Human review planner · provisional name
Review Gate
Put human judgment where an error can still be caught and changed.
Sets review depth, timing, coverage, evidence, and escalation rules from the work's impact, reversibility, uncertainty, verification, and access. A gate is useful only when it sits before the drop, has a named keeper, and can actually stop the load.
Input bench
Name the real material
Name the artifact, recommendation, or system action a person will review.
Choose the furthest point the system reaches in the workflow.
Score the largest plausible consequence, not the most common one.
How completely can a mistaken result be repaired?
How settled are the governing facts, rules, and interpretation?
Choose what can reliably detect a material error.
Choose the most sensitive input or consequential permission involved.
Wider or individualized use usually needs stronger release control.
What must be true for the reviewer to approve the result?
List the sources, calculations, tests, comparisons, or logs the reviewer receives. One per line.
List the errors the reviewer must actively look for. One per line.
Name the qualified role, what they can approve, and who handles escalations.
For Tier 4 dual control, name a different authorized person or role who gives the final release approval. Leave blank when separate approval is not required.
PRIVACY: INPUTS STAY IN THIS BROWSER. SAVE USES LOCAL STORAGE ON THIS DEVICE. NOTHING IS SENT BY THIS INSTRUMENT.
Output desk
Your result appears here
Load the sample or bring real material.
THE TOOL TRANSFORMS WHAT YOU ENTER. IT DOES NOT CALL A MODEL OR INVENT MISSING FACTS.
Worked example
Starting material
A production billing migration can change customer balances at scale, and a plausible rollback must be proven before any live write.
Useful change
Tier 4 review: full pre-use engineering verification, separate billing-owner go/no-go approval, reconciled test evidence, a timed rollback rehearsal, and a hard stop when authority or evidence is missing.
Put the result to work
A method to follow. A prompt to take with you.
Prompts
Human Approval Gate Designer
Place human review where it changes risk rather than adding ceremonial approval to every step.
Skills & workflows
Verification Budget
Decide how much verification an AI agent's output actually needs based on real stakes, instead of checking everything or nothing.
Workflow recipes
Add Human Approval Gates to an Agent Workflow
Place deliberate review points before an agent creates cost, changes records, reaches people, or crosses a privacy boundary.
Tutorials
Agent Reliability for Nondevelopers
Design agent work so a nondeveloper can see the contract, authority boundary, evidence, failure, and recovery path.
StrAItegy writing ↗
The Rules Nobody Wrote Down
The unwritten operating rules of working with AI agents, written down.