# Purchase Approval

> Decide whether one proposed purchase is allowed under an explicit procurement policy before any spend flow starts, refusing budget overages and unapproved vendors instead of guessing.

- Skill: `runxhq/purchase-approval` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add runxhq/purchase-approval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/runxhq/purchase-approval/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Finance & Business
- Author: runxhq (https://skillmd.com/u/runxhq)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/runxhq/purchase-approval

---


# Purchase Approval

Purchase Approval is the pre-spend decision gate. It answers one question, is
this purchase allowed under the supplied procurement policy, and it answers it
deterministically. It never moves money; `spend` and the payment rails own
execution, and this skill's sealed decision is what a caller carries into that
flow.

## Procedure

1. Native `data.digest` binds the exact purchase request and procurement policy
   before any judgment.
2. Deterministic rules apply in order: the vendor must be in
   `approved_vendors`, the request currency must match the policy currency, and
   the amount must fit both `remaining_budget` and `single_purchase_cap`. The
   first failed rule refuses with a named `refused_reason`.
3. An in-policy amount that reaches `approval_threshold` returns `needs_human`
   with the policy's `human_lane` (default `procurement-review`); approval below
   the threshold is `approved`.
4. A policy missing any required authority field refuses with
   `missing_policy_authority` and findings; nothing is inferred or invented.

`spend_executed` is always false. The decision packet carries both input
digests so a downstream spend can prove which request and policy were decided.

## Output

`purchase_approval` (`runx.purchase_approval.v1`) carries `decision`
(`approved`, `refused`, `needs_human`), `reason`, `vendor`, `amount`,
`currency`, `ceiling_amount`, `human_lane`, `refused_reason`, `policy_refs`,
`validation`, and the two digests.

Inputs are `purchase_request` and `procurement_policy`.