Infrastructure · Verification
HumanApprove
Pause for human approval before critical actions
Definition
Mechanism
A checkpoint gate where execution pauses and awaits explicit human approval before proceeding. The agent presents its proposed Task, rationale, and risk assessment to a human operator. Only upon receiving affirmative consent does execution continue. Critical for high-stakes actions (financial transactions, deployments, irreversible changes).
Invariants
- Presents the proposed Task, its rationale and its risk assessment before asking.
- Halts execution until a human answers, or until `timeout`.
- Requires a context-specific challenge code when `challenge_required`.
- Logs every approval and rejection with a timestamp and a rationale, in the Audit trail.
- On timeout the action does not proceed: silence is not consent.
Risks
Failure modes
Approval Fatigue
Humans rubber-stamp requests without review due to high volume.
Blocking
System halts indefinitely if human is unavailable.
Missing decision information
Human lacks sufficient information about the proposal, rationale, or risk to make an informed decision.
Caution
Bypassing removes the human-in-the-loop safety boundary.
Relationships
Builds on
Accepts
The atomic unit of intent
Used by
References
Mind
Value-based state evaluation
Society
Requesting elevated privileges for sensitive operations
Society
Reversibility Audit (Alias for Check(Reversibility))
Reference
Parameters
challenge_requiredBoolean, {true, false}- Whether the approver must match a challenge code, not merely click through.
timeoutDuration, unspecified- Longest the gate waits for a human before escalating. Deliberately uncapped.