Clear rules for data and changes.

How we handle data, tool access and sensitive changes.

Review the system first

We review code and documentation first, then agree on which changes we may make.

Least privilege

We limit access to code, tasks, documents and messages to the agreed work.

No training by default

Client data is not used for model training unless explicitly agreed in writing.

Approval of sensitive changes

A person reviews sensitive edits, public changes, data exports and production-impacting steps before execution.

Information with a source

Saved information needs a source, an owner, an allowed scope and a review date. New decisions replace older ones.

Private stays private

We do not publish client code, internal paths, operational details, secrets, or private repo history.

Six checks before the change

The same questions apply to code, a migration, an AI workflow, or a production operation. Speed only matters when the change stays understandable and verifiable.

Source

Which repo, ticket, incident, document, or decision supports the claim.

Freshness

Whether the information is still current or a newer record has replaced it.

Permission

Whether this context is allowed for this team, tool, actor, and task.

Risk

What could break production, leak data, or lead the agent to a wrong conclusion.

Change

What changed since the last run, deploy, incident, or recorded decision.

Trace

Whether a person can replay the path: sources, permissions, decision, and result.