v2.0 · 13 Chapters · 8 Appendices · 7 Skill Packs

AI Agent Cultivation
Field Manual v2.0

Don't just use AI. Cultivate it.
Train a generic chatbot into your domain engineering expert — 128 days of real-world battle testing.

Get the Manual → Free Preview: 5-Min Self-Assessment

Sound familiar?

If any of these hit home, this manual is for you.

🤦

Re-teaching every session

Every new conversation, you explain who you are, what you do, what font to use, what format you need. Again. And again.

📊

Excel breaks every time

AI-generated Excel files: column widths lost, merged cells broken, images vanished. You spend more time fixing than creating.

📄

PDF extraction is a nightmare

200-page tender document. 16 hours of manual extraction. One mistake and you start over.

📐

"AI can't do CAD"

Everyone says AI can't read drawings. What if you could teach it to read CAD content, identify components, calculate power, batch replace text, fill in numbers, modify line types and blocks — with high control rate, high stability, high recognition?

Real efficiency data

Not theoretical. Measured from actual engineering workflows in Macau.

Task Manual With cultivated AI Result
Word commissioning manual 2–3 days 30 minutes 10x+
Excel material submittal (28 sheets) 14 hours 30 minutes 28x
PDF tender extraction (200 pages) 16.7 hours 40 minutes 25x
CAD agent operation Manual only Read · Identify · Calculate · Batch replace · Modify blocks High control · Stable · Precise

What's inside

5 parts · 13 chapters · 8 appendices. Every chapter has real rules, real failures, real code.

Chapter 6

Word Document Processing

4 Iron Rules + 3 Painful Lessons + Full Case
  • Why docx-js breaks Word files
  • XML validation: 4 must-check items
  • Case: Commissioning manual (2 days → 30 min)
Chapter 7

Excel Table Processing

5 Iron Rules + 4 Painful Lessons + Full Case
  • Pure ZIP engine: why not openpyxl save
  • Material submittal 5-step pipeline
  • Case: 28 sheets, 14 hrs → 30 min
Chapter 8

PDF File Processing

3 Iron Rules + 3 Painful Lessons + Full Case
  • Why camelot needs JDK (use pdfplumber)
  • OCR parameter tuning
  • Case: 200-page tender, 16.7 hrs → 40 min
Chapter 9

CAD File Processing

5 Iron Rules + 6 Fatal Pitfalls + 3 Skills + 2 Cases
  • Agent reads CAD: content, components, power calculation
  • Batch replace text, fill numbers, modify line types & blocks
  • High control rate · high stability · high recognition
  • ezdxf vs COM: when to use which + DWG → DXF pipeline

Free preview: 5-minute self-assessment

Which level is your AI right now? Be honest.

The 8-Level AI Cultivation Model

This is Chapter 1 of the manual — free to read. Most people are stuck at Level 1–3. The manual shows you how to climb to Level 8.

  • Level 1
    Blank Slate Reads/writes files. Every new conversation = re-introduce yourself from scratch.
  • Level 2
    First Skill Can complete one specific task. Switch tasks = teach from zero again.
  • Level 3
    Document Capability Handles Word/Excel/PDF but makes dumb mistakes. You correct the same error repeatedly.
  • Level 4
    Red Lines & Rules Pitfalls have become permanent rules. Doesn't repeat known mistakes.
  • Level 5
    Systematic Organization Skills trigger automatically. No need to manually specify which tool to use.
  • Level 6
    Disaster & Rebirth Survived a catastrophic failure. Rules are now even stronger (e.g., some tools permanently banned).
  • Level 7
    Packaged Transformation Experience packaged into reusable skill packs. Cross-project deployment.
  • Level 8
    Autonomous Evolution AI proactively internalizes new regulations, new skills. You almost forget it's AI.

The manual teaches the complete path from Level 1 → Level 8, with real code, real failures, and real workflows.

7 installable skill packs

Plug and play. Each one solves a real engineering documentation problem.

1

material-approval

Batch submittal sheet generation

2

material-approval-workflow

Full pipeline + BQ page merge

→ GitHub repo (free code)
3

dem-material-approval

DEM-specific submittal format

4

commissioning-manual-generator

Auto-generate Word commissioning manuals

5

daily-report-generator

Construction daily reports

6

tender-intel-extractor

Extract data from tender PDFs

7

vba-reader

Read/modify/execute VBA macros

→ GitHub repo (free code)

Free code samples on GitHub

These repos are the real working code. The manual teaches you how to use them and build your own.

material-submittal-generator
Batch submittal sheet generation + BQ page merge. Battle-tested in MEP construction.
Python
excel-template-filler
Dual-engine (openpyxl + ZIP) batch template filling. Images & print settings preserved.
Python
gantt-chart-pro
Professional Gantt charts in Excel. No MS Project required. Multi-color, WBS, milestones.
Python
VBA-Macro-Reader
Read, modify & execute VBA macros from .xlsm/.xlam files. Dual-mode: oletools + win32com.
Python

Pricing

Start free. Upgrade when you're ready. One-time purchase — yours forever.

Free Trial — No Strings Attached

Get Chapter 1 (8-Level Self-Assessment) + 1 skill pack (material-approval).
Try the real code, real workflow. See if it works for you.

Read Chapter 1 Free → Get Skill Pack Free (GitHub) →

After trying, share your feedback (good or bad) → get 50% off the Complete Edition.

Starter

¥9.9 / MOP 12
  • HTML manual (3 essential chapters)
  • 3 core skill packs
  • Quick start guide
Email to Order
Best Value

Complete

¥49 / MOP 59
  • Full manual (HTML + PDF + EPUB)
  • All 13 chapters + 8 appendices
  • All 7 skill packs
  • Complete Python case studies
  • CAD chapter (DWG/DXF automation)
Email to Order

Professional

¥199 / MOP 239
  • Everything in Complete
  • 1-on-1 AI cultivation consultation
  • 1 custom skill pack (your task)
  • Priority support
Email to Order

Prices are public. Contact via email to purchase — payment details discussed privately.
Free updates within the same major version.

About the author

Mike — MEP Engineering Project Manager based in Macau. 128 days of real-world AI cultivation, documented from Level 1 blank slate to Level 8 autonomous evolution. Every rule, every pitfall, every efficiency gain — measured from actual construction site workflows.

GitHub Email