One set of rules for every agent.
Developers' desktop agents and cloud agents work under the same policies, follow the same process, and leave the same record.
Guardrails scan both sides
Requests
Scanned on the way to the LLM.
Responses
Scanned on the way back.
Prompt injection
Stopped before an agent acts on it.
Bad data injection
Stopped in both directions.
The same policies and process for every agent
Consistent policies
Every agent, desktop or cloud, works under the same policies.
Consistent development process
Every change follows the same process, whichever agent made it.
Say what done means, then check it
A process lists the steps a change needs for a repository or team. Cortega records the evidence for each step on every task.
Tests passed
Read from the CI check on the pull request.
Scan clean
Read from code scanning or dependency alerts after the pull request.
Review by someone else
A reviewer who is not the author and not the person who ran the task.
Second agent review
A review from a separate review agent.
Documentation updated
Changed files include docs, or a reviewer marked docs as not needed.
Linked ticket
The pull request references the work that asked for it.
Cortega posts the result as a check on the pull request. Branch protection can require it.
Who can do what
Developers opt in
Each developer chooses to take part and can withdraw at any time.
Ticket text is data
Text from tickets, comments and chat is marked as untrusted data in the task prompt.
A different agent reviews
The review goes to a different agent from the one that did the work. Agent reviews are comments, and people approve.
A record of every step
Every offer, acceptance, run, write-back and evidence record is stored on the task with who did it.
Cloud agents behind a gateway key
Each cloud agent runs behind a gateway key with guardrails for tool calls and secrets bound to it.
Access by role
Reading task data and managing agents are separate permissions.