Engineering study 02

Agents that run unattended

Scheduled agent work only counts if it survives a restart and its failures are visible rather than silent.

QAAPIAutomationTypeScriptPublic repository

The problem

An agent that runs on a schedule is easy to build and hard to trust. The failure mode is not a crash. It is silence. A scheduled job that stops running after a deploy is significantly harder to detect than one that never ran at all, because nothing appears in a log and no alert fires.

The job store is the point. It survives a restart, so a deploy cannot silently stop the schedule. Every result is persisted and queryable, which turns a regression into something visible rather than something discovered.

The model writes the scenarios

Upload an API collection and the agent generates the test cases, then adapts them when the API changes rather than failing against a stale suite.

Durable scheduling

The job store persists across restarts. Silence is treated as a failure mode, not an absence of one.

Every result queryable

Regressions surface as a trend rather than as a surprise in a release meeting.

Put your test coverage in front of us

Sixty minutes on what your suite misses today and what an unattended agent could hold for you.

A 60-minute architecture review · no charge · the notes are yours either way