continue_loop

Send the task once, then keep saying "continue" to the same session, which is what a person at a prompt would type. The agent keeps its whole conversation, and every round it is told to carry on rather than told the task again.

ships with humanize

Nothing to install: it is in hmz from the first run, with chat and the other loops, and its code is humanize's flows/builtin/continue_loop.

at the prompthmz exec

text

❯ $continue_loop port the test suite from unittest to pytest

hmz exec -f continue_loop -a agent=kimi/kimi-code/k3:high \
    -p budget.duration=6h "$(cat TASK.md)"

continue_loopsimulated

The whole run, at rest. Step through it with the buttons, or drag the bar.

Until a turn answers, the task is sent again rather than “continue”: the word means something only to a session that heard the task.

The run, turn by turn

agent — the task; a session opened for this turn
agent — “continue”; another turn of the session it already had
agent — “continue”; another turn of the session it already had
agent — “continue”; another turn of the session it already had

Then round again: until the budget is spent.

It ends when the budget runs out, or 3 failed turns in a row.

When to use it

Like stateful_ralph, it holds one session, so the agent remembers what it tried. The difference is the prompt: stateful_ralph repeats the task, which pulls the agent back to it every round, while "continue" lets it follow its own plan.

"Continue" only makes sense to a session that heard the task. Until a turn has answered, the flow sends the task again rather than "continue".

Roles and params

Role	What it is	How it is filled
`agent`	agent, required	`-a agent=…`	Takes every round, in the one session the run holds.
`workspace`	environment, local	the directory you start in; no `-e`	Where every round works.

Each agent role takes one -a role=CLI[@PROVIDER]/MODEL[:EFFORT]; several roles may share one -a, comma-separated. There is no -e to give: workspace is a local environment, the directory you start the run in, and an -e naming it is refused. See Command-line specs.

No params. The loop pauses 5 seconds between rounds.

What ends it

The budget. However much the agent says it is done, it is told to continue, so this is the usual end.
Three failed turns in a row. The run ends with the last failure. A turn that answers with nothing is not a failure; it is simply sent again.

Picking it up

--resume carries on the round count. The session is not picked up: the resumed run opens a new one, which has heard nothing, so it is sent the task first, exactly as the first run was. See Picking a run up.

continue_loop ​

When to use it ​

Roles and params ​

What ends it ​

Picking it up ​

See also ​

continue_loop

When to use it

Roles and params

What ends it

Picking it up

See also