Continuation
The hooks that refuse to let agents stop while work is unfinished — todo-continuation, ralph-loop, boulder state.
Continuation is OmO's core philosophy made operational. When work is unfinished, hooks push the agent back into motion. The Sisyphus metaphor is literal — the boulder never stops.
| Hook | Event | What it does |
|---|---|---|
todo-continuation-enforcer | Event | Yanks idle agents back to work when todos remain. Prevents "I'm done" lies. |
compaction-todo-preserver | Event | Preserves todo state across context-window compactions so continuation has something to resume on. |
ralph-loop | Event + Message | Manages the self-referential loop. Detects <promise>DONE</promise>, auto-continues on early stops. |
start-work | Message | Handles /start-work — activates Atlas on the latest Prometheus plan. |
auto-slash-command | Message | Detects slash-command intent in plain prompts and routes accordingly. |
stop-continuation-guard | Event + Message | Guards the /stop-continuation mechanism — when explicitly stopped, continuation does not silently re-engage. |
agent-usage-reminder | PostToolUse + Event | Periodically reminds the agent that specialized lanes (Oracle, Librarian, Explore) exist when the current path looks generic. |
When to disable
Almost never. Continuation is load-bearing for ultrawork, ralph loops, and /start-work flows. If a specific loop is misbehaving, disable that hook individually via disabled_hooks rather than turning off the whole continuation tier.
{ "disabled_hooks": ["ralph-loop"] }Source Notes
Aligned with upstream docs/reference/features.md#continuation and #productivity--control. The continuation principle itself lives in Manifesto.