korchasa/run

Run flowai-workflow

Execute a flowai-workflow DAG via the flowai-workflow engine binary (plugin precondition — FR-E78). The binary itself bundles the canonical workflows; the operator may also adapt a project-local copy under <project-root>/.flowai-workflow/<name>/.

Preflight

Before launching, verify the engine is on PATH:

command -v flowai-workflow >/dev/null 2>&1 || { echo "Error: flowai-workflow is required — see https://github.com/korchasa/flowai-workflow#install" >&2; exit 127; }

If the binary is missing, stop and surface the install link. Do not silently fall back to a different runtime.

Resolve the workflow

The first positional argument names the workflow. Resolve it in this order:

1. If the argument is an existing directory containing workflow.yaml, use it as-is.
2. If the argument matches a sibling under <project-root>/.flowai-workflow/, use the project-local copy.
3. Otherwise, treat the argument as a bundled workflow name and rely on the engine's built-in catalogue (see flowai-workflow init --list).
4. If none of the above resolves, list available workflows and ask the user which one to run.

Launch

flowai-workflow run "<resolved-workflow-path>" [extra args]

Forward any additional CLI flags the user provides (e.g. --prompt, --dry-run, --cycles). Stream output back to the user in normal verbosity unless they asked for -v / -q.

Common flags

• --prompt "<text>" — extra context for the PM agent
• --resume <run-id> — resume a previous run
• --dry-run — print the execution plan without running
• -v / -s / -q — verbose / semi-verbose / quiet
• --cycles <N> — repeat the run N times sequentially

Errors

• "Missing workflow.yaml" → the resolved path is wrong; list candidates again and re-prompt.
• "Workflow validation failed" → surface the engine's error verbatim; point the user at scaffold to adapt the config.
• "Permission denied: …" — the engine writes under <workflow>/runs/<run-id>/ and the per-run worktree; if the workflow lives in a read-only location, instruct the user to first /flowai-workflow:init into their project so runs can write into the project-local copy.