Oneshot
A Slack-native work agent that also keeps watch.
Andrew built Oneshot to strip the repetitive layer off his own work and give the team a capable operator inside Slack. A request can start in a DM, a mention, a watched thread, a GitHub event, or a scheduled check; the agent carries it into the right repository and reports back in the conversation it came from.
It is built to share Andrew's operating instincts: precise about who owns what, sceptical of state it has not verified, and expected to finish the whole authorised job rather than the interesting half. A merge is not a deployment and a deployment is not proof, so tests, review, release, and live behaviour stay separate gates.
Its watches cover Andrew's pull requests, release drift, critical product journeys, repository hygiene, team handovers, and the health of neighbouring agents. Most checks end in silence. Work reaches Slack when there is a result, a genuine blocker, or a decision that still belongs to a person.
Operating model
What makes Oneshot itself.
- 01
Slack-native cockpit
Thread-scoped context, streamed progress, cancellable work, and a private App Home, all where the team already works.
- 02
Watches that stay quiet
Event-driven and scheduled loops pick up eligible PRs, incidents, drift, and unanswered responsibilities. Most of them end without a message.
- 03
Skill-routed execution
Each job loads its own written procedure instead of asking one long system prompt to improvise engineering, QA, research, and comms.
- 04
Evidence-led delivery
Repository state, CI, deployment identity, live behaviour, and the authority to act are each verified before work is called finished.
