CLI reference¶
| Command | Purpose |
|---|---|
list PATH |
list tasks (--type, --include/-k, --json) |
info TASK |
print a task's statement and spec |
validate PATH |
validate tasks (--repeat N, --sandbox, -v, --json); exit 1 if any fail |
run PATH |
run an agent (--agent, --workers, --record, --output, --report, --keep, --sandbox); exit 1 if any task fails |
report RESULTS |
render results.json (-o report.html / .md) |
mine REPO |
mine tasks from git history (--limit/-l commits to scan, --max-tasks/-n, --rev, --test-command, --setup, --out) |
mutate REPO |
synthesise tasks by fault injection (--max-tasks/-n, --sample, --seed, --include, --exclude, --out) |
export PATH |
tasks → JSONL (--include-repo) or --trajectories DIR → JSONL (--only-passed) |
new PATH |
scaffold a task directory (--type) |
solution TASK --from DIR |
write solution.patch by diffing the snapshot against a fixed copy (--include-tests, -o) |
doctor |
show available toolchains and Docker status |
PATH may be a single task directory or a tree of them; -k accepts glob patterns
on task ids, --tag filters by tag, and both may be repeated. In mine and
mutate, -n always means the maximum number of tasks to write.
Agent specs¶
golden · noop · claude[:model] · shell:<command with {prompt}> ·
package.module:ClassName
Sandbox specs¶
local (default) · docker (python:3.12-slim) · docker:<image>
Exit codes¶
| Code | Meaning |
|---|---|
| 0 | success / all passed |
| 1 | some tasks invalid or failed |
| 2 | usage or configuration error |