Pro-Prompt — Local LLM Prompt Engineering & Manifest Generator

Generate structured, expert-level instruction manifests from any task description using local LLMs via Ollama — no cloud API keys required.

Pro-Prompt is a CLI tool and interactive launcher that transforms task descriptions into comprehensive, reproducible instruction manifests. It applies 173 prompt engineering techniques across 15 categories (Chain-of-Thought, Tree-of-Thought, ReAct, MECE decomposition, red teaming, and more) to force exhaustive, high-quality outputs from any Ollama-compatible model. Supports single-model streaming, dual-model parallel generation with split-screen display, and expert synthesis that merges two outputs into a superior unified document.

---

Key Features

Feature	Description
Interactive launcher	Numbered menu, no CLI flags to memorize
173 prompt engineering techniques	Organized in 15 categories with anti-patterns and quick-reference matrix
Single model generation	Real-time token streaming in terminal
Parallel dual-model generation	Split-screen display with two columns, live tokens
Expert synthesis	Merges two manifests into a unified, superior document with streaming
Full pipeline	Parallel generation + synthesis in one command
Web enrichment	Automatic DuckDuckGo search for real-world context injection
Model auto-detection	Lists locally installed Ollama models with numbered picker
Auto-install	Installs Ollama and Python automatically on macOS, Linux, and Windows
Auto-pull	Downloads missing models on demand via ollama pull
Session memory	Tracks past runs for cross-session coherence
Persistent settings	Models, techniques, temperature saved locally

---

Quick Start

Install

macOS / Linux:

git clone https://github.com/TFD-42/Pro-Prompt.git
cd Pro-Prompt
chmod +x install.sh
./install.sh

Windows (PowerShell):

git clone https://github.com/TFD-42/Pro-Prompt.git
cd Pro-Prompt
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
.\install.ps1

The installer handles everything: Ollama, Python 3, virtual environment, and dependencies.

Run

source .venv/bin/activate    # Linux/macOS
# .\.venv\Scripts\Activate.ps1  # Windows

python3 prompt_expert_enhence.py

Launches the interactive menu. No arguments needed.

CLI Mode

Pass arguments directly for scripting and automation:

# Single model generation
python3 prompt_expert_enhence.py generate "Design a REST API" --model llama3:latest

# Parallel dual-model generation
python3 prompt_expert_enhence.py parallel "Design a REST API" --model-a llama3 --model-b qwen2.5:7b

# Full pipeline (parallel + synthesis)
python3 prompt_expert_enhence.py full "Design a REST API" --techniques "1-30"

# List all 173 techniques grouped by category
python3 prompt_expert_enhence.py generate x --list-techniques

---

Interactive Menu

==============================================================
   MANIFEST GENERATOR  -  Prompt Expert Launcher
==============================================================
   Model A     : llama3:latest
   Model B     : qwen2.5:7b
   Synthesis   : qwen2.5:7b
   Temperature : 0.3
   Techniques  : 15 active / 173 available
   Internet    : ON   Web enrichment : ON   Streaming : ON
--------------------------------------------------------------

  1.  Single generation         (1 model, streaming)
  2.  Parallel generation       (2 models, split screen)
  3.  Full pipeline             (parallel + synthesis)
  4.  Synthesize 2 files

  5.  Configure models
  6.  Configure techniques
  7.  Browse available techniques
  8.  Advanced settings          (temperature, timeout, url, web, stream)

  9.  View memory
  10. Clear memory

  0.  Quit

Model Picker

When selecting a model, Pro-Prompt lists all locally installed models:

  -- Generation model --
  Locally installed models:
      1. llama3:latest                   4.7GB  2026-05-20  <-- current
      2. qwen2.5:7b                      4.4GB  2026-05-18
      3. dolphin3:latest                 4.6GB  2026-05-07
      4. [Enter a name manually / pull a new model]

  Choice [llama3:latest] >

Type a number to select, a model name to pull, or Enter to keep the current one.

---

Prompt Engineering Techniques

Pro-Prompt ships with 173 techniques across 15 categories in prompte_expert_methodologie.json, plus 8 anti-patterns and a quick-reference matrix for task-based technique selection.

Categories

#	Category	Techniques	Examples
1	Framing	6	Zero-shot, few-shot, many-shot, negative-shot, contrastive prompting
2	Directed reasoning	10	Chain-of-Thought (CoT), Tree-of-Thought (ToT), Graph-of-Thought (GoT), ReAct, Program-of-Thought (PoT), Skeleton-of-Thought, least-to-most
3	Depth forcing	11	Output length specification, recursive deepening, exhaustive enumeration, anti-lazy preamble
4	Constraint-based	21	Format forcing, vocabulary constraint, register constraint, perspective constraint, inverse prompting, rubber duck, constraint stacking
5	Multi-perspective	9	Multi-viewpoint analysis, counter-arguments, audience layering, cross-disciplinary
6	Meta / recursive	15	Self-critique, self-refine, self-ask, meta-prompting, constitutional prompting, recursive summarization
7	Structural	13	Instruction decomposition, strong delimiters, priority stacking, prompt chaining, conditional prompting
8	Persona & role	9	Expert persona, multi-persona debate, naive persona, devil's advocate, future historian
9	Emergent	14	Emotional priming, anchoring, semantic pressure, cognitive load offloading, counterfactual, steelmanning, pre-mortem
10	Cognitive decomposition	9	MECE, first principles, five whys, abstraction ladder, dual process, Socratic decomposition, ontology extraction
11	Adversarial	9	Red teaming, stress testing, bias hunting, assumption mapping
12	Hybrid multi-pass	8	Generate-then-filter, breadth-first/depth-first, adversarial refinement loop, perspective rotation, zoom protocol
13	Evidence & justification	16	Citation thresholds, uncertainty quantification, historical grounding, tiered evidence
14	Creative & narrative	8	Analogy generation, narrative embedding, timeline construction, forced self-interruption
15	Rarely explored	15	Formal logic coherence, invariant detection, test generation, tacit knowledge elicitation, weak signal detection, second-order effects, heuristic generation

Default Set (15 techniques)

Step-by-step reasoning, forced reframing, anti-lazy preamble, recursive deepening, counter-arguments, example-driven expansion, outline-then-expand, definition-first, first-principles, no-word-limit, Tree-of-Thought, constraint stacking, constitutional prompting, MECE decomposition, assumption mapping.

Selecting Techniques

--techniques "1,5,8,10,25"     # Specific IDs
--techniques "1-30"             # Range
--techniques "1-173"            # All 173 techniques

In the interactive menu, use option 6 to configure or option 7 to browse (grouped by category with anti-patterns and quick reference).

---

Web Enrichment

When internet is available, Pro-Prompt automatically searches DuckDuckGo for the task description, fetches top results, and injects relevant context into the prompt. This runs before generation and adds real-world grounding without any API keys.

Toggle via the advanced settings menu (option 8) or --no-web flag.

---

Output Structure

Generated manifests follow a 12-section structure:

1. Title & Executive Summary
2. Final Objective & Success Definition
3. Execution Context & Prerequisites
4. Ambiguity Zones to Resolve
5. Step Decomposition (Detailed Pipeline)
6. Control Loops & Scoring
7. Persistent Artifacts to Maintain
8. Constraints & Guardrails
9. Error Handling Strategy
10. Final Deliverable & Output Format
11. Reproducibility Checklist
12. Notes for the Target Agent

---

Project Structure

Pro-Prompt/
  prompt_expert_enhence.py          # Main application (~1700 lines)
  prompte_expert_methodologie.json  # 173 prompt engineering techniques (15 categories)
  requirements.txt                  # Python dependencies (requests)
  install.sh                        # Installer for macOS/Linux
  install.ps1                       # Installer for Windows
  .gitignore
  README.md
  LICENSE
  memory/                           # Session history (gitignored)
  outputs/                          # Generated manifests (gitignored)

Requirements

• Ollama — installed automatically by the installer
• Python 3.8+ — installed automatically by the installer
• requests — installed via pip install -r requirements.txt
• At least one Ollama model pulled (the launcher handles this interactively)

Configuration

All settings persist in settings.json (gitignored, local to each user):

Setting	Default	Description
model_a	llama3:latest	Primary generation model
model_b	qwen2.5:7b	Secondary model for parallel runs
synthesis_model	qwen2.5:7b	Model used for expert synthesis
temperature	0.3	LLM temperature (0.0–1.0)
timeout	600	Seconds per Ollama call
techniques	[1,5,8,10,12,14,18,25,40,47,108,121,125,147,153]	Active technique IDs (from 173 available)
use_web	true	Enable web enrichment
stream	true	Enable real-time streaming

How It Works

1. You describe a task — in natural language, as simple or complex as you want
2. Pro-Prompt builds an expert prompt — injecting selected techniques, web context, and session memory
3. Local LLM generates a manifest — a structured 12-section document describing the task with methodological precision
4. Optionally, two models generate in parallel — and a synthesis pass merges them into a superior unified document
5. The output is a reproducible instruction set — ready to be executed by any LLM agent (Claude, GPT, Gemini, Llama, Mistral, Qwen)

Use Cases

• Prompt engineering — Generate expert-level prompts for any LLM task
• Task specification — Create detailed, unambiguous task descriptions for AI agents
• Knowledge extraction — Force exhaustive exploration of any topic
• Comparative analysis — Run two models in parallel and synthesize the best of both
• Reproducible AI workflows — Manifests can be reused across models and platforms
• Learning prompt engineering — Browse 173 techniques with descriptions and categories

License

MIT

Contributing

1. Fork the repository
2. Create a feature branch
3. Run python3 -m py_compile prompt_expert_enhence.py before committing
4. Open a pull request

Keep settings.json, outputs/, and memory/sessions.json out of commits (they are gitignored).