Skip to content

Repository files navigation

Procurement Skills for Claude

An open-source pack of Claude Agent Skills for procurement professionals. Anthropic shipped agents for finance. This is skills for procurement.

License: MIT

What's in the pack

Five skills, each focused on one part of the source-to-pay lifecycle. Drop the folder into Claude's skill directory and Claude becomes a procurement specialist with your house playbook, your taxonomy, your scoring weights, and your tone.

Skill What it does When it fires
redline-sentry Contract redlining (NDAs, MSAs, DPAs, SaaS) with risk rating, suggested redlines, negotiation cheat sheet "review this contract", "is this NDA risky", "redline this MSA"
spend-prism Spend analysis from Coupa/Ariba/SAP/Oracle/Concur exports — categories, anomalies, tail spend, savings opportunities "analyse this spend", "find anomalies", "categorise these suppliers"
bid-compass Guided RFP generation and evaluation with configurable scoring and mandatory quality gates "draft an RFP", "evaluate these bids", "scoring matrix"
supplier-truthcheck Vendor master data validation — IBAN mod-97, VAT (VIES/HMRC/BZSt), company registers, sanctions (OFAC/EU/UK/UN), PEP "validate these suppliers", "check this IBAN", "sanctions screening"
procure-voice Tone overlay — honest, plain-English procurement writing that strips jargon and leads with the point applied automatically by the other skills; or "make this less corporate"

Every skill is configurable via its own config.yaml. You edit your playbook, your weights, your taxonomy — never the skill logic.

Why this exists

On 5 May 2026, Anthropic announced 10 agent templates for financial services — pitchbooks, KYC, month-end close, the works. Finance got the headline.

Procurement is the next-largest spend category in most enterprises and arguably has more day-to-day operational pain than finance. There's no equivalent pack for procurement. So here's one — not as agents (those need orchestration, connectors, subagents), but as skills: portable, transparent, configurable Markdown files that any Claude user can fork and adapt.

If you have Claude.ai Pro/Team or Claude Code, you can install these in 5 minutes and use them today.

Quick install

These skills work in three places. Pick the one you use most — full details in INSTALL.md.

Most users: Claude.ai web / Claude Desktop / mobile / Cowork

git clone https://github.com/YOUR_USERNAME/procurement-skills.git
cd procurement-skills
./install.sh --package

This produces 5 upload-ready zips in dist/. Then in Claude.ai: Settings → Capabilities → Skills → Upload skill. Upload each zip. Once uploaded, the skills appear in the slash menu (/redline-sentry, /spend-prism, etc.) across Claude.ai web, Desktop, mobile, and Cowork — anywhere your Anthropic account logs in.

Requires Claude.ai Pro/Max/Team/Enterprise with code execution enabled.

Developers: Claude Code

git clone https://github.com/YOUR_USERNAME/procurement-skills.git
cd procurement-skills
./install.sh

Copies all 5 skills to ~/.claude/skills/. They're available in your next Claude Code session as slash commands and auto-trigger from natural language.

API workspaces

See INSTALL.md for API upload. Note that skills uploaded to the API are not visible in Claude.ai web or Claude Code — each surface has its own skill store.

Important: skills do not sync between surfaces. If you use Claude.ai AND Claude Code, install in both. The skills themselves are identical; the storage is what differs.

Customising for your organisation

Every skill has a config.yaml you can edit:

  • redline-sentry: liability cap thresholds, jurisdiction, contract-type playbooks
  • spend-prism: your category taxonomy (UNSPSC, eClass, or custom), currency, anomaly thresholds
  • bid-compass: scoring weights per category, quality gates, RFP structure
  • supplier-truthcheck: which checks to run, fuzzy match thresholds, rate limits
  • procure-voice: tone enforcement (rules don't need configuring; they're the skill)

The skills are designed so that anyone — you, your team, a client — can fork the repo, edit only the config.yaml files, and have a customised procurement assistant without touching skill logic.

Worked examples

Each skill folder has an examples/ directory with realistic worked outputs:

These show what to expect when each skill fires. They're also the easiest way to evaluate whether a skill fits your workflow before you customise it.

What this is not

  • Not legal advice. redline-sentry flags risk; it doesn't replace a lawyer.
  • Not commercial sanctions screening. supplier-truthcheck uses free public data sources (OFAC SDN, EU Consolidated, UK Sanctions, UN Consolidated). For full Refinitiv/LexisNexis-grade screening, use a commercial provider.
  • Not a replacement for a P2P platform. spend-prism reads exports from Coupa/Ariba/SAP; it doesn't replace them.
  • Not agents. No orchestration, no scheduled runs, no connectors. Skills are invoked by you, in the moment, and complete in a single conversation.

How to verify a skill triggered

Skills run silently by default in Claude.ai. To make activation visible, each of these skills emits a one-line tag on first invocation in a conversation:

Skill Tag
redline-sentry 🛡️ redline-sentry activated — running contract review.
spend-prism 📊 spend-prism activated — running spend analysis.
bid-compass 🧭 bid-compass activated — guiding RFP workflow.
supplier-truthcheck 🔍 supplier-truthcheck activated — running vendor validation.
procure-voice ✍️ procure-voice activated — rewriting in plain English.

If you see the tag, the skill fired. The tag appears once per conversation per skill — re-invoking the same skill later in the same chat won't repeat it.

You can also invoke any skill explicitly via slash command in Claude Code: /redline-sentry, /spend-prism, /bid-compass, /supplier-truthcheck, /procure-voice. In Claude.ai (Pro/Team/Enterprise) skills auto-trigger from natural language matching the description.

Other ways to confirm a skill is active: in Claude Code or Cowork, expand the tool-call block above the response — you'll see Claude reading SKILL.md from the relevant skill folder. Or just ask Claude: "which skill did you use?"

Roadmap (community input welcome)

Things explicitly not in v1, considered for future:

  • DOCX templates shipping with the skills (currently text-based output; templates are TODO)
  • A quality-gates.yaml shared resource so redline-sentry and bid-compass can reference common organisational standards from one place
  • Local snapshot scripts for sanctions lists (for offline bulk screening)
  • A category-intel skill for weekly market briefs (deferred — needs a scheduler, not just a skill)
  • A supplier-qbr skill (deferred — needs structured supplier performance data inputs)

Open an issue or pull request if you want one of these or have other ideas.

Contributing

Pull requests welcome, especially:

  • Additional country support in supplier-truthcheck (more company registers, more tax ID formats)
  • Additional scoring templates in bid-compass for niche categories
  • Translations of procure-voice for non-English procurement teams
  • Worked examples in your industry vertical

The skills follow Anthropic's skill-creator conventions: YAML frontmatter with name and description, body in Markdown, references and templates in subfolders. Keep SKILL.md under 500 lines; push detail to references.

License

MIT. Use it, fork it, sell services around it, embed it in your own tools. No attribution required, but appreciated.

Acknowledgements

Inspired by Anthropic's Agents for Financial Services (May 2026) — they showed the pattern works for finance. This is the same idea, applied to the category they didn't cover.

About

Open-source Claude Agent Skills for procurement: contract redlining, spend analysis, RFP, vendor validation, and tone

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages