vibespot 1.0.9 → 1.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -25,18 +25,20 @@ npx vibespot
25
25
 
26
26
  Opens a browser with:
27
27
  - **Chat on the left** — describe your landing page in natural language
28
- - **Live preview on the right** — see your page render in real-time
28
+ - **Live preview on the right** — see your page render in real-time, plus a **Plan tab** and a **Code tab**
29
29
  - **Agentic pipeline** — multi-stage AI generation with real-time progress
30
+ - **Plan mode** *(new in v1.1)* — toggle on to deliberate before generating: the AI asks elicitation questions, builds a markdown plan in a dedicated pane, and only generates after you approve
30
31
  - **Project sidebar** — create, open, resume, or delete projects
31
32
  - **Module management** — reorder via drag-and-drop, edit fields, delete modules from module list or module library
32
33
  - **Starter templates** — SaaS, Portfolio, Restaurant, Event
33
- - **GitHub import** — convert existing React projects
34
+ - **From Figma** *(Beta)* paste a Figma URL to extract design tokens, text, and assets, then generate a full page that translates the design verbatim
35
+ - **From React** *(Beta)* — convert existing React/Lovable projects from a Git URL
34
36
  - **Field editor** — tweak text, colors, images directly
35
37
  - **File uploads** — attach images and documents via drag-and-drop or paperclip button
36
38
  - **Upload to HubSpot** — per-file progress, auto-fix, celebration popup with direct portal link
37
39
  - **Version history** — per-template git commits with rollback
38
40
  - **Light/dark mode** — toggle or auto-detect system preference
39
- - **Tabbed settings** — AI engines, HubSpot accounts, GitHub, vibeSpot config
41
+ - **Tabbed settings** — AI engines (with extended-thinking toggle), HubSpot accounts, Figma, GitHub, vibeSpot config
40
42
  - **ZIP download** — export your theme as a ZIP file
41
43
 
42
44
  ### Agentic Pipeline
@@ -50,6 +52,50 @@ When you send a message, vibeSpot runs a 4-stage pipeline:
50
52
 
51
53
  Completed modules appear in the live preview immediately as each finishes, with themed skeleton placeholders for modules still generating.
52
54
 
55
+ ### Plan Mode
56
+
57
+ Plan mode is a deliberation phase that runs *before* any code is generated. Toggle it from the prominent **Plan Mode** pill above the chat input. While active:
58
+
59
+ - The AI asks elicitation questions to surface gaps in your brief: who's the audience, what's the primary CTA, what sections should the page have, what's the brand voice.
60
+ - Each response builds up a markdown plan, persisted to `{theme}/.vibespot/plan.md` and rendered in a dedicated **Plan** tab in the right pane.
61
+ - The AI may emit clickable answer chips for common questions (e.g. "What's your primary goal? [Lead capture] [Sign-ups] [Demo bookings]") to keep the conversation fast.
62
+ - You can **edit the plan inline** (pencil icon → textarea → Save) — refinements without typing chat messages.
63
+ - Generation is **hard-gated**: the agentic pipeline is refused while plan mode is on. The only way to generate is the explicit **Approve plan** button, which prepends the plan to the user message as a design brief and runs the pipeline against the approved spec.
64
+ - **Discard & start over** clears the plan and exits plan mode.
65
+
66
+ Plan mode is ideal when:
67
+ - You're starting from a vague brief and want the AI to help you scope the page.
68
+ - Multiple stakeholders need to review the plan before code is generated.
69
+ - You're generating a high-stakes page and want to control content/structure decisions explicitly.
70
+
71
+ When plan mode is off, vibeSpot generates immediately — the existing fast path. Mode state persists across sessions.
72
+
73
+ ### Figma Import (Beta)
74
+
75
+ Translate a Figma design into a HubSpot CMS theme without re-implementing it. The pipeline preserves the design's exact colors, typography, copy, and section order — the AI is only used to translate each section to HubL, not to make creative decisions.
76
+
77
+ **Setup:**
78
+ 1. Open **Settings → Figma** and paste a Figma Personal Access Token (generate one in your Figma account settings).
79
+ 2. Click **From Figma** on the setup screen and paste a Figma file URL or a specific frame URL.
80
+
81
+ **What the importer extracts:**
82
+ - **Design tokens** — colors (with usage counts), typography styles (family, size, weight per role), spacing scale, effects (shadows, border radii). These map mechanically to `:root` CSS variables — no AI guessing.
83
+ - **Section structure** — top-level frames become modules, in their original Figma order.
84
+ - **Per-section content** — exact headlines, body copy, and CTAs are extracted with role and font-size annotations, fed to the AI as field defaults.
85
+ - **Image assets** — downloaded from Figma's image API as PNGs (2× resolution).
86
+
87
+ **Two image modes (toggle on the import screen):**
88
+ - **Import images as assets** *(default)* — images land in `{theme}/assets/` and are referenced via `get_asset_url()`. The page looks identical to Figma out of the box.
89
+ - **Image fields with placeholders** — modules use HubSpot image fields with placehold.co defaults. Content editors swap in their own images via the HubSpot editor.
90
+
91
+ **The conversion pipeline (streamlined):**
92
+ 1. **Generate shared CSS** — deterministic mapping of design tokens to CSS variables and utility classes. No AI.
93
+ 2. **Map sections to module specs** — each section's exact text and layout becomes a module specification.
94
+ 3. **AI translation** — for each module, the AI converts the Figma section into HubL. Parallel, concurrency-limited. The system prompt forbids inventing content; everything comes from Figma.
95
+ 4. **Quality check** — the same auto-fix layer used by the agentic pipeline.
96
+
97
+ The result is a HubSpot theme where every color, word, and pixel of spacing matches the original design — fully editable in HubSpot's drag-and-drop editor.
98
+
53
99
  ### Classic Wizard Mode
54
100
 
55
101
  ```bash
@@ -122,17 +168,28 @@ Most users only need `npx vibespot` — the web UI handles everything.
122
168
 
123
169
  ## Configuration
124
170
 
125
- Settings are managed in the **Settings** panel (tabbed: AI, HubSpot, GitHub, vibeSpot) and saved in `~/.vibespot/config.json`:
171
+ Settings are managed in the **Settings** panel (tabbed: AI, HubSpot, Figma, GitHub, vibeSpot) and saved in `~/.vibespot/config.json`:
126
172
 
127
173
  - `aiEngine` — Your preferred AI engine (`anthropic-api`, `claude-oauth`, `openai-api`, `gemini-api`, `claude-code`, `gemini-cli`, `codex-cli`)
128
174
  - `anthropicApiKey`, `openaiApiKey`, `geminiApiKey` — API keys (stored locally, never sent except to the provider)
129
175
  - `hubspotAccounts` — Connected HubSpot accounts (PAK-based auth)
130
176
  - `hubspotUploadMode` — `api` (default, direct API) or `cli` (legacy, requires HubSpot CLI)
131
177
  - `agenticConcurrency` — Max parallel module generation calls (default: 20)
178
+ - `figmaToken` — Figma Personal Access Token for design import
132
179
  - `enabledCLITools` — Which CLI tools to detect on settings load
133
180
 
134
- ## What's New (v1.0)
181
+ ## What's New
182
+
183
+ ### v1.1.0
184
+ - **Plan mode** — deliberation phase before generation. Prominent toggle above chat input; AI asks elicitation questions, builds a markdown plan in a dedicated tab, only generates after Approve. Plan persists to `.vibespot/plan.md` across sessions.
185
+ - **Streamlined Figma import** — replaces the old agentic-pipeline path with a translation pipeline that preserves Figma's exact design tokens, copy, and section order. Image assets copy into the theme automatically.
186
+ - **Anthropic SDK upgrade** (0.39 → 0.91) — unlocks extended thinking, improved prompt caching, and tool/schema-level cache control.
187
+ - **AI Capabilities settings panel** — toggle Extended Thinking (Anthropic API engines, with Low/Medium/High budget) and Web Search (Anthropic API + Claude Code CLI). Prompt Caching shows as auto-active; Citations is the remaining "coming soon" item. Claude Code engines see honest "Auto (CLI-managed)" badges where the CLI handles the feature internally.
188
+ - **Cache control on tool definitions** — Module Developer's identical tool schema is now cached, saving schema-encoding tokens on every parallel call after the first.
189
+ - **Claude Code stream-json output** — both single-call mode and the agentic CLI path now use `--output-format stream-json --include-partial-messages --verbose`. Tool calls (`Read`, `Edit`, `Bash`, `WebSearch`, `WebFetch`, `Grep`, `Glob`) flow into the pipeline status pane as concrete activity ("Searching: '…'", "Reading hero/module.html") instead of generic rotating placeholders.
135
190
 
191
+ ### v1.0.x
192
+ - **Figma design import** (v1.0.10) — paste a Figma URL to extract design tokens, text, section structure, and image assets, then generate a full HubSpot page via the agentic pipeline
136
193
  - **Template & module deletion** (v1.0.4) — delete templates from disk, option to delete exclusive modules, delete module button in module library preview
137
194
  - **Brand assets redesign** (v1.0.2) — hover-expand cards with per-asset Upload/Extract, Extract All, brand voice extractor, cross-template product context sharing via rendered preview HTML
138
195
  - **Agentic pipeline** (v1.0.0) — 4-stage AI generation: Intent Analyzer → Page Architect (Design System + Module Planner) → Module Developer (parallel) → Quality Check (auto-fix)