cli.md

CLI reference

Global

flow --version
flow --help

Run

flow run workflow.py \
  --args '{"target":"src"}' \
  --budget '{"max_usd":1,"max_calls":8,"max_tokens":50000}' \
  --run-id release-audit \
  --executor thread \
  --max-workers 8

Natural-language authoring plus execution:

flow run --nl "audit this repo across correctness/security/performance" \
  --budget '{"max_usd":1,"max_calls":8}'

Dry-run authoring without model execution:

flow run --nl "audit this repo" --dry-run --estimate-cost

Resume

flow resume <run_id> workflow.py

Trace

flow trace <run_id>
flow trace <run_id> --json

Status

flow status <run_id>
flow status <run_id> --json

Shows node states and what resume would rerun.

Author

flow author "classify these files and summarize the risky ones" -o workflow.py

Writes a validated workflow script without running it.

List

flow list

Doctor

flow doctor

Prints environment, config path, data root, and provider diagnostics. It must never print secret values.

Init

flow init
flow init --force

Writes a starter config. JSON is used when YAML support is unavailable. YAML is used when PyYAML is installed.

Self-test

flow self-test --offline
flow self-test --online
flow self-test --json

--offline performs import, config, data-root, and local DAG checks with no network. --online adds one real model call through the configured router.