':1704,1809,1826 'bullet':161,177,190,584,668,686 'buri':1311 'cach':1755 'calcul':970 'calibr':859,1173 'call':1838 'candid':981 'case':753,833,1356,1589 'cash':606,653 'catalog':816,818,1513 'categori':1464,1477,1583,1604 'caus':1466 'ceil':1708 'chain':136,778,863,919 'chain-of-thought':135,862 'chang':179,277,391,415,576,674,741,743 'chat':1230 'chatbot':1409 'check':443,465,796,1211,1242,1254,1367 'checklist':691,1207,1383,1469 'choos':229 'ci':1640 'citat':552,1300,1756 'citation-without-sourc':1299 'cite':1389 'claim':300,338,470,559,565 'clarifi':262 'clariti':85,1256,1415 'classif':843,1078,1139 'claud':1684 'claus':353 'clear':67 'code':1018,1141,1550,1850 'collabor':1806 'combin':130 'common':1446 'commonsens':874,915 'compact':199,1741,1796,1848 'compani':638 'comparison':587 'complementari':1452 'completeness/verification':1794 'complex':167,757,887,921 'compon':224 'comprehens':817 'concis':616,664 'concret':427 'confabul':1486 'confid':1174 'config':374 'conflict':1265 'consist':112,894 'constrain':206 'constraint':148,333,733,811,1266,1830 'consult':1677 'context':87,297,304,455,852,1112,1228,1306,1314,1424,1440,1522,1539,1757 'contract':1795 'control':143,759,806,857,1610 'control-plan':1609 'control/data':1284 'convers':1140 'core':55 'cot':866,897,967,996,1056 'count':1318 'cover':1096,1156,1608,1648 'craft':13,49 'creat':66 'creep':773 'crisi':1510 'criteria':1261,1353 'critic':1248 'crop':1760 'cve':1625 'cyber':1747 'dan':1164 'data':123,343,371,483,492,610,611,645,787,955,1025,1114,1537,1614 'data-plan':1613 'date':351 'debt':1324,1545,1551 'debug':1054 'decis':1039,1401 'decision-mak':1038 'deep':1461 'deep-div':1460 'deeper':1399 'default':155 'defens':1167 'definit':1259 'deleg':882 'deliber':1518 'deploy':1227 'deprec':1751 'depth':711 'deriv':558 'design':1117,1190 'detail':350,823 'dialogu':1011 'dimens':1241,1251,1366,1403 'direct':838,1001,1009 'direct/indirect':1563 'discov':412 'distribut':1181 'dive':1398,1462 'doc':1317 'document':379,926,947 'domain':800,890 'drift':204,767 'eager':805,1792 'easi':99 'echoleak':1624 'effect':50,1166,1185 'effort':1697,1778 'element':243,1088,1110 'embellish':226 'enabl':105,792 'end':682,1832 'enforc':1358 'engin':3,6,38,45,976,1540,1758 'error':1713 'eval':720,737,1352,1638 'evalu':1350,1432 'evaluation-redteaming.md':1646 'exact':215,286,495 'exampl':90,588,824,842,992,1092,1135,1851 'execut':840 'exemplar':855,1180,1183 'expand':435 'expect':82 'expert':885 'explicit':120,147,205,247,334,477,769 'explor':936 'extern':946 'extra':222,498 'extract':108,481,484,491,783,1080,1137 'fabric':285 'face':1408 'fact':274 'factscor':1650 'factual':776,916,1148,1169,1295 'failur':1048,1492 'failure-taxonomy.md':1593 'famili':1670,1685,1769,1816 'featur':223,772 'feedback':1051 'few-shot':846,998,1187 'field':499,500,504,508,513,531,1799 'figur':287 'file':1393 'final':140,451 'financi':457,595,623 'fine':349 'first':75,316,912,1607,1802 'fit':1336 'fix':1370,1375 'flexibl':98 'flow':607,654 'follow':1203,1737 'format':151,856,1089,1264,1269,1503 'found':429 'fragil':1502 'frame':1297,1498 'frequency/presence':1107 'fresh':366,954 'full':822,1206,1856 'fulli':1220 'function':1837 'fundament':1084 'gate':1641,1749 'gemini':1814,1818 'general':280 'generat':597,627,907,909,949,978,1008,1017,1061,1142,1843 'good':724 'gotcha':1342,1674 'gpt':1767,1771,1844 'graph':1070,1074,1081 'graph-bas':1073 'ground':305,313,1170,1172,1298,1321 'growth':650 'guarante':476 'guardrail':1659 'guess':524 'guid':1007,1347,1667 'guidanc':1811 'haiku':1693 'hallucin':249,1294,1413,1444,1481 'handl':245,774,1304 'hardcod':1325 'header':189,583 'help':35,193 'heurist':1617 'hierarchi':129 'high':446,798 'high-risk':445,797 'highlight':656 'hint/keyword':1005 'id':376 'ident':704 'identifi':990,1221 'ignor':1177 'imag':1066,1721,1842 'implement':214 'imprecis':1124 'improv':25,661,681,1053,1200 'in-context':850 'incid':1623 'includ':425,551,603,647,675 'independ':382 'indic':1116 'individu':102 'induc':1485 'info':1308 'inform':72 'inject':1160,1289,1564,1575 'input':307,1113 'inputs/outputs':81 'insight':644 'instead':1019 'instruct':34,88,314,1111,1474,1736 'instruction-follow':1735 'intens':952 'interact':1229 'interleav':1030 'intern':320,363 'interpret':233,269 'invent':236 'investor':602,632,639 'isol':705 'issu':1149,1245,1364,1369,1388,1479 'item':677,1592 'iter':1052 'jailbreak':1163,1565 'json':785 'judg':1657 'keep':702 'key':323,564,1478 'knob':1782 'knowledg':364,908,911,951 'knowledge-intens':950 'label':271,858 'ladder':1519 'languag':474 'last':1780 'last-mil':1779 'layer':1278,1453 'leak':1162 'leakag':1568 'learn':853 'legaci':1709 'legal':456 'length':150,760,1104 'level':1829 'librari':1660 'like':24 'limit':763 'liter':1734 'llm':10,54,879,977,1085,1097,1655 'llm-as-judg':1654 'llm07':1570 'lm':1016 'load':1538 'logic':873,1328,1482,1491 'long':195,303 'long-context':302 'lookahead/backtracking':941 'loop':1707 'lost':1524 'maintain':1323,1426,1434 'major':410,900 'make':238,612,1040 'manag':1307,1743 'manipul':1026 'margin':605,652 'markdown':187,582 'mata':1628 'match':713,861,1457 'materi':575 'math':905,1024 'max':1103 'may':275 'measur':1636 'mechan':1467 'memori':1745 'meta':877 'metric':671,1639 'middl':1313,1527 'migrat':690,1762,1812 'mile':1781 'mini/nano':1810 'miss':530 'mistak':1447,1476 'mistakes-context.md':1542 'mistakes-debt.md':1560 'mistakes-hallucinations.md':1499 'mistakes-security.md':1578 'mistakes-structure.md':1520 'misus':1145 'mitig':1152 'mix':95,1277 'model':20,93,209,696,701,715,1225,1329,1334,1338,1345,1616,1665,1669,1683 'model-specif':1333,1337,1344,1664 'modifi':101 'monitor':1359 'monolith':1276 'mrps':1586 'multi':169,889,929,1555 'multi-ag':1554 'multi-domain':888 'multi-step':168,928 'multimod':1055,1574,1841 'multipl':561,896,937 'must':424,1822 'naiv':1536 'name':114,501,1787 'narrat':196,419 'navig':547 'need':680,1459 'negat':1279 'nest':126 'never':284 'new':409,1699,1716 'next':182 'node':1077 'note':1244 'null':506,510,521 'number':468,509 'numer':507 'one':742,1456 'open':184,1661 'oper':388,421 'optim':14,32,984 'option':503 'opus':1687 'orchestr':881,964 'order':373,1184 'outcom':428,1801 'outcome-first':1800 'outlin':321 'outlook':687 'output':107,141,144,448,580,1115,1263,1753,1840 'over':472 'over-retriev':1533 'over-trigg':1730 'overconfid':251 'overload':1489 'overview':174,1599 'owasp':1569 'p':1102 'pal':1012 'paragraph':175,197,554 'parallel':381 'param':1340 'paramet':1673 'paraphras':348 'pars':788 'parseabl':68,104 'part':103 'patch':1790 'path':898,1763,1813 'pattern':751,1178,1450,1517,1676 'penalti':1108 'per':1210 'per-check':1209 'perform':395 'permiss':1292 'person':1804 'persona':809,1835 'phase':411,1798 'pick':899 'pin':709,1330 'pipelin':1234 'pitfal':1658 'placement':1309,1831 'plan':417,942 'plane':1611,1615 'plausibl':268 'point':669 'poison':1573 'possibl':385 'practic':110,1091 'precis':261 'prefer':198,292,360 'prefil':1750 'present':265,516 'prevent':92,202,248,764,766,1468 'primari':567 'principl':56 'prior':714 'priorit':566,1405,1410,1421,1429,1439,1602 'prioriti':1836 'produc':317,1361 'profession':614,665 'profil':717 'profit':651 'program':1014 'program-aid':1013 'prompt':2,5,17,27,31,37,41,44,51,69,355,536,689,694,703,730,758,814,820,837,849,878,886,918,975,982,983,989,1071,1076,1083,1087,1109,1158,1159,1161,1191,1202,1219,1274,1288,1372,1490,1511,1544,1567,1754,1803,1834,1855 'prompt-audit-checklist.md':1213 'prompt-engin':1 'prompt/hyperparameter/framework/cost':1553 'promptbench':1652 'provid':296,487 'purpos':1223 'puzzl':943 'q2':599,629 'qa':917,1037,1138 'queri':548,777 'question':164,185,255,263,1663 'quick':747,1400,1580 'quot':346 'rag':944,1236,1316,1437,1532,1572 'rag-aug':1235,1436 'rational':1060 're':312,330,460,526,593,621,736,1320 're-ev':735 're-ground':311,1319 're-scan':459,525 're-stat':329 'react':1029 'read':383,1217 'readi':1351 'real':1621 'real-world':1620 'reason':781,871,875,906,938,958,1023,1031,1068,1143,1777 'reasoning/thinking':710 'recommend':546 'red':1643 'red-team':1642 'refer':118,289,380,748,827,1212,1384,1392,1480,1581,1679 'references/claude-family-prompting.md':1686 'references/evaluation-redteaming.md':1647 'references/failure-taxonomy.md':1594 'references/gemini3-family-prompting.md':1817 'references/gpt5-family-prompting.md':1770 'references/gpt5-prompting-guide.md':1847 'references/mistakes-context.md':1543 'references/mistakes-debt.md':1561 'references/mistakes-hallucinations.md':1500 'references/mistakes-security.md':1579 'references/mistakes-structure.md':1521 'references/prompt-audit-checklist.md':1214 'references/prompting-introduction.md':1094,1095 'references/prompting-risks.md':1154,1155 'references/prompting-techniques.md':829,830 'reflect':1046 'reflexion':1042 'regen':1549 'regener':1327 'regress':728 'relat':1079 'relev':325,910,1391 'repair':1559 'report':600,630,1363,1379 'reproduc':1509 'request':23,220,328 'requir':969 'research':535,569,789,1662,1853 'resili':1290 'restat':389 'result':722 'retent':344 'retriev':945,1535,1808 'return':533 'revenu':604,649 'review':16,42,1193,1198 'rewritten':1371 'risk':181,447,799,1144,1587,1618 'role':1296,1497 'root':1465 'rot':1523 'routin':420 'rule':358 'run':719 'saa':637 'safeti':458,1282,1411,1422,1442 'sampl':80,895 'samsung':1631 'say':1125 'scale':986 'scan':461,527 'schema':489,496,1357 'scope':203,436 'score':980,1588,1619 'search':572,971,1036 'second':78 'secret':1286 'section':97,324,340,345,648,688,755,1270 'secur':1283,1562 'see':1093,1153,1343 'select':962,994 'self':442,795,893,1045 'self-check':441,794 'self-consist':892 'self-reflect':1044 'selfcheckgpt':1651 'send':399 'sensit':454,544,1504 'sentenc':158,166,404 'separ':86,754,1271,1285 'sequenc':1106 'sequenti':924 'server':1739 'server-sid':1738 'set':519,1086,1098 'sever':1247,1368 'shadow':1632 'shape':145,1530 'shift':1541 'ship':725 'short':173,319 'shot':836,848,1000,1189 'side':1740 'simpl':162,1120 'simplest':231 'size':1315 'skew':1182 'skill' 'skill-prompt-engineering' 'small':740 'smell':1275,1512 'soften':471 'solut':1557 'someth':413 'sometim':117 'sonnet':1690 'sourc':518,528,562,568,1302 'source-codealive-ai' 'special':884 'specif':370,375,1122,1257,1335,1339,1346,1666,1846,1858 'specifi':146 'stage':1059 'start':407,1119,1595 'state':282,331,478 'step':76,79,170,183,868,870,930 'step-by-step':867 'stimulus':1002 'stop':1105 'strength':657 'string':502,505 'strong':473 'structur':58,83,106,153,200,480,579,591,618,786,1072,1268,1430,1501,1752,1839 'studi':1590 'style':813,860,1807 'subtask':925 'success':1260 'suffix':1577 'suggest':1250,1376 'summar':845,1010,1136 'switch':700 'symbol':1027 'system':30,804,1232,1418,1566 'tabl':191,585,1365 'tactic':1168 'tag':61,64,113,119,125,127,750,1273 'target':1224,1374,1682 'task':171,485,782,839,876,891,922,953,968,1028,1069,1082,1134,1258,1495,1703 'tax':1547 'taxonomi':1552,1584 'team':1644 'techniqu':7,47,133,749,815,821,831 'temperatur':1099,1821 'templat':1380,1793 'term':281,550 'test':1355,1637 'testabl':1331 'text':1021,1065,1719 'text.verbosity':1786 'think':1696,1825,1828 'thinking.budget':1710 'thought':138,780,865,934 'three':1451 'threshold':352 'throughout':115 'ticket':372 'time':543,746 'time-sensit':542 'tip':1118 'token':309,1546,1711,1717 'tone':666,812 'tool':356,361,801,959,965,1034,1291,1420,1645,1722,1761,1788 'top':1101 'top-p':1100 'topic':545 'topic-agent-safety' 'topic-agent-skills' 'topic-ai-coding' 'topic-ai-driven-development' 'topic-ai-safety' 'topic-antigravity' 'topic-bash' 'topic-claude-code' 'topic-codex-cli' 'topic-cursor' 'topic-developer-tools' 'topic-gemini-cli' 'tot':935 'trace':1032 'translat':844 'transpar':641 'tree':932 'trigger':21,1725,1732 'truth':1171 'truthfulqa':1649 'tune':729 'two':1058 'two-stag':1057 'u':1529 'u-shap':1528 'uncertain':291,991 'uncertainti':283,775,1303 'under-trigg':1723 'unground':467 'uniqu':1672 'univers':4,46 'unrequest':242 'unstat':463 'updat':397,401,423,431 'url':377 'usag':357,802 'use':11,62,111,121,186,560,608,663,667,752,832,1377,1454 'user':332,369,396,1396,1407 'user-fac':1406 'user-specif':368 'v':1629 'valid':232,394 'valu':237,640,1326 'variabl':707 'varianc':1508 'verbal':1050 'verbos':765 'verbosity/format/scope':732 'verif':1494,1748 'version':698 'via':854 'vision':1759 'visual':1067 'vs':1612,1728,1805,1827 'walk':1239 'waluigi':1165 'warn':1249 'web':534,540,557,791,1852 'web-deriv':556 'web-en':790 'without':590,841,1301 'won':573 'word/bullet':762 'workflow':931,1205,1215 'world':1622 'write':28,387 'x':430 'xhigh':1700 'xml':60,63,70,154,213,252,315,359,398,449,490,537,619,1272,1857 'y':432 'yes/no':163 'yoy':673 'zero':835 'zero-shot':834","prices":[{"id":"39655338-09a8-486d-8a1e-fc503cec5f15","listingId":"439abbf4-b003-42f1-8e07-294496d8b0a5","amountUsd":"0","unit":"free","nativeCurrency":null,"nativeAmount":null,"chain":null,"payTo":null,"paymentMethod":"skill-free","isPrimary":true,"details":{"org":"CodeAlive-AI","category":"ai-driven-development","install_from":"skills.sh"},"createdAt":"2026-05-04T06:56:23.611Z"}],"sources":[{"listingId":"439abbf4-b003-42f1-8e07-294496d8b0a5","source":"github","sourceId":"CodeAlive-AI/ai-driven-development/prompt-engineering","sourceUrl":"https://github.com/CodeAlive-AI/ai-driven-development/tree/main/skills/prompt-engineering","isPrimary":false,"firstSeenAt":"2026-05-04T06:56:23.611Z","lastSeenAt":"2026-05-18T18:57:07.245Z"}],"details":{"listingId":"439abbf4-b003-42f1-8e07-294496d8b0a5","quickStartSnippet":null,"exampleRequest":null,"exampleResponse":null,"schema":null,"openapiUrl":null,"agentsTxtUrl":null,"citations":[],"useCases":[],"bestFor":[],"notFor":[],"kindDetails":{"org":"CodeAlive-AI","slug":"prompt-engineering","github":{"repo":"CodeAlive-AI/ai-driven-development","stars":67,"topics":["agent-safety","agent-skills","ai-coding","ai-driven-development","ai-safety","antigravity","bash","claude-code","codex-cli","cursor","developer-tools","gemini-cli","hooks","mcp","multi-agent","opencode","plugins","prompt-engineering","skills","subagents"],"license":"mit","html_url":"https://github.com/CodeAlive-AI/ai-driven-development","pushed_at":"2026-05-12T20:04:46Z","description":"Practices, protocols, and skills for AI-driven software development. 18 skills + 1 Bash safety hook for Claude Code, Codex CLI, OpenCode, Cursor, Gemini CLI, Antigravity, and any agent supporting the Agent Skills standard.","skill_md_sha":"742151bd917e80e946da866e863428f2ca86a968","skill_md_path":"skills/prompt-engineering/SKILL.md","default_branch":"main","skill_tree_url":"https://github.com/CodeAlive-AI/ai-driven-development/tree/main/skills/prompt-engineering"},"layout":"multi","source":"github","category":"ai-driven-development","frontmatter":{"name":"prompt-engineering","description":"Universal prompt engineering techniques for any LLM. Use when crafting, optimizing, or reviewing prompts for AI models. Triggers on requests like \"improve this prompt\", \"write a system prompt\", \"optimize my instructions\", \"help me prompt engineer\", \"audit this prompt\", \"review my prompt\", or when building agentic systems that need structured prompts."},"skills_sh_url":"https://skills.sh/CodeAlive-AI/ai-driven-development/prompt-engineering"},"updatedAt":"2026-05-18T18:57:07.245Z"}}