Egypt Mentor Agent

English | Arabic

Open-source prompts, skills, and evaluation assets for building a mentor-style AI agent focused on Egyptian users.

A disciplined prompt pack for builders who want better diagnosis, tighter prioritization, and more honest follow-up than a generic advice bot can provide.

---

🎯 Why This Exists

Most AI "coach" systems fail in the same way:

• they jump into advice too early
• they ask broad, repetitive questions
• they open too many paths at once
• they confuse motivation with strategy
• they generate plans that sound good but collapse in real life

egypt-mentor-agent is built to do the opposite.

It is designed to help an AI system answer questions like:

• What should this user focus on now?
• What is the actual bottleneck?
• What should happen next given time, money, and life constraints?
• What should the agent do when the user comes back with weak execution or changed reality?

The project is Egypt-first by default, but structured so the mentoring brain can be reused across different hosts and products.

✨ What Makes It Different

Typical AI advice bot	Egypt Mentor Agent
Gives advice before understanding the case	Builds a case map before serious planning
Offers many paths at once	Defaults to one real priority
Uses generic motivation language	Uses evidence, bottlenecks, and constraints
Treats planning as success	Treats execution evidence as success
Restarts the conversation every time	Preserves continuity when the host supports memory
Sounds helpful but stays vague	Produces reviewable next steps

🧠 What This Repo Is

This repository is a host-agnostic prompt pack for a long-term mentor-style agent.

It is meant to be reused across:

• file-aware chat tools
• editor agents
• custom GPT-style environments
• local AI wrappers
• future runtime integrations

It currently includes:

• a bootstrap entrypoint
• a system prompt
• a response contract
• modular core skills
• modular domain skills
• governance guardrails
• realistic examples
• evaluation rubrics and dry runs
• builder-facing architecture notes

🗃️ Repository Map

Path	Purpose
START-HERE.md	Bootstrap entrypoint for hosts that support file loading or mentions
prompts/system.md	Core mentoring behavior and operating rules
prompts/response-contract.md	Internal quality contract for strong responses
skills/core/	Reusable cross-domain decision and execution skills
skills/domains/	Egypt-aware domain lenses such as real estate, fintech, ecommerce, and more
governance/	Safety, memory, review, and long-term quality controls
manifests/	Load order and host capability references
examples/	Realistic worked cases that show intended behavior
evaluations/	Checklists, scoring rubrics, and dry runs
architecture/	Builder-facing structural references
CONTRIBUTING.md	Rules for extending the repo without breaking its philosophy

🔄 How It Works

flowchart TD
    A[START-HERE.md] --> B[manifests/load-order.md]
    B --> C[prompts/system.md]
    B --> D[prompts/response-contract.md]
    B --> E[skills/core]
    B --> F[skills/domains]
    B --> G[governance]
    B --> H[architecture]
    E --> I[Diagnosis, prioritization, execution, review]
    F --> J[Real-estate, fintech, ecommerce, education, and more]
    G --> K[Safety, memory, continuity, quality]
    I --> L[Examples and evaluations]
    J --> L
    K --> L

Loading

🚀 Quick Start

If your host supports file mentions or file loading

1. Start from START-HERE.md
2. Follow manifests/load-order.md
3. Load only the domain files relevant to the current case

If your host does not support file mentions

1. Manually load START-HERE.md into the host context
2. Then follow manifests/load-order.md
3. Preserve the same structure and behavior rules

If you want to evaluate the system quickly

1. Read examples/founder-no-revenue.md
2. Read evaluations/dry-runs/founder-no-revenue-dry-run.md
3. Review evaluations/checklist.md
4. Review evaluations/scoring-rubric.md

🧭 Core Design Principles

• Egypt-first by default
• diagnosis before planning
• one main priority by default
• current-phase plans over giant roadmaps
• execution evidence over intention
• follow-up that reacts to reality, not wishful thinking
• actor-aware domain reasoning
• host-agnostic packaging

📦 Current Coverage

Core capabilities

Area	Coverage
Diagnosis	intake, guided questioning, case mapping
Decision quality	validation, monetization, prioritization
Execution	planning, focus, review, replanning
Human layer	behavior, friction, avoidance, continuity
Governance	safety, memory, review loops, update policy

Domain lenses

Domain	Primary lens
Real estate	actor clarity, workflow value, monetization path
Fintech	trust, compliance friction, financial workflow value
Ecommerce	conversion, fulfillment, margin, repeat behavior
Education	learner vs buyer, attendance, outcomes, monetization
Career & work	employability, proof of capability, realistic path selection
Creator & media	attention vs trust vs monetizable value
Local services	inquiry-to-booking flow, quality, referrals, repeat
Healthcare	safety-first separation between operations and medical judgment
Tech	lane selection and Egypt-fit entry paths

🧪 Quality Standard

A strong implementation of this repo should consistently produce responses that:

• reduce confusion
• identify the real bottleneck
• fit the user's actual constraints
• avoid spreading effort across many tracks
• create a reviewable next step
• become smarter over time through continuity and evidence

This repo already includes:

• realistic examples in examples/
• scoring assets in evaluations/
• a worked dry run in evaluations/dry-runs/

🚫 What This Repo Does Not Include

This repository does not currently ship with:

• a bundled runtime
• a UI
• a database or persistence layer
• a memory backend
• a hosted API
• a provider-specific integration

That is intentional. The goal is to keep the mentoring brain reusable across different environments.

🛠️ For Builders

Integrating the repo into a product

Start here:

• START-HERE.md
• manifests/load-order.md
• prompts/system.md
• prompts/response-contract.md
• architecture/agent-spec.md

Extending the repo itself

Start here:

• CONTRIBUTING.md
• examples/README.md
• evaluations/README.md

🤝 Contributing

Contributions are welcome, but this project should stay disciplined.

Before changing behavior, read:

• CONTRIBUTING.md

For meaningful changes, prefer updating at least one of:

• an example
• an evaluation artifact

That keeps the project grounded in real behavior instead of theory alone.

📄 License

MIT License. See LICENSE.