An AI agent can be approved and still leave you with an unknown result

Description and use

An approval that an AI agent may attempt an action does not guarantee the remote service completed it. In Agent Work Layer, operations that end with an unknown result are marked as "unknown" and are not retried automatically, allowing operators to inspect and retry if needed.

Problem

An approval that an AI agent may attempt an action does not guarantee the remote service completed it, leading to uncertain outcomes.

Outcome

A reproduction of PR #43 on Linux x86_64 with Node v24.22.1 ran 42 passing tests in ~21.95 seconds, confirming that unknown outcomes are correctly handled and that retries are not automatically triggered.

Workflow

• 1. Gateway records operation start.
• 2. Receiver performs the effect.
• 3. Receiver returns a result.
• 4. Gateway records the result.
• 5. If gateway dies before step 4, operation status is set to "unknown".
• 6. Operator can inspect logs and decide to retry or settle the unknown outcome.

Evidence

An approval answers whether an agent may attempt an action. It does not tell you whether a remote service completed it.

Source

The useful outcome was a change we could inspect and rerun.

Source

On October 27, 2026, the maintainer-side reproduction checked out PR #43's exact head, 493157aee3391ec89fe1ffd301a402ae6e61861c, on Linux x86_64 with Node v24.22.1.

Source

The focused suite finished with 42 passing tests, zero failures and zero skips, exit 0, in about 21.95 seconds.

Source

First, recovery from the killed process includes manual lock removal.

Source

Second, the gateway test representing the before/after-send ambiguity injects a running log entry.

Source