The Procurement OS for Codex and ChatGPT: Nine AI Skills for the Full Source-to-Contract Lifecycle

The Molecule One Procurement OS is a free set of nine production-ready procurement skills covering source-to-contract, from spend analysis through negotiation, contracts, supplier management and ESG. We first released it as a Claude plugin. Many of the procurement teams we work with run on ChatGPT and Codex instead, so we have adapted the same skills for OpenAI platforms.

There are two editions in one download. The Codex edition installs the nine skills into a project's .agents/skills/ folder, with saved settings, run history and the bundled Python helpers. The ChatGPT edition gives you nine Custom GPT drafts (instructions plus knowledge files) for eligible Business, Enterprise and Edu workspaces. The methodology, templates and guardrails are the same as the Claude edition.

This article covers what each skill does, how they chain together into a coherent procurement workflow, and how to install and run them in Codex or ChatGPT. If you have already read The Codex Playbook for Procurement Teams, think of this as the working operating system that the playbook described in the abstract. If you want to understand how to implement AI in procurement more broadly, start there.

The rest of this document is structured for two reading paths. If you want the why, read sections 1 through 3. If you want the how, jump to sections 4 through 12 and use the sidebar to navigate to whichever skill you need today. If you just want to install it, sections 15 and 16 cover setup and the download.

Most procurement teams using AI today are using it the way you would use a calculator: pull it out for a discrete task, get the answer, put it away. McKinsey research consistently shows that the procurement functions delivering the most value from AI are the ones treating it as infrastructure, not as a tool. Someone uses ChatGPT to summarise a contract. Someone else uses a different tool to draft an RFP cover letter. The category lead has a private prompt library for spend questions that nobody else knows exists. None of it is connected. None of it is repeatable. Each prompt starts from zero.

This is not nothing. It is faster than not using AI at all. But it is structurally limited in three ways:

The result is what we have been calling the procurement-AI fragmentation problem. Teams have access to capable AI, but the work product they get out of it is uneven, inconsistent, and difficult to defend. The technology is doing well. The operating model around it is doing badly.

When we started building AI workflows for procurement clients, we quickly learned that the right unit of procurement AI is not the prompt. It is the skill. A skill is a structured, reusable workflow with its own configuration, its own templates, its own guardrails, and its own outputs. You install it once and use it forever. The next person on your team uses the same one. The methodology is shared, the memory is persistent, and the outputs are consistent.

The Procurement OS is nine of those skills, designed and tested as a set, packaged as a single plugin. We built these skills originally for client engagements. We are releasing them publicly because the procurement community deserves access to the same methodology we use with our consulting clients.

What an OS gives you that a stack of prompts doesn't

This is the same shift you would make if you were going from spreadsheets-as-data-pipelines to a real ETL tool. The work is the same. The infrastructure around the work changes. And once that infrastructure exists, you stop spending energy on the parts of the workflow that don't deserve any.

The nine skills cover the Source-to-Contract lifecycle in the order you would normally run it: from understanding spend, through category planning and sourcing, to negotiation, contracts, supplier risk and sustainability.

Each skill is built to stand alone. You can run the Spend Analyzer on its own without ever touching the others. You can use the Negotiation Playbook Generator the night before a renewal call without ever having run a single line of spend through the Spend Analyzer. They are independently useful.

But the design intent is that you also run them in sequence when the work calls for it. The output of Spend Analyzer becomes the input of Category Strategy Builder. The category strategy informs the RFP. The RFP defines the criteria the Response Evaluator scores against. The selected supplier feeds the Scorecard Engine. The supplier's performance, plus the original commercial terms, feed the Negotiation Playbook for the renewal. End to end.

The three newer skills plug into that chain. The Contract Lifecycle Manager reviews the paper that comes out of a negotiation and tracks it after signature. Supplier Management and the ESG & Sustainability Assessor sit alongside the Scorecard Engine and cover supplier risk and sustainability.

At a glance

What is new in v1.1

The Codex and ChatGPT edition starts at version 1.1, the same release as the Claude plugin. It includes all nine skills, including the three added in v1.1: the Contract Lifecycle Manager, Supplier Management, and the ESG & Sustainability Assessor. It is released under the Apache License 2.0, so you can run, modify and build on it inside your organisation.

Beyond that, we are working on the full Source-to-Pay Procurement OS. It will add invoice matching, payment optimisation, and compliance monitoring to these nine skills, covering the complete procurement cycle from spend analysis through to payment. Follow us on LinkedIn for release updates.

Spend Analyzer is usually the first skill teams reach for, because spend visibility is usually the first thing a procurement function lacks. Most organisations have spend data sitting somewhere, like an ERP export, a P-card report, or an invoice register, but the data is messy, the category structure is informal, and pulling a clean "where are we spending" picture out of it takes a week of analyst time.

The Spend Analyzer skill takes whatever export you have, parses it, normalises supplier names, applies a category taxonomy (yours if you have one, a sensible default if you don't), flags maverick and tail spend, and surfaces ranked savings opportunities. The output is a categorised spend cube in Excel and an executive summary in Word.

Category strategy is the work that procurement leaders consistently say they don't have time for. It's the deliberate, deliberate planning of how a category will be sourced, segmented, and managed over a horizon of 12-24 months. Most organisations skip it. The ones that do it well usually do it with the help of an external consultant, at significant cost.

The Category Strategy Builder skill brings that work in-house. Give it a category name, a one-line objective, and any spend data you have for the category (an export, a contract list, a supplier roster, whatever is to hand). It returns a Kraljic-segmented view, supply market intelligence, three to five savings levers ranked by effort and impact, and a sourcing roadmap. The more spend data you feed in, the more the recommendations are grounded in what your organisation is actually buying instead of generic category benchmarks. You can iterate the output, push back on the recommendations, and ship a defensible category plan in an afternoon.

Writing an RFP from a blank page is one of the most over-engineered tasks in procurement. Most teams have a template that's three years old, written by someone who's left the company, that nobody fully understands. The RFP that actually goes out the door is a Frankenstein of that template, copy-pasted clauses from the last three RFPs, and whatever the requesting business unit wrote in an email.

The RFP Generator skill replaces that with a structured pipeline: feed it your source material (SOW, contract, brief, meeting notes, or just a description) and it returns an issuable RFP, RFI, or RFQ with all the standard sections, plus a supplier cover letter, plus an evaluation scoring template aligned to what you're asking. You decide the type (RFP / RFI / RFQ); the skill handles the structure.

Once the RFP is back in, the work that follows is the work nobody wants to do. Twenty pages per supplier, multiplied by three to five suppliers, in different document formats, with answers in different orders, scattered across attachments. Comparing them line by line is the kind of work that takes a senior analyst three days and still produces a comparison everyone disagrees with.

The RFP Response Evaluator skill compresses that to a few hours. Drop the supplier responses into a folder, give the skill the original RFP, and it produces section-by-section scoring, a side-by-side comparison highlighting strengths and gaps, a ranked recommendation, and an Excel scoring grid with RAG colours. You can override weightings, ask follow-up questions about specific responses, and generate a recommendation memo that survives stakeholder scrutiny.

Once a supplier is on the books, performance management is supposed to be the disciplined backbone of supplier relationships. In practice, it tends to be ad hoc: a quarterly review meeting where everyone shows up with their own numbers, no shared scorecard, and an action list that gets forgotten by the next quarter.

The Supplier Scorecard Engine skill builds the scorecard, runs it, and reports on it. You give it your SLA and KPI data (or describe it verbally if your data is informal) and it produces a scorecard with RAG ratings, trend analysis if historical data is provided, an improvement plan with owners and milestones, and a benchmark comparison against peers in the same category. The skill is designed to be run quarterly so that performance management stops being something everyone improvises in the meeting.

Negotiation is the highest-leverage hour of work in procurement. It is also the most under-prepared. The pattern most teams fall into: you do the spend analysis, build the strategy, run the RFP, evaluate responses, run the scorecard… and then walk into the renewal meeting with a one-page summary the analyst threw together the night before. The leverage you built up in the prior six months evaporates because nobody had time to convert it into a structured playbook.

The Negotiation Playbook Generator skill exists to close that gap. You feed it the supplier proposal, your target position, the contract terms, and any market intel you have. It returns a structured playbook: a quantified BATNA, a sequenced set of negotiation levers ordered by impact, pre-drafted counter-proposal language for the contentious clauses, and a one-page briefing card you can take into the meeting itself. Used well, the skill changes the meeting itself: instead of reacting to what the supplier puts on the table, you walk in with the levers, the numbers, and the language already prepared, and you run the conversation.

A signed contract is where negotiated value either holds or leaks. Notice periods get missed, auto-renewals roll over, and obligations sit in a PDF nobody reopens. The Contract Lifecycle Manager reviews a supplier contract clause by clause against your standard positions, pulls out the obligations and key dates, and helps you decide what to do at renewal.

It works from the buyer's side. Every clause gets a Green, Yellow or Red rating, and the items that need changes come with a ready-to-use redline, the reasoning behind it, and a fallback position, grouped by negotiation tier. It supports procurement and legal workflows, but it does not give legal advice. Qualified counsel should review Red items before anything is signed.

Supplier problems rarely announce themselves. A single-source dependency, a weak financial position, or a vendor handling employee data without the right checks usually surfaces only after something goes wrong. Supplier Management gives you a consistent way to check suppliers on the way in and to watch the portfolio afterwards.

For a new supplier, you choose the depth of due diligence (Light, Standard or Enhanced) and get a checklist and an approve, approve-with-conditions or reject decision. For existing suppliers, it builds a risk register scored by severity and likelihood, tests concentration and single-source exposure from a spend list, and tiers suppliers so you know where relationship effort belongs. It does not run credit, sanctions or adverse-media checks itself. It tells you which to run and records the results you provide.

Sustainability questions now land on procurement's desk. Customers send ESG questionnaires, regulators ask about supply-chain due diligence, and finance wants a view of Scope 3 emissions. Most teams answer each one with a spreadsheet built from scratch.

The ESG & Sustainability Assessor screens suppliers by risk, scores them on the evidence you provide, escalates red flags, and builds tailored questionnaires and improvement plans. From a spend file, it produces a spend-based Scope 3 screen using your emission factors and ranks the hotspots. It also maps the rules you tell it apply to you onto procurement tasks, tender criteria and contract clauses. It does not supply emission factors or decide which laws apply to you: you provide the factors, and Legal confirms applicability. Spend-based results are screening estimates, not a reportable inventory.

To show how the skills chain together, here is one scenario we have worked through in the field. Your AWS contract is up for renewal in six months. The CFO wants 15% off. You have a year of cloud invoices and a draft renewal proposal from AWS sitting on your desk.

If you have already read The Codex Playbook for Procurement Teams, the Procurement OS will feel like the natural next step. The playbook makes the case that procurement teams should build skills, automate analyst-time work, and govern AI use carefully. The Procurement OS is nine of those skills, pre-built, ready to install.

If you haven't read the playbook, you can still use the Procurement OS on its own. The playbook gives you the wider context: how Codex works for non-developers, how to write AGENTS.md, how to budget credits, how to roll it out across a team, and how to plan a 30/60/90-day adoption arc.

On Claude instead? The same nine skills are available as the Procurement OS Claude plugin.

Which edition should I use?

The download has two ZIP files, m1-procurement-os-codex-v1.1.0.zip and m1-procurement-os-chatgpt-v1.1.0.zip, plus a PDF user guide. Use the edition that matches where your team works.

In Codex

1. Open the project where you will do procurement work. Create a .agents/skills/ folder in it if there isn't one.
2. Copy each of the nine folders from m1-procurement-os-codex/skills/ into .agents/skills/. Keep each folder's contents together.
3. Copy AGENTS.md to the project root. If the project already has one, merge the Procurement OS section into it rather than replacing it.
4. Start a new Codex chat in that project. If the skills don't appear, restart Codex or reopen the project.
5. Invoke a skill by name, for example $spend-analyzer, or describe the task and let Codex choose.

Python 3 is needed for the bundled helper scripts. Word, Excel and PDF outputs depend on the tools available in your Codex environment.

In ChatGPT

1. In an eligible Business, Enterprise or Edu workspace, open Explore GPTs, then Create.
2. Create one GPT per workflow you need. Paste the matching instructions/<skill>.txt into Instructions.
3. Upload knowledge/<skill>-reference.txt (and the helper reference, where there is one) as Knowledge.
4. Turn on Code Interpreter & Data Analysis for spend, scoring and scorecard work.
5. Test in Preview with sample data, then save privately or share within your workspace.

New GPT creation is currently limited to eligible Business, Enterprise and Edu workspaces, and OpenAI has announced plans to retire Custom GPTs. Check your workspace before choosing this route. The ChatGPT drafts cannot run the Python helpers or keep run history; paste a short organisation profile at the start of each chat instead.

First run setup

The first time you use a skill in Codex, it asks a few short setup questions: your organisation name, primary currency, ERP system, financial year start month, and spend taxonomy if you have one. Answers are saved in the project's .procurement-os/ folder and reused on later runs. To change them, say so in the chat: "Update my settings, we've moved to NetSuite."

The full user guide

The download includes a PDF user guide covering installation for both editions, privacy and data handling, which workflow to use, first prompts, and a review checklist. Each ZIP also contains the same guide as INSTALLATION_AND_USAGE.md.

The Procurement OS v1.1 for Codex and ChatGPT is available now. The download includes both editions and the PDF user guide. Enter your work email to access all three.

What happens next

1. You install it. About ten minutes in Codex; a few minutes per GPT in ChatGPT.
2. You run your first skill. Not sure where to start? Take our AI Readiness Assessment first. Otherwise, we recommend starting with $spend-analyzer if you have spend data on hand, or $negotiation-playbook-generator if you have a renewal coming up. Both produce a usable artifact in under an hour.
3. You tell us what worked. We are continuously improving the skills based on real procurement use. Email hello@moleculeone.ai with what worked, what didn't, and what you wish was there.

Need help with rollout?

If you want help installing, configuring, or rolling the Procurement OS out across a procurement team, our AI procurement consulting team offers implementation support: installation, custom skill development, and governance setup. We also run AI training for procurement teams: hands-on workshops that get your analysts productive with the OS in days, not weeks. Email hello@moleculeone.ai or visit moleculeone.ai/contact.