Execution identity
The agent proposes work.
A scoped execution API key starts Tasks, requests checkpoints, and records results. It cannot approve its own REVIEW action.
Security and trust
Team Guardrails adds human-approved project rules to the existing supported control boundary. The Rulebook is locally validated but remains pending hosted migration, deployment, and production smoke.
Execution identity
A scoped execution API key starts Tasks, requests checkpoints, and records results. It cannot approve its own REVIEW action.
Human identity
A short-lived human session identifies the reviewer. The server rechecks organization membership and existing decision permissions.
Verification source
The independent verifier reads actual Git and filesystem state and runs the declared checks. Agent completion claims cannot certify success.
Team Guardrails
An organization owner creates, versions, or disables a rule. New controlled Tasks inherit a bounded frozen snapshot. Existing Tasks keep the version they started with.
A matching supported mutation is held for an authorized human decision before execution.
A matching supported mutation is denied. Task expected scope cannot weaken the project rule.
Evidence records the frozen rule version, match, decision, and actual execution result without rewriting history.
This behavior is a release candidate capability, not yet a production availability claim.
Before mutation
The Codex alpha uses a dedicated configuration with PreToolUse interception and PostToolUse result reconciliation. Required hooks, entrypoints, credentials, and verifier availability are checked before controlled work starts.
Codex reads through the bounded reader and proposes a supported apply_patch update.
Minimized operation metadata is checked against deterministic Task scope before mutation.
Authorized work continues through its intended invocation. PROTECTED stays blocked; REVIEW waits for a human.
When required control components cannot be validated, the supported launcher refuses to start the controlled Codex session.
Minimized engineering data
Truthful results
Authorization, execution, and verification are distinct states. Generic external callbacks and result recording are not one atomic transaction; an uncertain result must not trigger a blind retry.
Every required deterministic check and declared repository outcome passed.
A required engineering condition was conclusively false, even if the patch executed successfully.
Repository truth or required checks could not be established because of environment or infrastructure failure.
Access and current limits
Execution keys are limited by organization, project, agent, and integration channel. A key authenticates instrumented traffic; it does not automatically observe an agent.
Task and Evidence links require authentication and authorized workspace access. Possession of a URL grants no access.
Codex 0.157.1, one repository, and one existing-file apply_patch update per operation. Team Guardrails remains a pilot preview. REVIEW continuation is process-bound. Verification proves only the declared deterministic contract.
Keep local human credentials separate from agent execution credentials. Evaluate the repository, declared checks, reviewer access, and supported launch configuration before a pilot.
The alpha assumes a trusted developer host, reviewed repository and verification scripts, and a dedicated worktree without concurrent writers. It does not defend against malicious repository code running with the verifier's host-user privileges.
Tracking ACTIVE does not mean every Codex action or host process is observed or controlled. Missing required evidence remains UNCERTAIN. Uncertain recording is not a reason to retry blindly.
Founding-team pilot
Team Guardrails is pending release. Apply with one supported Codex workflow and the review rules your team already repeats.
Need a local evidence report instead? Change Proof is locally validated and founder-assisted, not publicly distributed. Read the guide.