ralph_loop
Leave one agent on a long task. Every round is a fresh session that starts from the task and the repository, so a run can go on for days without drowning in its own context. The price is that an agent that forgets may redo work, so it suits tasks whose progress shows in the repository itself, such as tests that pass one by one.
In hmz, open /flow → Flowverses → official and install ralph_loop. Or run the release without installing anything:
hmz exec -f 'git+https://github.com/humanfia/ralph-loop-flow@v0.1.0#ralph_loop' …- repository
- humanfia/ralph-loop-flow
- release
- v0.1.0 · commit
7bd23b1 - 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.
Every round starts from the task and from whatever the round before it left in the working directory — never from what it said. With every test passing, the next round starts anyway; a turn that fails ends the run.
The run, turn by turn
- agent — fixes the imports; a session opened for this turn; hands agent in the tree: 31 tests failing
- agent — fixes the parser; a session opened for this turn; hands agent in the tree: 12 failing
- agent — fixes the fixtures; a session opened for this turn; hands agent in the tree: 3 failing
- agent — fixes the last 3; a session opened for this turn; hands agent in the tree: all passing
- agent — all pass: goes on; a session opened for this turn
Then round again: nothing carries over but the repository.
It ends when the budget runs out, or a turn fails.
See also: stateful_ralph · goal · flame_chasoid:agent_cleanup_ralph_loop