Implement fail-closed agent decisions

Missing authority facts should not silently become ALLOW for consequential actions.

01 / The operating case

What changes in a real workflow?

A restart agent loses access to its policy service during a production incident.

Trace the authority boundary
  1. 01Authority origin
  2. 02Task and grant
  3. 03Current decision
  4. 04Observed outcome

02 / Decision contract

What the executor must check

Classify effects, deny restart on indeterminate decision, preserve low-risk reads only under separately valid local policy and escalate through a documented emergency path.

InputPrincipal · task · action · target
DecisionGrant ancestry · policy · current state
ResultEnforce · record · verify outcome

03 / Failure and evidence

What goes wrong, and what can be proven?

Failure mode

A network timeout is interpreted as approval or yesterday’s ALLOW is reused.

Evidence to retain

Log dependency failure, local mode, denied request and later reconciliation.

04 / Canonical scope

Why this reference stands alone

Degraded semantics owns alternate reduced operation; this page sets the mandatory failure default.

Illustrative implementation artifact

Unavailable-decision matrix

low-risk read + valid local grant => limited read
production write + PDP unavailable => DENY
unknown revocation state + high impact => DENY
explicit break-glass activation => separate audited path

Primary references

Read the underlying material

Architecture discussion

Choose one consequential action and make its boundary explicit.

Request a Conversation