Drop API mode

CHANGELOG.md

@@ -6,9 +6,13 @@ Each released version is tagged in git (`v0.0.1`, `v0.1.0`, etc.) and includes t
 
 ## [Unreleased]
 
+### Removed
+
+- **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
 
-- **`draftwise skills install` auto-detects which AI harnesses are on the machine instead of installing to all three by default.** Previously the default behavior was to write SKILL.md into every known harness's user-level skill dir (`~/.claude/skills/draftwise/`, `~/.cursor/skills/draftwise/`, `~/.gemini/skills/draftwise/`) regardless of whether the user actually had Claude Code, Cursor, or Gemini CLI installed. Now `detectInstalledProviders` (in `src/utils/skill-providers.js`) checks which of `~/.claude` / `~/.cursor` / `~/.gemini` exist at the chosen scope root and installs only to those. The detected set is logged so the user sees *why* a particular harness was picked. `--provider=all` is the explicit opt-in for the old behavior; `--provider=<name>` still targets one harness regardless of detection. When detection finds nothing the command errors with a hint pointing at both override flags. `skills help` now also prints "Detected harnesses (user scope): …" and the project-scope equivalent so the auto-detect set is visible without having to run install. Why: this is the same behavior impeccable's install uses (via the `vercel-labs/skills` package — auto-detect, with `--all` as the explicit override) and it's friendlier than littering provider dirs with files for harnesses the user doesn't have. CLAUDE.md's "Standalone skill" section and README's slash-command callout updated to match. `skills uninstall` keeps its existing "iterate every known dir and skip ones with nothing to remove" behavior — different goal (clean up stale Draftwise installs whether or not the harness is still on disk), so detection-on-uninstall would miss the cleanup case. — Ankur
+- **`draftwise skills install` auto-detects which AI harnesses are on the machine instead of installing to all three by default.** Previously the default behavior was to write SKILL.md into every known harness's user-level skill dir (`~/.claude/skills/draftwise/`, `~/.cursor/skills/draftwise/`, `~/.gemini/skills/draftwise/`) regardless of whether the user actually had Claude Code, Cursor, or Gemini CLI installed. Now `detectInstalledProviders` (in `src/utils/skill-providers.js`) checks which of `~/.claude` / `~/.cursor` / `~/.gemini` exist at the chosen scope root and installs only to those. The detected set is logged so the user sees *why* a particular harness was picked. `--provider=all` is the explicit opt-in for the old behavior; `--provider=<name>` still targets one harness regardless of detection. When detection finds nothing the command errors with a hint pointing at both override flags. `skills help` now also prints "Detected harnesses (user scope): …" and the project-scope equivalent so the auto-detect set is visible without having to run install. Why: auto-detect with `--provider=all` as an explicit override is friendlier than littering provider dirs with files for harnesses the user doesn't have. CLAUDE.md's "Standalone skill" section and README's slash-command callout updated to match. `skills uninstall` keeps its existing "iterate every known dir and skip ones with nothing to remove" behavior — different goal (clean up stale Draftwise installs whether or not the harness is still on disk), so detection-on-uninstall would miss the cleanup case. — Ankur
 
 ## [0.2.1] — 2026-04-29 — Ankur
 
@@ -26,7 +30,7 @@ After merge, ritual: `git tag -a v0.2.1 -m "v0.2.1"`, `git push origin v0.2.1`,
 
 ### Added
 
-- **`draftwise skills <install|uninstall|help>` — standalone slash-command skill across Claude Code, Cursor, and Gemini CLI.** Three subcommands grouped under `skills` (the `git remote <sub>` / `gh pr <sub>` / `impeccable skills <sub>` pattern): `install` writes `plugin/skills/draftwise/` (SKILL.md + per-verb references) into each known harness's user-level skill dir (`~/.claude/skills/draftwise/`, `~/.cursor/skills/draftwise/`, `~/.gemini/skills/draftwise/`); `uninstall` removes them; `help` prints a 3×2 state table (3 harnesses × 2 scopes). `--provider=claude|cursor|gemini|all` narrows; `--scope=user|project` switches between `~` and `<cwd>`; `--force` overwrites on install. Why: (1) Claude Code's marketplace plugin path forces a `<plugin>:<skill>` namespace prefix on every plugin skill regardless of `/plugin install` scope ([anthropics/claude-code#15882](https://github.com/anthropics/claude-code/issues/15882), closed: not planned), so the marketplace install gives users `/draftwise:draftwise <verb>` instead of `/draftwise <verb>` — a standalone SKILL.md written directly into `.claude/skills/` has no plugin manifest and resolves bare. (2) Same SKILL.md works across harnesses with a tiny per-provider frontmatter trim (Claude-only `user-invocable` / `argument-hint` / `allowed-tools` stripped for Cursor + Gemini); body identical. Same model impeccable uses (`npx impeccable skills install` writes to ~11 provider dirs). The marketplace plugin is independent and may coexist; both forms appear side-by-side in Claude Code's slash menu. Conflict detection lists every existing target in one error (no half-installs). Subcommand routing lives in `src/index.js` via a small `SUBCOMMAND_GROUPS` table; provider mapping + frontmatter trim in `src/utils/skill-providers.js`; commands at `src/commands/skills/{install,uninstall,help}.js`. `package.json` `files` ships `plugin/skills/` so the install can resolve its source from the npm install. Tests inject temp `home`, `cwd`, and `sourceDir` to keep the real `~/.claude/` (and equivalents) untouched. Plugin SKILL.md learns `skills` as a verb routing to `reference/skills.md` so the marketplace plugin's chat-driven flow handles `skills install` / `skills uninstall` / `skills help` natively. — Ankur
+- **`draftwise skills <install|uninstall|help>` — standalone slash-command skill across Claude Code, Cursor, and Gemini CLI.** Three subcommands grouped under `skills` (the `git remote <sub>` / `gh pr <sub>` pattern): `install` writes `plugin/skills/draftwise/` (SKILL.md + per-verb references) into each known harness's user-level skill dir (`~/.claude/skills/draftwise/`, `~/.cursor/skills/draftwise/`, `~/.gemini/skills/draftwise/`); `uninstall` removes them; `help` prints a 3×2 state table (3 harnesses × 2 scopes). `--provider=claude|cursor|gemini|all` narrows; `--scope=user|project` switches between `~` and `<cwd>`; `--force` overwrites on install. Why: (1) Claude Code's marketplace plugin path forces a `<plugin>:<skill>` namespace prefix on every plugin skill regardless of `/plugin install` scope ([anthropics/claude-code#15882](https://github.com/anthropics/claude-code/issues/15882), closed: not planned), so the marketplace install gives users `/draftwise:draftwise <verb>` instead of `/draftwise <verb>` — a standalone SKILL.md written directly into `.claude/skills/` has no plugin manifest and resolves bare. (2) Same SKILL.md works across harnesses with a tiny per-provider frontmatter trim (Claude-only `user-invocable` / `argument-hint` / `allowed-tools` stripped for Cursor + Gemini); body identical. The marketplace plugin is independent and may coexist; both forms appear side-by-side in Claude Code's slash menu. Conflict detection lists every existing target in one error (no half-installs). Subcommand routing lives in `src/index.js` via a small `SUBCOMMAND_GROUPS` table; provider mapping + frontmatter trim in `src/utils/skill-providers.js`; commands at `src/commands/skills/{install,uninstall,help}.js`. `package.json` `files` ships `plugin/skills/` so the install can resolve its source from the npm install. Tests inject temp `home`, `cwd`, and `sourceDir` to keep the real `~/.claude/` (and equivalents) untouched. Plugin SKILL.md learns `skills` as a verb routing to `reference/skills.md` so the marketplace plugin's chat-driven flow handles `skills install` / `skills uninstall` / `skills help` natively. — Ankur
 
 ### Changed
 
@@ -62,7 +66,7 @@ The "Mode 1, for real" release. Draftwise now ships a Claude Code plugin so PMs
 
 ### Added
 
-- **Claude Code plugin: `/draftwise` slash commands.** New `.claude-plugin/marketplace.json` at repo root declares a single `draftwise` plugin with `source: ./plugin`. Inside `plugin/` lives the install manifest (`.claude-plugin/plugin.json`) plus a single skill `skills/draftwise/SKILL.md` that routes user input to per-verb references at `skills/draftwise/reference/<verb>.md` (one per CLI verb: init, new, scan, explain, tech, tasks, list, show, scaffold). Each reference walks the model through how to drive that verb in chat — collect inputs, shell out to the npm-installed `draftwise` CLI, parse the structured handoff or streamed output, report back. Pattern mirrors impeccable: 1 skill / N commands / shells out to the underlying CLI. Users install via `/plugin marketplace add 4nkur/draftwise` then `/plugin install draftwise` in Claude Code; the npm CLI install (`npm i -g draftwise`) is the prerequisite. Plugin name = skill name = `draftwise` so the slash form is `/draftwise <verb>` (no `<plugin>:<skill>` namespacing in command logs). Plugin is distributed separately from the npm package — `package.json` `files` does not include the plugin directories. SKILL.md includes a "Setup gates" table making the verb-dependency chain explicit (`init` is bootstrap; `tech` requires a product spec; `tasks` requires a tech spec; etc.) and a "Conversation standards" section that pulls in the eight collaboration principles from `src/ai/prompts/principles.js` by reference, so the chat-driven flow matches what the CLI's api-mode synthesis enforces. Per-verb references add pre-flight gate checks (catch missing prerequisites in chat instead of letting the CLI throw), tone-shaping guidance for ambiguous flag questions, idea-concreteness check for `new` (one-word ideas get one elaboration ask before invoking), and review nudges for `tech` / `tasks`. README's Quick Start gains a 3-line install snippet for the plugin path. Closes #42. — Ankur
+- **Claude Code plugin: `/draftwise` slash commands.** New `.claude-plugin/marketplace.json` at repo root declares a single `draftwise` plugin with `source: ./plugin`. Inside `plugin/` lives the install manifest (`.claude-plugin/plugin.json`) plus a single skill `skills/draftwise/SKILL.md` that routes user input to per-verb references at `skills/draftwise/reference/<verb>.md` (one per CLI verb: init, new, scan, explain, tech, tasks, list, show, scaffold). Each reference walks the model through how to drive that verb in chat — collect inputs, shell out to the npm-installed `draftwise` CLI, parse the structured handoff or streamed output, report back. Pattern: 1 skill / N commands / shells out to the underlying CLI. Users install via `/plugin marketplace add 4nkur/draftwise` then `/plugin install draftwise` in Claude Code; the npm CLI install (`npm i -g draftwise`) is the prerequisite. Plugin name = skill name = `draftwise` so the slash form is `/draftwise <verb>` (no `<plugin>:<skill>` namespacing in command logs). Plugin is distributed separately from the npm package — `package.json` `files` does not include the plugin directories. SKILL.md includes a "Setup gates" table making the verb-dependency chain explicit (`init` is bootstrap; `tech` requires a product spec; `tasks` requires a tech spec; etc.) and a "Conversation standards" section that pulls in the eight collaboration principles from `src/ai/prompts/principles.js` by reference, so the chat-driven flow matches what the CLI's api-mode synthesis enforces. Per-verb references add pre-flight gate checks (catch missing prerequisites in chat instead of letting the CLI throw), tone-shaping guidance for ambiguous flag questions, idea-concreteness check for `new` (one-word ideas get one elaboration ask before invoking), and review nudges for `tech` / `tasks`. README's Quick Start gains a 3-line install snippet for the plugin path. Closes #42. — Ankur
 
 ### Changed