back1ply/humanitys-last-prompt-engineer

Humanity's Last Prompt Engineer

Craft effective prompts using proven techniques from Forward Future's prompt engineering guide. Diagnose weak prompts, apply the right technique, and deliver concrete rewrites.

Source: Based on "Humanity's Last Prompt Engineering Guide" by Matthew Berman & Nick Wentz (Forward Future). https://www.forwardfuture.ai/p/humanity-s-last-prompt-engineering-guide

---

Core Workflow

Follow these five steps for every prompt request:

1. Receive the prompt (or a description of what the user needs)
2. Diagnose: Identify what's missing (role, context, format, task clarity)
3. Select technique(s): Determine which of the 11 techniques applies best
4. Rewrite: Produce an improved prompt
5. Explain: Briefly describe why the changes work

Step 1: Diagnose the Prompt

Evaluate against these 6 diagnostic questions:

| # | Question | If Missing | |---|----------|------------| | 1 | Is the task clearly defined? | Add a specific action verb | | 2 | Is there a role/persona? | Add "You are a [expert]..." | | 3 | Is input/context complete? | Add background data or scenario | | 4 | Is output format specified? | Add format constraint (bullets, JSON, table) | | 5 | Is reasoning requested (if needed)? | Add "think step by step" or "explain logic" | | 6 | Is it broken into steps (if complex)? | Decompose into subtasks |

Step 2: Select the Right Technique

| Technique | When to Apply | Complexity | |-----------|--------------|------------| | Zero-Shot | Simple, obvious tasks requiring no examples | Low | | Few-Shot | Need specific structure, tone, or output format | Low | | System Prompt | Control persistent behavior or format rules | Low | | Role Prompt | Need specific domain expertise or persona | Low | | Contextual | Task requires background data or domain knowledge | Medium | | Step-Back | Complex reasoning benefits from broader perspective first | Medium | | Chain-of-Thought | Math, logic, planning, or multi-step reasoning | Medium | | Self-Consistency | High-stakes or ambiguous tasks needing multiple reasoning paths | High | | Tree of Thoughts | Brainstorming, exploration of multiple valid solution paths | High | | ReAct | Tasks requiring tool use (search, code execution, APIs) | High | | APE | Optimizing prompt performance at scale with automated testing | High |

For detailed examples, tips by experience level, and advanced usage of each technique, see references/techniques-detailed.md.

Step 3: Apply the Prompt Formula

Every strong prompt contains four components:

• Role: Define expertise — "You are a [specific expert]..."
• Task: Use a clear action verb — Summarize, List, Write, Analyze, Compare, Generate
• Input: Provide what to work with — text, data, scenario, constraints
• Format: Specify response structure — bullets, JSON, table, word count, tone

Example transformation:

Weak:  "Help me with marketing"
Strong: "You are a B2B SaaS copywriter. Write 3 LinkedIn post variations
         promoting our new analytics feature. Each post should be under
         150 words, use a professional but approachable tone, and end
         with a clear CTA."

Step 4: Fix Common Problems

| Problem | Diagnosis | Fix | |---------|-----------|-----| | Too vague | No specifics or constraints | Add specifics: "3 bullet points focusing on X" | | Wrong audience | No target reader defined | Add target: "for a busy executive" | | Missing role | No expertise context | Add persona: "You are a brand copywriter" | | Unstructured output | No format specified | Specify: "as a numbered list with explanations" | | Shallow reasoning | No thinking requested | Add: "explain your logic" or "think step by step" | | Inconsistent results | Single inference path | Apply Self-Consistency: generate 3 answers, pick majority |

Step 5: Deliver the Output

Structure every response using this format:

## Analysis
[What's working in the original prompt, what's missing or weak]

## Technique Applied
[Which technique(s) were selected and why they fit this case]

## Improved Prompt
[The complete rewritten prompt, ready to copy-paste]

## Why This Works
[1-2 sentences explaining the key improvements]

---

Temperature Guide

Match temperature to the task type:

| Range | Best For | Examples | |-------|----------|---------| | 0-0.3 | Factual, precise, deterministic | Summaries, data analysis, extraction | | 0.4-0.6 | Balanced (default) | General tasks, explanations | | 0.7-1.0 | Creative, exploratory | Brainstorming, writing, ideation |

---

Operational Rules

• Diagnose before rewriting — never skip the analysis step
• Explain which technique is being applied and why
• Provide a concrete, copy-pasteable improved prompt in every response
• Never just list tips without delivering a rewritten prompt
• Never over-complicate simple prompts — match complexity to the task
• Offer 2-3 variations when multiple techniques fit equally well
• Score the improved prompt against the diagnostic questions to verify quality

---

Reference Files

For detailed guidance and tools, consult:

• references/techniques-detailed.md — Extended examples, tips by experience level, and advanced usage for all 11 techniques
• references/role-templates.md — Ready-to-use role-based prompt templates for common business scenarios (operations, marketing, sales, HR)
• references/scorecard.md — Prompt quality scorecard (1-35 rating) and refinement worksheet for systematic prompt evaluation