Trade Pipeline Skill

Trade Pipeline — Trade Documentation Automation Tool

English | 中文 | Русский | Technical Deep-Dive

📖 中文用户请点这里：完整中文文档 → README_CN.md —— 一份询价单，自动生成报价单 / PI / CI / 装箱单，改一处四单联动。

One set of data, auto-generates your Quotation, PI, CI, and Packing List.
No more copy-pasting between a dozen Excel files.

How to use it: an AI Skill

This is an AI Agent Skill (in SKILL.md format). Once installed, you just talk to it in plain language — "Turn this inquiry into a Quotation," "The customer confirmed, generate the PI/CI/PL" — and the AI handles the whole process. No commands to memorize. SKILL.md is a cross-agent standard — besides Claude Code and WorkBuddy, agents like Codex, Kimi Code, and DeepSeek Harness can load it too.

👉 Install and use it: with Claude Code · with Tencent WorkBuddy · other agents

Does this sound like your week?

• Customer sends an inquiry Excel → you manually copy product names, specs, and quantities into a Quotation
• Customer confirms → you manually copy the same data again to make a PI
• Goods are ready → you copy it all again to make the CI and Packing List
• A quantity typo in the PI → the CI, PL, and customs paperwork all inherit the error, and you have to hunt them down one by one
• Different customers need different PI formats (Russian customers want Russian, European customers want English) → you end up maintaining several template sets

Every document is essentially the same data in a different layout, but you rebuild it from scratch every single time.

Trade Pipeline's solution is simple: maintain a single order record, and generate every document from it automatically. Change one thing, and all four documents update together.

What it does for you

Automatically turns a customer's inquiry Excel into the full set of documents:

Quotation — fill in unit prices in the yellow-highlighted column (amounts calculate automatically); write them back to the order record with one click:

Proforma Invoice (PI) — seller/buyer header, grouped product rows, payment terms:

Commercial Invoice (CI) — bilingual Chinese/English, shipping marks, shipment info, SAY amount in words:

Packing List (PL) — grouped line items, carton count/weight/quantity, net/gross weight, pallet summary:

Who this is for

Best-fit profile

• Annual order volume: 50–500 orders
• Product SKUs: < 5,000
• Primary markets: Russia / Eastern Europe / South America / Middle East
• Team size: 1–3 trade sales staff

No coding background at all? That's fine. Once you've read the sections above, you'll know whether this is for you. The install and setup steps involve a bit of command-line work, but no programming knowledge is required — just copy and paste as you go.

How it relates to your ERP

Trade Pipeline is not a replacement for your ERP. ERPs handle inventory, finance, receivables/payables, and tax — this tool doesn't do any of that, and isn't meant to.

What it fills in is the front and back ends that ERPs tend to handle poorly:

What it is not

• ❌ Not an ERP — doesn't manage inventory, finance, or tax
• ❌ Not an online system — no accounts, no customer support portal; all data stays on your own computer
• ❌ Not a multi-user collaboration platform — best suited for teams of 1–3 people
• ❌ Not "AI doing your trade business for you" — the AI only helps parse messily formatted inquiries; everything else is deterministic, rule-based logic
• ❌ Does not connect directly to customs — it can prepare the data your customs broker needs, but it cannot submit directly to the Single Window system

To be upfront: what industry this is built for

This tool was built for my own company, so it's shaped around the characteristics of the hardware / fastener industry:

• Pricing is primarily weight-based (KG / metric ton)
• Product specs are relatively fixed (the DIN / ISO / GOST standard-parts world)
• Packaging conventions are fairly standardized

If you're also in hardware / fasteners (bolts, nuts, washers, standard parts, wire products, pipe fittings, flanges, bearings, seals, etc.), you can probably get it running just by updating your company info and the translation tables.

If you're in a different industry, your specs, pricing basis (per piece / per set / per meter), and weight/packaging logic will all differ, so this tool won't work out of the box — you'd need to adapt it, mainly in these three places:

The framework itself is solid, but how much adaptation work you need depends on how complex your industry is.

In short: hardware / fasteners can use it as-is; other industries should treat it as an adaptable foundation.

Industries that would take more work: apparel (complex size matrices), electronic components (nested BOMs), bulk commodities (futures-based pricing), cross-border e-commerce (tens of thousands of SKUs) — these require substantial rework, so go in with that expectation.

What's planned for the future

These are future directions — none of them exist in the current version yet. Items marked 🔜 are not coming immediately.

git clone https://github.com/Dangooy/trade-pipeline-skill.git
cd trade-pipeline-skill

# Create a virtual environment — don't skip this, see the note below
python3 -m venv .venv && source .venv/bin/activate    # Mac / Linux
# Windows PowerShell: py -m venv .venv; .\.venv\Scripts\Activate.ps1

pip install -e .

Initialize configuration                             → triggers trade-pipeline-init, walks you through your company info step by step
Turn examples/sample_inquiry.xlsx into a Quotation    → triggers trade-pipeline-run, runs the full pipeline automatically
The customer confirmed, write back prices and generate the PI/CI/PL   → automatically writes back prices + generates the formal documents

# 1. Download the project and install the engine
git clone https://github.com/Dangooy/trade-pipeline-skill.git
cd trade-pipeline-skill
py -m venv .venv; .\.venv\Scripts\Activate.ps1
pip install -e .

# 2. Copy both skills into the WorkBuddy skills directory
mkdir "$env:USERPROFILE\.workbuddy\skills" -Force
Copy-Item -Recurse -Force ".\.claude\skills\trade-pipeline-init" "$env:USERPROFILE\.workbuddy\skills\"
Copy-Item -Recurse -Force ".\.claude\skills\trade-pipeline-run"  "$env:USERPROFILE\.workbuddy\skills\"

# 1. Download the project and install the engine
git clone https://github.com/Dangooy/trade-pipeline-skill.git
cd trade-pipeline-skill
python3 -m venv .venv && source .venv/bin/activate
pip install -e .

# 2. Copy both skills into the WorkBuddy skills directory
mkdir -p ~/.workbuddy/skills
cp -r .claude/skills/trade-pipeline-init ~/.workbuddy/skills/
cp -r .claude/skills/trade-pipeline-run  ~/.workbuddy/skills/

@trade-pipeline-run Turn this inquiry into a Quotation

# Phase 1: generate the Quotation only (for price negotiation with the customer)
python -m trade_pipeline --input inquiry.xlsx --order 2601 --buyer global_fasteners --quote-only

# Phase 2: after customer confirmation, write back prices + generate PI / CI / PL
python -m trade_pipeline --price-update output/2601/2601_quotation.xlsx --model output/2601/2601_model.json

# Generate everything in one go (for rush orders or demos):
python -m trade_pipeline --input inquiry.xlsx --order 2601 --buyer global_fasteners

# New customer (use a placeholder identity for now, fill in details later):
python -m trade_pipeline --input inquiry.xlsx --order 2601 --buyer _new --quote-only

# When the system can't match a customer name, it stops and generates review.json — confirm, then continue:
python -m trade_pipeline --input inquiry.xlsx --order 2601 --confirm output/2601/2601_review.json

output/2601/
├── 2601_rfq.json           # Data extracted from the inquiry
├── 2601_model.json         # Order record (every document is generated from this)
├── 2601_quotation.xlsx     # Quotation
├── 2601_pi.xlsx            # Proforma Invoice
├── 2601_ci.xlsx            # Commercial Invoice
└── 2601_pl.xlsx            # Packing List

How this tool came to be

I'm not a professional programmer — I work in trade sales at a manufacturing factory, or more precisely, I'm what's called a "second-generation factory owner's kid." Given the limited resources available to me, I really needed a powerful assistant-type tool: an AGENT.

The reason this tool exists is simple: I was spending several hours a week manually producing trade documents, and got sick of it. Claude Code helped me turn years of accumulated business experience into a working automated system.

Here's an example: the first version of the price write-back feature worked by row number — the price on row 3 got written back to row 3. Technically correct. Until one day a customer inserted a few group-header rows into the Quotation, and when I ran the write-back — every price got shifted out of place, and the PI inherited the same errors.

I told Claude about the bug, and its first instinct was "exclude header rows during parsing." I said that wouldn't work — there's no predicting how a customer will edit an Excel file. We went back and forth for three rounds, and I finally suggested, "What if every row carried a hidden ID card?" Claude implemented the hidden UUID column approach.

This kind of design doesn't show up in programming tutorials, because it came from a real production bug.

AI coding tools are at their most powerful when you bring deep domain knowledge — I know exactly what a PI should look like, because I've made hundreds of them by hand.

Where your data lives

All customer information, order data, and generated documents stay on your own computer — nothing gets uploaded to any server.

This means:

• ✅ You have complete control over your data
• ✅ You can track every change with Git
• ⚠️ You're responsible for your own backups (recommend periodically copying to an external drive or encrypted cloud storage)

⚠️ One exception — optional LLM parsing (--use-llm): the default parser runs entirely locally, with no network calls. Only when you explicitly enable --use-llm does the tool send the raw inquiry contents (including customer names, products, and quantities) to the Anthropic cloud API to help parse unusual formats. This switch is off by default and requires an explicit opt-in — for sensitive or confidential documents, stay on the default local mode.

About the author

Built by Yang Tian (Ethan Yang) — factory-side trade practitioner, not a professional programmer. More trade AI skills: tianji-skills (B2B lead research, RFQ intake, document verification). Public profile: @Dangooy.

License

MIT — free to use, free to modify.

Curious what changed in each release?
Release notes (full changelog in CHANGELOG.md)

Ran into a problem?
File an Issue (no technical background needed — just describe what you saw and what you expected)