Reference adapter

Watch the guard decide.

GitHub Gateway recorded decision flow

Recorded runs use sanitized Gateway results. Payloads, secrets, and raw repository contents stay out of view.

Self-hosted technical preview for controlled evaluation. No auto-merge. Human review remains required. The evaluation path does not prove semantic correctness.

Scenario: Review feedback updates the same Gateway PR

Agent intent

  • Reviewer asks for one missing config field.
  • Agent targets existing Gateway PR #30.
  • It validates the expected parent head before updating.

Gateway judgment

  • parent_pr_open: pass
  • parent_head: matched
  • path_policy and state_binding: pass

Visible result

  • PR #30 updated.
  • No duplicate PR created.
  • Review remains required.
agent-gateway-timeline
Click "Replay recorded run" to start the timeline.

Current checks

Active step: idle

Intent

not reached

Format

not reached

Rules

not reached

Target

not reached

Content

not reached

Drift

not reached

Replay

not reached

Decision

not reached

Outcome

not reached

Decision status

AdmittedImpact: Same PR updated
View sample output #30

Recent decisions

TimeOperationDecisionStatusReject stageImpactOutputNext action

Recorded output is sanitized. Payloads, secrets, runner keys, raw file contents, private repository names, internal identifiers, and IP addresses are not displayed.

What it does:

Creates reviewable pull requests.

What it doesn't:

It does not give agents write tokens.