Legal
Data processing addendum
A structural draft: the processing, the roles and the measures it can point at.
What this document will cover
- 1. Roles
- Which party is controller and which is processor, per category of data.
- 2. Subject matter and duration
- The processing: running agent Sessions against a customer's repositories and trackers, producing Artifacts, reviews and Evidence, and retaining the audit trail.
- 3. Categories of data and of data subjects
- Repository content, tracker content, and the identities of the people who approve, override and comment.
And the clauses a reviewer reads first
- 4. Instructions
- That we process on documented instruction only, and what happens when an instruction conflicts with law.
- 5. Confidentiality and security
- The technical and organisational measures, which are the ones described on the enterprise page rather than a generic list.
- 6. Subprocessors
- Authorisation, the current list, and notice before it changes.
- 7. International transfers
- The mechanism relied on.
- 8. Assistance, audit and deletion
- Help with data-subject requests, audit rights, and what happens at the end of the agreement.
- 9. Breach notification
- Timing and content of a notice to you.
Measures a drafted version can point at
These are implemented and checkable today, which is what makes them worth naming in a schedule rather than describing in the abstract:
- Per-Task confinement: every Session runs in its own pod under a gVisor RuntimeClass, asserted before an Implement Phase and refused if absent.
- No direct egress from a Session; credentials are held by the control plane.
- Per-Role tools, mount mode and secret scopes, so no single Role holds private data, Untrusted Content and a route out at once.
- An audit trail of who approved, started, overrode and changed what.