CodeHerderSearch⌘KRequest access →

Where your agents' audit trail goes

A security reviewer asks one question first. An agent did the work. Can we see what it did, and can we prove the record is complete?

You want a yes to both. This post shows how CodeHerder answers, and where the answer stops.

One record, two exports

CodeHerder writes every governance event to one append-only log for each workspace. Task changes land there. So do approvals and sign-ins. Two exports read from that same log. You can turn on either, both, or neither.

The first is a webhook feed that fits your SIEM. The second is OpenTelemetry trace export, and it is for self-hosted servers only.

SIEM export

Any webhook subscription can send its deliveries as OCSF 1.1.0 “API Activity” records. OCSF is an open schema for security events. Splunk and AWS Security Lake ingest it. Panther and Sumo Logic do too. Your existing detections need no new parser.

Webhooks are on the Starter plan and above. To switch the format, choose OCSF when you create or edit the subscription.

Sign-ins use the same feed. An auth.succeeded event tells you who signed in. You can alert when a member signs in from an address they never used before.

What leaves

A new subscription sends metadata only. Each delivery names the event and the time. It also names the place and the actor. The payload field is null, so no event details leave CodeHerder.

Choose Full only if the receiver needs the details. Even then, CodeHerder replaces secret values with [redacted] before it signs and sends the delivery.

This default is deliberate. Most detections only need to know that something happened.

Prove the copy is complete

A feed is only useful if you know nothing is missing. Each subscription also gets a signed webhook.digest delivery about every 15 minutes. The digest lists the hash of every earlier delivery. It also links to the digest before it, so the digests form a chain.

Your team can then check three things:

  • The signature on each digest is valid.
  • Every listed delivery is in your store, and its hash matches.
  • The chain has no gap in its sequence numbers.

A failed delivery shows up as a gap when the next digest arrives. The guide Verify the audit export has the full steps and a short script.

Self-hosted: OpenTelemetry traces

If you run CodeHerder yourself, you can also send traces to your own collector. Set CH_OTEL_TRACES_ENDPOINT and restart the server. A task stage becomes a span. So does an agent session.

Span attributes have fixed types: ids, tokens, numbers, flags and times. No free text can leave, so a shared collector learns nothing about your code or briefs.

This export is not part of the hosted product. The digests do not cover it either.

What it does not do

We state the limits, because your reviewer will ask.

  • Delivery is at-least-once. You may get an event twice. Use the event id to remove duplicates.
  • A digest proves loss or change after CodeHerder queued the delivery. It does not catch a database administrator who edits an event before that.
  • The event log is append-only for the app role. A database owner can still edit rows. The log has no hash chain.
  • The newest deliveries are unproven until the next digest arrives.
  • Trace export has one destination for the whole deployment, not one per workspace.

Next steps

On Enterprise, SSO and SCIM keep sign-ins and access in your identity provider. Read the webhooks guide to set up the feed. The security page lists every control. For your own network, see self-hosting. Need the full questionnaire answers? Start with the security review.

← Back to Insights

Test it on one team for two weeks.

Connect one repository and file one task. Read the cost and the diff yourself.

Try "pricing", "connect a device", or "who reviews the code"

↑↓ move · ↵ open · esc close