All commands
Every hillclimb command, generated from the CLI itself.
One page per command, with its arguments, options and defaults as the CLI itself declares them. The CLI reference is the shorter, hand-written view: which command you want, and why.
Getting a problem and running it
| command | what it does |
|---|---|
hillclimb init | Create a hillclimb dir. |
hillclimb problem … | Ready-made verifier problems shipped with hillclimb. |
hillclimb demo | Try hillclimb in one command: agents climb the circle-packing problem. |
hillclimb verify | Run a problem's verifier once, outside a search. |
hillclimb run | Start a run on a problem or a run-spec YAML. |
hillclimb resume | Resume a parked or interrupted search. |
hillclimb smoke | One real DRAFT call through the selected agent, end to end. |
Watching a climb
| command | what it does |
|---|---|
hillclimb status | Show the candidate tree of a search. |
hillclimb watch … | Live TUI: runs, searches, candidates, and candidate details. |
hillclimb chart | Live hillclimb chart: best score so far vs tested candidates. |
hillclimb tree | Live exploration tree of one search: what was expanded, what was left. |
hillclimb tree2 | Live archive tree of one search, drawn like the Darwin Gödel Machine's. |
hillclimb archive | Archive tree beside the progress chart, live — the Darwin Gödel Machine's two-panel figure for one search. |
hillclimb surface | Live 3D fitness surface of one search: candidates on the problem's terrain. |
hillclimb similarity … | Live 3D similarity views of a search's candidates: map (pairwise, default) and reference. |
hillclimb graph | Live knowledge graph (same screen as hillclimb knowledge graph). |
hillclimb show | Everything known about one candidate. |
Controlling a climb
| command | what it does |
|---|---|
hillclimb ps | Every process hillclimb is responsible for on this machine. |
hillclimb stop | Gracefully stop a running engine. |
hillclimb kill | SIGTERM a running engine; it can be resumed. |
hillclimb prune | Prune a candidate and its whole subtree. |
hillclimb summit | Copy the best solution found so far next to your hillclimb/ folder. |
hillclimb reset | Kill every engine of THIS hillclimb dir and delete the dir. |
Memory
| command | what it does |
|---|---|
hillclimb knowledge … | Cross-search learning: cards distilled from finished searches |
hillclimb paper … | Distill PDF papers into knowledge claims that seed future searches |
Experiments and policies
| command | what it does |
|---|---|
hillclimb experiment … | Compare setups: named arms of config overrides × repeats on a problem |
hillclimb store … | The record store behind the cross-run views (store.backend in config.yaml: files | sqlite) |
Everything else
| command | what it does |
|---|---|
hillclimb disconnect | Undo hillclimb connect <target> on hillclimb's side: unpin it as the default and forget the staged credentials, so the next connect sets it up from scratch again. |
hillclimb climber … | Climbers — the shareable bundle that decides HOW to hillclimb (policy, operators, prompts, tuner): list them, start your own, check one before spending budget on it |
hillclimb connect … | Connect the agents that run operators (claude, codex, pi) and the OpenRouter route that pays for them. |
hillclimb sandbox … | The sandbox agents and solutions run in: what it stops, shown on this machine |