goal
Hand the task to the agent as its own goal: the model keeps working, turn after turn, in one session, until it says the goal is met. It fits when the agent can tell for itself that the work is done; the catch is that it judges its own work, where rlar asks a separate reviewer instead. It needs a backend with a goal feature of its own.
In hmz, open /flow → Flowverses → official and install goal. Or run the release without installing anything:
hmz exec -f 'git+https://github.com/humanfia/goal-flow@v0.1.0#goal' …- repository
- humanfia/goal-flow
- release
- v0.1.0 · commit
3f516e5 - licence
- Apache-2.0
Read from humanfia/flowverse when this site was built.
The whole run, at rest. Step through it with the buttons, or drag the bar.
One turn of the flow. The ticks inside it are turns the backend started itself — profile, change, measure, again — each one that would have ended starting another, until the model says the goal is met.
The run, turn by turn
- worker — /goal bench under 200 ms; a session opened for this turn; hands the finish as words: met: 184 ms
It ends when the model says the goal is met, or the budget runs out.
See also: rlar · ralph_loop