Approval Gates (Governance Gates) — Approval gates are governance checkpoints in a software delivery pipeline that a change must pass before it can progress — for example, requiring human sign-off, a passing blast-radius check, or resolved spec drift before a work order advances or a release ships.
Gates turn policy into enforcement. Instead of relying on convention ('please get review'), a gate blocks progression until its condition is met, producing an audit trail of who approved what and why.
In AI-native delivery, gates are how you keep agent velocity from outrunning governance: an agent can generate change quickly, but a gate ensures a human or an automated check validates it before it counts as done or shippable.
Skaftor gates progression between delivery phases (requirements → blueprints → work orders) and on releases, combining human approvals with automated checks like readiness scoring and drift status.
A gate has three parts: a trigger (the transition it guards — a release, or a move between delivery phases), a condition (what must be true — an approval, a passing check, a threshold), and an outcome (block, warn, or allow with an audit record). Because the condition is explicit, a gate is enforceable and reviewable in a way an informal "looks good to me" never is.
Gates come in two forms that work together. Human gates require a named approver to sign off; automated gates require a check to pass — tests green, blast radius under a threshold, spec drift resolved, release confidence above the bar. AI-native teams lean on automated gates to keep pace with agent output, reserving human gates for the decisions that genuinely need judgment.
CI checks typically verify code mechanics (build, tests, lint). Governance gates enforce process and policy — human approvals, readiness thresholds, resolved drift, acceptable blast radius — and gate progression across the delivery lifecycle, not just the build.
Common gates include a human sign-off before a release ships; a blast-radius check that blocks a change touching too many service boundaries; a spec-drift gate that stops a work order whose implementation diverged from its requirements; and a release-confidence threshold that must clear before deploy. Each blocks progression until its condition is met.
AI coding agents can produce more change than a team can manually review. Gates keep that velocity governed: an agent implements quickly, but the change only counts as done once it clears its gates — a human approval, passing checks, resolved drift — so speed never bypasses policy or accountability.
Release confidence is a composite readiness score that answers 'should we ship this?' by c…
In software delivery, a change's blast radius is the full set of components, contracts, se…
The Intent-to-Production Thread is unbroken traceability from a unit of business intent to…
AI-native software delivery is a way of building software in which humans express intent a…