A coding agent can request a discount. Who gets to approve it?

An approval rule becomes useful when you can test what happens on both sides of it: the forbidden action is refused, and the permitted decision leaves evidence. A happy-path demo alone cannot show that distinction.

Here is a runnable example using Accordo, the open-source framework coding agents use to build custom CRMs. A synthetic customer wants 30 seats of an Enterprise Plan and requests 25% off. The existing policy permits automatic approval through 10%; above that, through 50%, it requires a user decision.

Run it locally

You need Git, Node.js 22.16 or newer, npm, and internet access for cloning and dependency installation. Start in an empty working directory:

git clone https://github.com/khaoss85/agent-crm.git framework-source
cd framework-source
git checkout 3b5b5f0c4c3e582e48d54501136024b064756daa
node --no-warnings examples/recipes/quote-approval/run.mjs ../my-quote-crm

The pinned recipe source creates a project, installs its dependencies and composes the existing commercial package. It then starts a temporary server on localhost and drives the public SDK through HTTP. The catalog is a fixture; the business journey does not call an external provider. It uses source from the checkout, independently of the npm scaffolder release.

Check the refusal, then the decision

The script contains assertions for each transition:

1. Server pricing produces EUR 3,750 once and EUR 2,400 per month after discount. These are synthetic quote amounts, kept in separate periods.
2. Submission under policy version 1 freezes a commercial snapshot and enters pending_approval.
3. An approval request from the simulated agent receives HTTP 403 with HUMAN_APPROVAL_REQUIRED. The quote and approval remain pending, and no business audit entry is added.
4. A simulated user approves. The quote becomes approved, with one user decision audit and a completed trace.
5. The submitted snapshot remains unchanged. There is one quote version and one approval record.

The refusal also has a failed trace. That is a useful distinction: recording an unsuccessful attempt should not manufacture a successful business decision.

Inspect the project you keep

cd ../my-quote-crm
cat data/quote-approval-receipt.json
npm run verify
npm run crm -- app inspect --json

In the receipt, inspect agentRefusal, refusalTrace, decision, humanAudit and approvalTrace. The quote, version and approval identifiers connect those records. sourceCommit and recipeSha256 identify the implementation used. The local database remains in data/accordo.sqlite; another replay needs a new target directory.

Both identities are scripted and asserted locally. No person authenticates or clicks Approve during this replay. Production needs verified identity and explicit authorization configuration; the sales-manager approval key is a label, not an authenticated role. This is not an agent-build benchmark or a production deployment.

Accordo vendors source into the project you own. You can review and adapt the rules while keeping the refusal checks. The full walkthrough includes the client brief and recorded execution.

We are looking for three teams with a real quoting or approval process to try a bounded pilot. If that sounds relevant, leave a comment with the process and the rule that must hold. Please use a synthetic example and omit customer records or other confidential information. We will agree the scope and success criteria before starting.