Questionnaire CLI workflow and context

Context

The questions CLI expects a JSON array of question definitions and uses Ink to walk a human operator through them. Users typically pipe JSON or point the CLI at a file (questions sample_questions.json). The entrypoint normalizes argv, reads the question set, validates each prompt, and boots the Ink app. Inside the TUI, the operator navigates with [/h and ]/l, arrow keys or j/k, and Enter/Space depending on the question type. Yes/no prompts always include a third "Other" row that opens a custom input so operators can supply free-form context without toggling configuration. Answers are cached locally until the final submit screen, where the UI is replaced with the generated markdown; the CLI simultaneously writes the same payload to --output when set (and only logs directly to stdout when running non-interactively). The finalOutput state doubles as the "submitted" flag—once it holds a string, prior screens unmount and the only available action is confirming exit.

Relevant Files

• src/cli.tsx – parses CLI flags, loads question JSON from stdin/file, validates structure, and renders the Ink app.
• src/components/App.tsx – orchestrates per-question state, keyboard handling, navigation, and submission lifecycle.
• src/components/{SingleSelect,MultiSelect,FreeForm,YesNo}.tsx – renderers for each supported question type, including custom-option UX.
• src/components/Header.tsx – displays global navigation affordances and current progress counter.
• src/components/SubmitScreen.tsx – review + submit view that surfaces every answer and triggers finalizing logic on Enter.
• src/utils/markdown.ts – turns (questions, answers) into the markdown payload written to stdout/file.
• sample_questions.json – reference dataset demonstrating the JSON schema accepted by the CLI.

Relevant Types

• Question union (src/types.ts) – guarantees each prompt carries id, question, and type-specific metadata (options, suggestions).
• Answer (src/types.ts) – union used across the UI to store the current value for each question id.
• parseQuestions(input) (src/cli.tsx) – runtime guard that enforces the schema before Ink ever mounts.
• generateMarkdown(questions, answers) (src/utils/markdown.ts) – shared formatter used by the submit handler to keep stdout and file output in sync.

Workflow

1. Input resolution – src/cli.tsx inspects process.argv, supports -f/--file and positional file args, falls back to piped stdin, or exits with help if no payload exists.
2. Question validation – raw input is parsed as JSON; every entry must have id, question, and a supported type, plus options/suggested arrays where required. Failures exit early so the Ink UI never loads bad data.
3. App initialization – render(<App ... />) seeds per-question UI state (selection indexes, custom strings, multi-select sets) and tracks currentStep to know which component to show.
4. Interactive answering – useInput inside App centralizes keyboard handling ([/h and ]/l to move steps, arrows or j/k to change the highlighted option, Space for multi-select toggles, Enter/Space to confirm). Components receive state slices to display cursors, custom text fields, and selection markers.
5. Saving answers – saveAnswer() maps current UI state into the canonical Answer union per question id; multi-select answers are arrays, custom entries persist when Enter/Space is pressed, yes/no returns booleans for Yes/No or a custom string when "Other" is confirmed, and skipped questions remain null.
6. Submission and output – when the operator advances past the last question, SubmitScreen shows a summary. Pressing Enter/Space once generates the markdown via generateMarkdown, stores it in finalOutput, and swaps the UI to a read-only Markdown view; finalOutput !== null now indicates the session is locked and no prior step can be revisited. Any subsequent Enter/Space (or key press) triggers the final onSubmit/exit path, which writes to --output (or stdout) and terminates the Ink app.

Edge Cases

• No input / unreadable file – CLI errors with guidance if stdin/file is empty or cannot be read.
• Schema violations – missing ids, unsupported types, or absent options arrays throw descriptive errors before entering the interactive flow.
• Custom option handling – multi-select custom entries are only stored once the user confirms the typed text; hitting Escape clears the draft.
• Empty selections – multi-select answers serialize as _none selected_/(none selected) in markdown so downstream readers know it was intentional.
• Output failures – disk write errors propagate from Bun.write; the CLI surfaces them after printing to stdout so callers can retry another path.

Questions to Clarify

• Should the CLI ship with additional built-in templates beyond sample_questions.json, or fetch remote question sets?
• Do we need persistence/resume capabilities so partially answered sessions can be reloaded?
• How should automated pipelines (non-interactive environments) provide answers—should we allow defaults or scripted input?
• Is markdown the only required output, or should alternate formats (JSON/HTML) be supported behind additional flags?