Skip to main content
make-workflow-tutorial is the guided first-run path for turning “what should I automate?” into a real Smithers workflow. It reads the Smithers docs, scans the current repo, samples recent external coding-agent sessions, recommends a ranked list of workflow ideas, asks you which one to build, then delegates the build to create-workflow. 
bunx smithers-orchestrator workflow run make-workflow-tutorial
You can bias the recommendation step with a hint: 
bunx smithers-orchestrator workflow run make-workflow-tutorial \
  --input '{"hint":"Find a workflow for release prep and changelog publishing"}'

​
Stages

  1. bootstrap: read the repo file tree, package.json, recent commits, and the Smithers docs index.
  2. sessions: bounded, read-only scan of recent Claude Code, Codex, and Pi session JSONL files, with secrets redacted.
  3. recommend: rank five candidate workflows for this codebase and recent work.
  4. pick: pause for the human to choose or modify one recommendation.
  5. build-launch: start create-workflow with the selected goal and require a custom .smithers/ui/<id>.tsx.
  6. monitor-loop: poll the child run, narrate progress, resume stale work, and triage failures when possible.
  7. dive-deeper: read human guide docs and preview other “You say -> Smithers runs” capabilities.

​
Inputs

InputTypeDefault
hintstring or nullnull (infer from repo and sessions)

​
What it reads

The session reader is read-only and bounded. It samples recent files from ~/.claude/projects, ~/.codex/sessions, ~/.codex/history.jsonl, and ~/.pi/agent/sessions, extracts user and assistant text where schemas are known, and redacts API keys and bearer tokens before passing snippets to the recommendation task. Use this when you have just initialized Smithers and want your agent to discover a high-leverage workflow from your actual repo and recent work instead of starting from a blank catalog.