Three ways to start a run
- Now. Open a workflow, give it a brief, run it.
- On a schedule. Attach a cron. Forge ticks every schedule, humanises the cron expression in the dashboard, and shows last-run status on one cross-workflow Schedules page.
- On an event. A webhook or an upstream run kicks it off. Runs can generate child runs with their own inputs — bounded by loop-detection so an ancestor can’t be regenerated into an infinite chain.
The Roadmap Driver
This is the autonomy story most teams come for. Point Forge at a roadmap and the driver runs the loop end to end: reconcile what’s done, pick the next actionable item, dispatch the SDLC workflow , and — when the change passes its own isolated-stack verification and review — auto-merge the PR. It keeps the loop moving without a human nudging every turn, and it picks up the right next thing rather than just the next-numbered thing.
The mission is a fully trustworthy autonomous SDLC where the operator’s attention is the scarce resource — spent only at the human-in-the-loop gates that genuinely need a human, never on babysitting each turn.
Autonomy with guardrails
Autonomous doesn’t mean unsupervised:
- A management agent validates every handoff — advance, enrich, or escalate. Bad output doesn’t flow downstream.
- A liveness watchdog enforces a wall-clock progress timeout with tiered recovery escalation, so a stuck run surfaces instead of hanging.
- Operator decisions are protected — once you unstick a run, the stale-sweeper won’t prematurely terminate it.
- Escalations reach you in chat — approve, retry, or terminate right from Telegram .
Describe the cadence once; Forge keeps the loop honest.