The fraction of agent-generated code that entered the main branch without any design governance review.
This page describes the intended design. It requires correlating merge events with audit completion timing, which isn't built yet. It doesn't appear on your dashboard yet, and no org currently has a live value for it.
AGVR is not about code quality — it is about process integrity. A non-zero AGVR means the CDA enforcement layer can be circumvented, intentionally or accidentally, by agent workflows. The audit gate that every agent-authored PR is supposed to pass through was bypassed: the PR merged without a completed CDA audit event ever being recorded.
This metric is especially critical because agents merge PRs at high velocity. A process gap that would be rare at human scale — one missed review in a quarter — becomes frequent at agent scale, where dozens of PRs may merge per day. A 5% bypass rate that seems minor in isolation translates to hundreds of unreviewed agent PRs per year entering the main branch.
AGVR should be structurally zero in a well-configured repository. The target is not near-zero — it is zero. Any non-zero value is a signal that the enforcement layer has a gap: a misconfigured webhook, a missing branch protection rule, or a repository that was connected to rkito after agent workflows were already active. Each gap is diagnosable and fixable. The metric exists to surface them before they become a pattern.
AGVR is detected by correlating merge events with CDA pipeline completion events. Agent authorship is identified from commit signatures and MCP logs. A merge with no associated COMPLETED audit record increments AGVR. Every such PR is individually flagged for retroactive review.
Any non-zero AGVR means AI-generated code is entering production without architectural review. This is not a code quality finding — it is a structural risk signal. The governance process that is supposed to gate every agent PR was bypassed. At high agent velocity, even a small bypass rate compounds into a significant unreviewed change surface.
AGVR > 0 means security-relevant agent code may have merged without Lens evaluation. This is an audit finding, not just a metric alert. Each bypassed PR is a potential unreviewed security-relevant structural change. In a regulated environment, a non-zero AGVR requires documented retroactive review of every flagged PR.
AGVR can be non-zero due to configuration gaps rather than intentional bypass. The most common causes are infrastructure, not bad actors: a repository connected to rkito after agent workflows were already running will show historical bypasses for every agent merge that predates the connection. A GitHub webhook temporarily unavailable during a merge window will produce a bypass event even though the pipeline was functioning correctly.
Always investigate whether a non-zero AGVR is a gap or a bypass before escalating. The investigation path is deterministic: check the webhook delivery log for the flagged merge timestamp, check whether the repository was connected before or after the merge, and check whether the branch protection rule requiring the rkito/cda status check was active at merge time. Each root cause has a different remediation.
Step-by-step response playbook for each signal state.
The correct state. Every agent-authored PR that merged had a completed CDA audit on record.
Monitor that webhook delivery remains healthy at /settings/integrations. A healthy AGVR of zero is contingent on reliable webhook delivery — a silent webhook failure is the most common path from zero to non-zero.
Confirm branch protection rules are enforced on all repositories with active agent workflows — not just the ones you know about. A repository added to an agent workflow without a corresponding rkito connection will not appear in AGVR at all; it will appear as a coverage gap in ABR instead.
At least one agent-authored PR merged without a completed CDA audit. Determine the root cause before the next merge window.
For each bypassed PR, determine the root cause from three possibilities: webhook failure (configuration issue — fix webhook and verify delivery), force-merge (process gap — enforce branch protection rules to prevent future force-merges), or repository not connected (coverage gap — connect the repository to rkito immediately).
Every bypassed PR requires a retroactive CDA review regardless of root cause. The review determines whether ungoverned architectural changes were introduced. Do not close the bypass finding without a completed retroactive audit on record.
Systematic bypass in progress. More than two percent of agent merges are reaching main without governance review.
Audit all recent agent-authored merges — not just the flagged ones. A systematic bypass pattern often indicates a configuration problem that has been active longer than the current AGVR window reflects.
Enforce branch protection rules requiring the rkito/cda status check to pass before merge. Review GitHub repository settings to confirm the status check is required, not optional — an optional status check can be bypassed by merging while the check is pending or not yet triggered.
Where AGVR measures whether the audit ran at all, ADRIFT measures what the audit found. A low AGVR with a high ADRIFT means audits are running but agents are failing them — a governance gap in design context, not enforcement.
AHCR compares agent and human conformance quality on PRs that completed a CDA audit. AGVR is upstream: it measures whether that audit happened at all before merge.
ABR covers all PR types — agent and human. AGVR is the agent-specific decomposition. When ABR is non-zero, check AGVR to determine how much of the bypass surface is agent-driven versus human process failure.
rkito detects governance bypasses automatically by correlating merge events with CDA audit completion records. Connect your repo and see your bypass rate within minutes.