Drop @inquirer/prompts

CHANGELOG.md

@@ -8,6 +8,8 @@ Each released version is tagged in git (`v0.0.1`, `v0.1.0`, etc.) and includes t
 
 ### Removed
 
+- **`@inquirer/prompts` dropped; CLI is purely flag-driven now.** Every interactive prompt is gone: init's idea input, tech/tasks' spec picker, and scaffold's confirm. Each was a TTY-only convenience layer on top of flags that already drove the canonical input path. With api mode gone (see entry below), the remaining prompts had nothing meaningful to gate — the CLI's job is to load context and hand off to the host coding agent, not run a Q&A loop. Concrete behavior changes: `tech` / `tasks` with multiple specs and no slug arg now error with the available list (was: TTY → inquirer picker, non-TTY → error); `scaffold` requires `--yes` to confirm before writing (was: TTY → inquirer confirm, non-TTY → error); `init` greenfield without `--idea` always prints the structured agent handoff (was: TTY → inquirer input, non-TTY → handoff). The TTY/non-TTY distinction is gone entirely — `src/utils/tty.js`, `test/setup.js`, and `vitest.config.js` are deleted; `deps.isInteractive` is no longer a dependency-injection seam. CLAUDE.md and README.md updated to drop "TTY-only fallback" framing and replace with "flags drive input; CLI never blocks on stdin." — Ankur
+
 - **API mode dropped — Draftwise runs only inside coding agents now.** Every command (`scan`, `explain`, `new`, `tech`, `tasks`, `init`) used to support two execution paths: agent mode (print scanner data + instruction, let the host coding agent do the synthesis) and api mode (call Claude directly via `@anthropic-ai/sdk` and write the spec from the CLI). The api path is gone. `src/ai/provider.js`, `src/ai/providers/claude.js`, and the `@anthropic-ai/sdk` dependency are removed. `init` loses its `--ai-mode`, `--provider`, `--api-key-env`, `--stack`, and `--answers` flags plus the corresponding interactive prompts; `new`/`tech`/`tasks` lose `--force` and (for `new`) `--answers`; `scaffold` is unchanged. `config.yaml` schema simplifies — the `ai:` block (`mode`/`provider`/`api_key_env`/`model`/`max_tokens`) is no longer used; `loadConfig` returns only `projectState`/`stack`/`scanMaxFiles`. Existing configs with an `ai:` block keep working: `loadConfig` prints a one-line notice telling the user the block is safe to delete, then ignores it. The shared overwrite-guard and answers-flag utilities and the `src/ai/prompts/principles.js` + `src/ai/prompts/spec-quality.js` modules are removed — every prompt module now exports just `buildAgentInstruction` (or `AGENT_INSTRUCTION` for `scan`), since the host coding agent is the only consumer. CLAUDE.md and README.md updated end-to-end. — Ankur
 
 ### Changed

CLAUDE.md

@@ -40,7 +40,7 @@ src/index.js              → command router (dynamic imports, help)
 src/commands/             → one file per CLI command, default export async fn
 src/core/scanner.js       → codebase scanning (frameworks, routes, components, models)
 src/ai/prompts/           → one prompt module per command. Each exports a `buildAgentInstruction(...)` (or `AGENT_INSTRUCTION` constant for `scan`) that the host coding agent reads — section structure, hard rules, save path. No SDK call from the CLI; the agent does the synthesis.
-src/utils/                → config.js (yaml loader; returns projectState/stack/scanMaxFiles), specs.js (list .draftwise/specs/), slug.js, overview.js (read .draftwise/overview.md for greenfield context), scan-cache.js (fingerprinted scan cache, drop-in for scan()), flow-filter.js (narrow scan to flow-relevant items), scan-warnings.js (truncation + missing-framework messages), fs.js (shared pathExists), scan-projection.js (shared compactScan that trims a raw scan into a prompt-sized projection), scan-context.js (shared greenfield/brownfield branch for new/tech/tasks), draftwise-dir.js (`requireDraftwiseDir` guard), tty.js (isInteractive helper), agent-handoff.js (shared orienting prefix logged before every agent-mode handoff), project-state.js (filesystem auto-detect for `init` — bail-fast walk for source files using scanner's IGNORE_DIRS + CODE_EXTENSIONS), skill-providers.js (provider dir mapping + Claude-only frontmatter trim + `detectInstalledProviders` filesystem check shared across `skills install` / `uninstall` / `help`)
+src/utils/                → config.js (yaml loader; returns projectState/stack/scanMaxFiles), specs.js (list .draftwise/specs/), slug.js, overview.js (read .draftwise/overview.md for greenfield context), scan-cache.js (fingerprinted scan cache, drop-in for scan()), flow-filter.js (narrow scan to flow-relevant items), scan-warnings.js (truncation + missing-framework messages), fs.js (shared pathExists), scan-projection.js (shared compactScan that trims a raw scan into a prompt-sized projection), scan-context.js (shared greenfield/brownfield branch for new/tech/tasks), draftwise-dir.js (`requireDraftwiseDir` guard), agent-handoff.js (shared orienting prefix logged before every agent-mode handoff), project-state.js (filesystem auto-detect for `init` — bail-fast walk for source files using scanner's IGNORE_DIRS + CODE_EXTENSIONS), skill-providers.js (provider dir mapping + Claude-only frontmatter trim + `detectInstalledProviders` filesystem check shared across `skills install` / `uninstall` / `help`)
 test/                     → vitest, mirrors src structure
 .claude-plugin/           → plugin marketplace declaration (see "Claude Code plugin" below)
 plugin/                   → plugin source tree shipped via the marketplace
@@ -65,10 +65,10 @@ The single most important module is `src/core/scanner.js` — it parses the user
 - **eslint + prettier** for code style
 - **YAML** for config (`yaml` package)
 - **Markdown** for all spec documents
-- **`@inquirer/prompts`** for interactive prompts (init's idea prompt, tech/tasks spec picker, scaffold's confirm) — strictly the TTY-only convenience layer; flags drive the canonical input path (see "How input works" below).
-- **No AI SDK.** Draftwise no longer calls models from the CLI. Every command prints scanner data plus an instruction for the host coding agent (Claude Code, Cursor, Gemini CLI, etc.), which does the synthesis using its own model.
+- **No interactive prompts.** Every command takes its input via flags or positional args. Missing required input → error with a usage hint, or (for `draftwise init` greenfield without `--idea`) a structured agent handoff. No `@inquirer/prompts`, no `node:readline` — the CLI never blocks on stdin.
+- **No AI SDK.** Draftwise doesn't call models from the CLI. Every command prints scanner data plus an instruction for the host coding agent (Claude Code, Cursor, Gemini CLI, etc.), which does the synthesis using its own model.
 
-**On dependency pinning:** `@inquirer/prompts` and `yaml` use caret ranges because they're stable 1.x+ packages where minor bumps follow semver. Dependabot (`.github/dependabot.yml`) opens PRs for both.
+**On dependency pinning:** `yaml` uses a caret range because it's a stable 2.x package where minor bumps follow semver. Dependabot (`.github/dependabot.yml`) opens PRs for it.
 
 No TypeScript for v1 — keep it simple. May migrate later if the codebase grows.
 
@@ -104,7 +104,7 @@ scan:
 
 **Conversation lives in the host agent.** The CLI is a non-conversational executor: it loads context (scanner data or greenfield plan), prints an instruction, and exits. The host coding agent walks the PM through clarifying questions in chat, then writes the spec to disk. `draftwise new` doesn't run a Q&A loop from the CLI itself — its instruction tells the agent to.
 
-**Flags drive input; inquirer is a TTY-only fallback.** Every command takes its full input set as flags (`--mode`, `--idea`, `--yes`, etc.) parsed via Node's built-in `util.parseArgs`. When stdin is a TTY and a required value is missing, inquirer fires to fill it in — that's the only place inquirer lives. When stdin is not a TTY (CI, coding-agent shell), most commands error with a specific usage hint instead of hanging on a prompt; `draftwise init` is special — when it can't proceed without asking the user something (greenfield without `--idea`), it prints a structured **agent handoff** (the question in chat-friendly format + a re-invocation template, all under `AGENT_HANDOFF_PREFIX`) and exits cleanly, so the host coding agent reads stderr, asks the user in chat, and re-invokes with the collected flag. TTY-detection helper in `src/utils/tty.js`; tests opt into either path via `deps.isInteractive`.
+**Flags drive input; no interactive prompts.** Every command takes its full input set as flags (`--mode`, `--idea`, `--yes`) or positional args, parsed via Node's built-in `util.parseArgs`. Missing required input errors with a specific usage hint — the CLI never blocks waiting for stdin. `draftwise init` is the one special case: greenfield without `--idea` prints a structured **agent handoff** (the question in chat-friendly format + a re-invocation template, all under `AGENT_HANDOFF_PREFIX`) and exits cleanly, so the host coding agent reads stderr, asks the user in chat, and re-invokes with the collected flag. A plain-terminal user reads the same handoff as a usage hint.
 
 **Single repo, single feature spec at a time.** No cross-spec dependency tracking. No multi-repo. Keep scope tight.
 
@@ -127,7 +127,7 @@ draftwise skills uninstall [--provider=...] [--scope=...]           → remove s
 draftwise skills help                                                → list known harnesses + install state
 ```
 
-Each command is a separate file under `src/commands/` with a single `export default async function(args, deps = {}) {}`. The `deps` object is the dependency-injection seam used by tests — `cwd`, `log`, `scan`, `loadConfig`, `readOverview`, `isInteractive`, and per-command prompt overrides.
+Each command is a separate file under `src/commands/` with a single `export default async function(args, deps = {}) {}`. The `deps` object is the dependency-injection seam used by tests — `cwd`, `log`, `scan`, `loadConfig`, `readOverview`, `listSpecs`, etc.
 
 ---
 
@@ -216,7 +216,7 @@ Don't try to make v1 universal. A great experience for JS/TS is better than a me
 
 ## Conventions
 
-- **File per command.** `src/commands/<name>.js` with `export default async function(args, deps = {}) {}`. The `deps` argument is how tests inject `cwd`, `log`, `scan`, `loadConfig`, `readOverview`, `isInteractive`, and prompts.
+- **File per command.** `src/commands/<name>.js` with `export default async function(args, deps = {}) {}`. The `deps` argument is how tests inject `cwd`, `log`, `scan`, `loadConfig`, `readOverview`, `listSpecs`, etc.
 - **Async/await everywhere.** No `.then()` chains.
 - **Console output for CLI feedback.** Plain text for now — colored output (kleur/chalk) is deferred until there's a need.
 - **Errors bubble up to `src/index.js`.** It catches and prints a friendly message, then exits non-zero.

README.md

@@ -236,7 +236,7 @@ Claude Code · GitHub Copilot · Cursor · Gemini CLI · Codex CLI · Antigravit
 
 In practice it has only been smoke-tested in a plain terminal so far. If you run it inside one of these and something breaks, please open an issue.
 
-**Non-interactive use.** Every command runs without a TTY — pass values as flags instead of letting inquirer prompt for them. That's what makes the agent integration possible: a slash-command wrapper (or any host agent) can collect answers in chat and re-invoke `draftwise <command>` with `--mode=...`, `--idea="..."`, `--yes`, etc. Run `draftwise <command> --help` for the per-command flag list. When `draftwise init` is run in a non-TTY shell as greenfield without `--idea`, it prints a structured handoff with the question to ask the user — copy it into your AI assistant if you're not already inside one.
+**Flags-driven, never blocks.** Every command takes its input via flags or positional args; missing required input errors with a usage hint instead of waiting on stdin. That's what makes the agent integration possible: a slash-command wrapper (or any host agent) collects answers in chat and re-invokes `draftwise <command>` with `--mode=...`, `--idea="..."`, `--yes`, etc. Run `draftwise <command> --help` for the per-command flag list. `draftwise init` greenfield without `--idea` is the one case that prints a structured handoff instead of erroring — copy it into your AI assistant if you're not already inside one.
 
 ---