Three quotes, different columns, missing freight, mixed tax treatment, and no clean way to compare.
Procurement skill
Package readyv0.1.0Vendor Quote Comparator
A ready-to-use AI Skill for comparing PDF, spreadsheet, and image-based vendor quotes. It standardizes the important fields, flags missing information, and keeps uncertain values visible instead of guessing.
Before / After
See the work this Skill removes
One normalized comparison with missing fields highlighted and follow-up questions surfaced.
The same quote, without the Skill vs with the Skill
Both runs used the same source material and Gemini Flash. The left is a direct AI summary; the right is the actual Vendor Quote Comparator output.
Source coverage
Lists each quote and its readability before comparison begins.
Scope alignment
Places comparable scope, specifications, and quantities together without forcing a match.
Missing information
Leaves unstated validity, freight, and warranty details as not provided.
Arithmetic checks
Recalculates subtotal, tax, and totals without treating calculations as source facts.
Follow-up
Turns unresolved items into questions ready to send to each vendor.
This testThe Skill found a TWD 200 arithmetic discrepancy in the quotation.
Want to test your own quotations?
Start a testA/B test: Oct 2, 2026 · Gemini Flash
Scope
Supported scope and limits
- Compare multiple vendor quotes in one run
- Normalize common prices, units, lead times, and line items
- Flag missing values and mismatched specifications
- Keep uncertain values explicit instead of inventing answers
- Produce a structured comparison ready for review
Included
Complete Skill package
SKILL.md
The full comparison workflow, decision rules, and conservative handling principles.
References
Output schema and normalization rules for repeatable results.
Examples
Sample inputs and expected outputs for quick testing and onboarding.
Quick Start
Setup guide, version information, license, and changelog.
FAQ
Before you buy
Will it overwrite my original file?
The default product principle is to preserve the original input and return a separate result. Exact behavior will be documented per Skill.
What if my file is unusual?
Supported formats are listed above. Unsupported or ambiguous inputs should be surfaced clearly instead of pretending the result is reliable.




