Merged
Conversation
Replace ~80 user-facing string references across 20 files: - Landing pages (navbar, hero, footer, features) - App metadata (layout title) - Auth pages (login, signup, OAuth, electron callback) - Chat UI (project-chat placeholder/buttons, studio advisor) - Dashboard (nav, quick-start, MCP connection, local LLM) - Content pages (docs, download, setup, addons, project detail) - CSS design system comments NOT changed (deferred for separate migration): - window.modelforge Electron IPC API - modelforge-addon.py download URLs - --forge-* CSS variable prefix
…ename - Electron bridge: preload.js key, electron.d.ts interface/Window prop - Frontend consumers: login-form.tsx, electron-auth-listener.tsx, setup/page.tsx - Desktop app: main.js env vars, protocol, title, HTML auth pages, addon paths - Desktop package.json: name, appId, productName, copyright, dev script - Addon: renamed modelforge-addon.py -> vipermesh-addon.py (desktop/assets + public/downloads) - Addon internals: bl_info, User-Agent, panel classes, operator bl_idnames, sidebar category - Download URLs: docs/page.tsx, quick-start-card.tsx - Build verified: exit code 0, zero remaining modelforge refs in code
…ero panel - Position viper illustration behind the hero StudioPreview panel - Add clipped overlay copy (left half) that renders above the panel - Create 3D weaving illusion: snake body goes under panel then emerges over it - Move panel down (translateY: 100px) so snake curve clears above panel edge - Fine-tune image size (185%), position (bottom: -95%, marginLeft: -90%) - Add teal drop-shadow and fade-in animation to viper illustration
… and tooltips - Replace generic hexagon icons with custom hand-crafted SVGs: Geometry (stacked layers), Texture (UV grid), Materials (shaded sphere), Lighting (sun rays), Animate (waveform) - Leave 2nd box empty (covered by snake overlay) - Add cursor-following tooltip showing tool description on hover - Add hover scale and shadow effects on tool boxes - Remove drop-shadow filter from snake image for performance - Add useState import for tooltip tracking
- Add activeTool state to track selected tool (defaults to Geometry) - Clicking a tool box highlights it with teal accent border and background - Previously selected tool deactivates automatically - Empty box (snake-covered) is not clickable
- Replace ModelForge branding with ViperMesh globally - Add documentation for the new premium Teal design system - Add documentation for the Autopilot and Studio dual modes - Document the new Hybrid Neural 3D Pipeline (Hunyuan3D, TRELLIS 2, YVO3D) - Update Quick Start and Roadmap to reflect current state
- Replace Hammer icon with AnimatedLogo on login page - Replace Hammer icon with AnimatedLogo on signup page - Replace old 'M' gradient square with ViperMesh logo on OAuth start page - Replace old 'M' gradient square with ViperMesh logo on electron callback page - Remove unused Hammer import from dashboard-nav
Renamed ~50 ModelForge references to ViperMesh in 13 doc files: - docs/future-plans.md, architecture.md, architecture-notes.md - docs/3d-pipeline-strategy.md, 3d-pipeline-integration.md - docs/HANDOFF.md, research-pipeline-techniques.md, test-prompts.md - docs/addon-integration-roadmap.md - SETUP.md, SECURITY.md, deploy/runpod/README.md, gemini.md Zero remaining ModelForge references in the entire project.
…reation) - Test 17: multi-object fruit market stall with complex spatial relationships, directional sun + fill lighting, 7+ material types, camera framing - Test 18: image-reference recreation using generated reading nook reference (tests vision analysis → Blender object translation) - Added note marking Tests 14-16 as future-only (require neural models)
- Hide per-message TOOL CALLS during active streaming (only AgentActivity shows live) - Thinking indicator only appears when agent is reasoning, not during tool calls - AgentActivity collapses into closed details dropdown when done - MCP execution summary now uses details element (closed by default) - Expanded friendlyToolMap (7 -> 27 entries) - Follow-up prompt uses grouped human-readable descriptions instead of raw tool names - Fallback message uses grouped labels - System prompt branding: ModelForge -> ViperMesh
…oning guide - New section: Proximity & Contact Relationships - 'Leaning against' / 'Beside' (surface contact with margin) - 'Resting on' / 'Sitting on' (vertical contact) - 'Inside a container' (containment bounds) - 'Clustered' (minimum separation checks) - Added scaled dimension awareness (radius × scale) - Added safety margin reference table - Added 3 new common mistakes (scale-aware offsets, surface-to-surface, intersection prevention) - Exact watermelon/banana example from Test 17 bug - Re-ingested 16 guides into vectorstore
- Add optional referenceImage input to blender-agent-shape in tool-catalog - Wire functional file upload in studio-workspace.tsx (was decorative only) - File input reads image as base64 data URL - Shows thumbnail preview with remove button - Forward image attachments in executeStep via studio-layout.tsx - Extracts image data from tool inputs as attachments array - Sends to /api/ai/chat in same format as Autopilot mode - Add image attachment button to follow-up chat in step-session-drawer - Image icon button opens file picker - Shows pending image preview chip with remove - Sends attachments through onSendMessage callback
- route.ts: a.data from FileReader is already a full data URL (data:image/png;base64,...) — was wrapping it again causing Gemini to reject with 'Base64 decoding failed' - step-session-drawer.tsx: show reference image thumbnail below the prompt text in the session drawer
- route.ts: use a.data directly (already a full data URL from FileReader) instead of wrapping with another data: prefix — fixes Gemini 400 error - studio-layout.tsx: strip base64 image data from inputs before persisting steps to database (can be megabytes) - step-session-drawer.tsx: detect data URLs in message content and render as <img> instead of raw base64 text; show reference image thumbnail below prompt
Image fixes: - route.ts: use a.data directly (already full data URL) — fixes Gemini 400 - studio-layout.ts: strip base64 from persisted inputs (saves DB payload) - step-session-drawer.tsx: detect data URLs in messages, render as <img> RAG fix: - agents.ts: detect short follow-up messages like 'try again' and use the original prompt for vector search instead of the vague text
- step-session-drawer.tsx: AgentActivity now renders in both running AND completed states (uses built-in <details> collapse). commandResults section only shown as legacy fallback when no agentEvents exist. - blender-agent-system.md: explicit instruction to describe WHAT was built (objects, materials, composition) instead of listing tool counts
…ement - lighting-guide.md: added LIGHT-EMITTING OBJECT GEOMETRY section with open shade rule, emission materials, point light placement checklist. Expanded tags for better RAG retrieval (lamp, candle, lantern, etc). - render-guide.md: expanded film_transparent section with when-to-use vs when-NOT-to-use guidance (interior/exterior scenes = false). - blender-agent-system.md: viewport screenshot is now MANDATORY at 3 checkpoints (geometry, lighting, pre-render). Clarified that render_image does NOT return visual data. - Re-ingested all 16 tool guides into vectorstore.
…, and use Gemini embeddings instead of Together.ai
github-actions
bot
added
documentation
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (9)
📝 WalkthroughWalkthroughRebrands ModelForge → ViperMesh across docs, web UI, desktop/electron, and prompts; adds image-attachment support end-to-end, RAG follow-up detection, interactive hero/navbar UI, new AnimatedLogo and LineShadowText components, studio image upload flow, and several UI/agent behavior adjustments. Changes
Sequence Diagram(s)sequenceDiagram
participant User as User
participant UI as Step Drawer UI
participant FileAPI as FileReader
participant Chat as /api/ai/chat
participant LLM as LLM
User->>UI: Select image file
UI->>FileAPI: readAsDataURL(file)
FileAPI-->>UI: data:image/...
UI->>UI: store pendingImage
User->>UI: Send message + image
UI->>Chat: POST /api/ai/chat (message, attachments)
Chat->>LLM: Process message + image data URL
LLM-->>Chat: response (tool calls / streaming)
Chat-->>UI: stream response
UI->>UI: clear pendingImage, render result
sequenceDiagram
participant User as User
participant ChatComp as Chat Component
participant Agent as Agent Middleware
participant RAG as RAG Search
participant LLM as LLM
User->>ChatComp: Send short/vague follow-up
ChatComp->>Agent: processMessage()
Agent->>Agent: detect short follow-up (RETRY_PATTERNS)
Agent->>Agent: select bestQuery from prior messages
Agent->>RAG: search(bestQuery)
RAG-->>Agent: context
Agent->>LLM: inject context + prompt
LLM-->>Agent: response
Agent-->>ChatComp: deliver response
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Suggested labels
Poem
🚥 Pre-merge checks | ❌ 3❌ Failed checks (2 warnings, 1 inconclusive)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 20
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (8)
app/globals.css (1)
28-44: 🧹 Nitpick | 🔵 TrivialConsider renaming CSS variables for brand consistency.
The comment is updated to "ViperMesh Design System", but the CSS custom properties still use the
--forge-*prefix (e.g.,--forge-accent,--forge-surface). While this is an internal implementation detail that doesn't affect users, updating the variable names to align with the new brand (e.g.,--viper-*or--vm-*) would improve codebase consistency.This is non-blocking as it would require coordinated changes across multiple files that reference these variables.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@app/globals.css` around lines 28 - 44, Update the CSS custom property names to match the new brand by renaming each --forge-* variable (e.g., --forge-accent, --forge-accent-hover, --forge-surface, --forge-border, --forge-text, --forge-glass, --forge-shadow, etc.) to a consistent ViperMesh prefix such as --viper-* or --vm-* and then update every reference across the codebase to use the new names to avoid breakage (search for occurrences of the specific symbols --forge-accent, --forge-accent-hover, --forge-accent-subtle, --forge-accent-muted, --forge-surface, --forge-surface-dim, --forge-surface-raised, --forge-border, --forge-border-strong, --forge-text, --forge-text-muted, --forge-text-subtle, --forge-glass, --forge-glass-border, --forge-shadow, --forge-shadow-lg and replace them with the chosen --viper- or --vm- equivalents).lib/orchestration/tool-catalog.ts (1)
1-6:⚠️ Potential issue | 🟡 MinorMissed branding update in JSDoc comment.
The file-level JSDoc comment still references "ModelForge tools" on line 2, which is inconsistent with the PR's rebranding to "ViperMesh".
Proposed fix
/** - * Tool Catalog — Central knowledge base for all ModelForge tools & categories. + * Tool Catalog — Central knowledge base for all ViperMesh tools & categories. * * Provides beginner-friendly metadata, input definitions, and descriptions * for every tool and pipeline category in the platform. */🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@lib/orchestration/tool-catalog.ts` around lines 1 - 6, Update the file-level JSDoc at the top of lib/orchestration/tool-catalog.ts to use the new branding: change the phrase "ModelForge tools" to "ViperMesh tools" (and any other "ModelForge" mentions in that same JSDoc block) so the comment reads "Tool Catalog — Central knowledge base for all ViperMesh tools & categories." Locate the top-of-file JSDoc comment (the block describing "Tool Catalog") and make the textual replacement to reflect the rebranding.SECURITY.md (1)
73-74:⚠️ Potential issue | 🟡 MinorUpdate SECURITY.md to reflect Supabase Auth migration.
Line 73 incorrectly references NextAuth.js, which has been fully removed from the project. The codebase now uses Supabase Auth as confirmed in
docs/architecture.md(line 137) and evidenced by Supabase imports throughout the authentication layer.📝 Suggested fix
* Password hashing using bcryptjs with appropriate salt rounds -* CSRF protection via NextAuth.js +* CSRF protection via Supabase Auth * SQL injection prevention through Prisma ORM🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@SECURITY.md` around lines 73 - 74, Replace the outdated reference to NextAuth.js in SECURITY.md with Supabase Auth: change the bullet "CSRF protection via NextAuth.js" to something like "CSRF protection via Supabase Auth (session and token handling provided by Supabase)" and ensure the document references the current auth implementation (see docs/architecture.md and Supabase imports in the auth layer) so the security entry accurately reflects the migration.SETUP.md (1)
26-26: 🧹 Nitpick | 🔵 TrivialDatabase name still references
modelforgein example connection strings.Several example connection strings use
modelforgeas the database name (lines 26, 91-92, 220, 223). While the database name doesn't need to match the app name, this creates inconsistency with the rebrand. Consider updating tovipermeshfor consistency, or add a note that the database name is configurable.📝 Proposed fix for line 26
# Database - Update with your PostgreSQL credentials -DATABASE_URL="postgresql://user:password@localhost:5432/modelforge" +DATABASE_URL="postgresql://user:password@localhost:5432/vipermesh"Apply similar changes to lines 91, 92, 220, and 223.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@SETUP.md` at line 26, Update the example PostgreSQL connection strings that still use the database name "modelforge" (specifically the DATABASE_URL example and the other examples referenced on lines 26, 91-92, 220, 223) to use the new brand-consistent database name "vipermesh" or alternatively add a short note that the database name is configurable; locate and update the DATABASE_URL example and any other explicit "modelforge" occurrences in SETUP.md to either replace them with "vipermesh" or append a parenthetical like "(replace with your database name)" to make it clear the name is configurable.desktop/package.json (1)
106-110:⚠️ Potential issue | 🟠 MajorPublish metadata still points at the pre-rebrand GitHub repo.
Line 109 still targets
ModelForge, but this PR is inKer102/ViperMesh. Electron Builder and updater metadata will be generated against the wrong GitHub repository.Suggested fix
"publish": { "provider": "github", "owner": "Ker102", - "repo": "ModelForge" + "repo": "ViperMesh" }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@desktop/package.json` around lines 106 - 110, Update the "publish" metadata in the package.json publish block so the GitHub target points to the new repo: change the "repo" value in the "publish" object (the keys "publish", "provider", "owner", "repo") from "ModelForge" to "ViperMesh" so Electron Builder and auto-update metadata are generated against Ker102/ViperMesh.app/api/ai/chat/route.ts (2)
1068-1080:⚠️ Potential issue | 🟠 MajorDon't overwrite the already-streamed assistant text in the empty fallback branch.
If the follow-up model returns zero chunks,
assistantText = emptyFallbackdiscards any response that was already streamed earlier in the request. The user saw that content live, but the Prisma record will only keep the fallback sentence after reload. Append the fallback the same way as the non-empty branch instead of replacing the buffer.Suggested fix
} else { // The LLM yielded 0 chunks — send a scene-aware fallback console.warn("[Chat] Follow-up stream returned empty text, sending scene-aware fallback") const sceneSummary = groupedLabels.length > 0 ? `I've set up your scene — ${groupedLabels.slice(0, 3).join(", ")}${groupedLabels.length > 3 ? ` and more` : ""}.` : "I've set up the scene." const emptyFallback = overallSuccess ? `${sceneSummary} What would you like to adjust or add next?` : `Some steps didn't go as planned. Would you like me to retry with a different approach?` - send({ type: "followup_delta", delta: emptyFallback }) - assistantText = emptyFallback + const fallbackDelta = assistantText.trim() + ? `\n\n${emptyFallback}` + : emptyFallback + send({ type: "followup_delta", delta: fallbackDelta }) + assistantText = assistantText.trim() + ? `${assistantText.trimEnd()}\n\n${emptyFallback}` + : emptyFallback }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@app/api/ai/chat/route.ts` around lines 1068 - 1080, The empty-fallback branch currently replaces any already-streamed assistant text by doing assistantText = emptyFallback; instead, append the fallback like the non-empty branch to preserve previously streamed content: build the appended text using (assistantText || "").trimEnd() + "\n\n" + emptyFallback and assign that to assistantText, and continue sending the followup_delta via send({ type: "followup_delta", delta: emptyFallback }); refer to assistantText, followUpText, groupedLabels and overallSuccess in the existing branch to locate and update the logic.
799-809:⚠️ Potential issue | 🟠 MajorAdd bounds and format validation for image attachments before forwarding to the provider.
The attachment schema (line 270) accepts a
datafield with no size, count, or MIME type constraints. After the full JSON body is materialized viareq.json(), attachments are forwarded directly at lines 804–809 without validation. Add:
- Maximum attachment count limit
- Maximum decoded image size (e.g., 5–20 MB per image)
- MIME type validation (restrict to
data:image/*;base64,...)- Total payload cap to prevent memory exhaustion before reaching the provider
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@app/api/ai/chat/route.ts` around lines 799 - 809, Validate and sanitize image attachments before building messageContent: before the mapping that uses attachments and a.data (the hasImageAttachments branch using agentPrompt and attachments.map), enforce a max attachments count (e.g., 5), validate each a.data matches the data URL MIME pattern /^data:image\/[a-zA-Z0-9.+-]+;base64,/, decode and check each image's decoded size does not exceed the per-image limit (e.g., 5–20MB) and track a running total to enforce a global payload cap to avoid memory exhaustion; if any check fails, short-circuit with a clear 4xx error response instead of forwarding, and only build messageContent (the array used for provider forwarding) after all attachments pass validation.README.md (1)
162-177:⚠️ Potential issue | 🟡 MinorFinish the Quick Start rebrand in the clone step.
This installation section was updated for ViperMesh, but the clone command still points users at
ModelForgeand tells them tocdinto that old directory name. Update the repository path and folder name here so the onboarding flow is internally consistent.Also applies to: 186-188
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@README.md` around lines 162 - 177, Update the README clone/install steps to use the new ViperMesh branding: change the git clone URL and the cd target from "ModelForge" to "ViperMesh" in the clone step shown around the current block (the lines with "git clone https://github.com/Ker102/ModelForge.git" and "cd ModelForge") and also update the identical occurrence referenced at lines ~186-188 so both places consistently point to "https://github.com/Ker102/ViperMesh.git" and "cd ViperMesh".
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@app/auth/start-oauth/page.tsx`:
- Around line 54-57: Replace the native img tag in the start-oauth page
component with Next.js's optimized Image: add an import for Image from
"next/image" at the top of the component file and swap the <img ... /> element
inside the div (the element rendering "/vipermesh-logo-transparent.png") for an
<Image> usage that supplies width and height (64 each) and retains the
"object-contain" styling and alt text; ensure the file remains a client
component so the Image import is allowed.
In `@components/landing/hero.tsx`:
- Around line 184-205: The tool tiles rendered in tools.map should be real
buttons to provide keyboard focus and activation: replace the outer <div> used
in the mapping with a <button type="button"> while preserving existing behavior
from setActiveTool, handleMouseMove, setTooltip, and tool.icon; add
aria-pressed={isActive} to expose selected state to assistive tech; keep the
conditional cursor/hover classes but also add matching focus/focus-visible
styles so keyboard focus mirrors hover (e.g., same
scale/shadow/border/background changes when focused); ensure the key prop
(tool.name || `empty-${idx}`) remains and that click activation continues to
call setActiveTool(tool.name).
In `@components/landing/navbar.tsx`:
- Around line 66-94: The CTAs use motion.button inside Link which creates nested
interactive elements; change both motion.button usages (the "Log in" and "Get
Started" CTA blocks) to non-interactive anchors by rendering motion.a (or using
Link with an <a> child) instead of motion.button, keep the same className,
motion props (whileHover/whileTap) and the inner motion.div for the animated
sheen, and add an appropriate aria-label if the text isn't sufficient;
alternatively wrap Link around a non-interactive motion.div and put the anchor
inside, but do not nest a <button> inside Link to resolve the
accessibility/focus issues.
In `@components/projects/agent-activity.tsx`:
- Around line 146-208: The collapsed “Done” branch currently treats any terminal
non-active state the same and hides failures; update the conditional rendering
around isActive so terminal failure cases are handled separately: keep the
existing Done summary only when !isActive && completedCount > 0, add a new
branch for !isActive && completedCount === 0 && failedCount > 0 (or derive
failed entries from allCompleted/allFailed) that renders a failure summary and
lists failed tools (use a different ToolIcon status like "failed" and list from
allFailed), and for mixed results when !isActive && failedCount > 0 &&
completedCount > 0 render a combined summary that shows both success and failure
lists; ensure these branches short-circuit so the live “Agent working” header
never renders after terminal failure.
In `@components/projects/project-chat.tsx`:
- Around line 1481-1493: The current logic builds a Set<string> (startedTools)
from agentEvents and toggles presence by tool name, which fails when multiple
overlapping calls for the same tool exist; replace the Set with a Map<string,
number> (or a per-tool counter) and iterate agentEvents to increment the count
for a tool on a "started" event and decrement on completion/failure/finished
events, removing the entry when the count reaches zero; then compute
hasActiveTool by checking if any map value > 0 (or the total sum > 0) and keep
isThinking = agentActive && !hasActiveTool, using the existing agentEvents,
startedTools (renamed), hasActiveTool and isThinking identifiers to locate where
to change the logic.
In `@components/projects/studio-layout.tsx`:
- Around line 36-49: The current slimming logic in the steps.map (variable slim)
strips data:image/... payloads by replacing them with "[image attached]", which
loses the actual image needed to rebuild attachments and breaks
StepSessionDrawer previews and calls to /api/ai/chat; update the logic in the
map callback to either (A) persist a durable attachment reference/thumbnail
instead of the raw base64 (e.g., store an attachments handle/id or small
thumbnail string alongside inputs so the image can be rehydrated later), or (B)
if no durable handle exists, mark the restored step as non-runnable (add a flag
like inputs._runnable = false or step.meta.runnable = false) so callers
(StepSessionDrawer and the code that posts to /api/ai/chat) know not to attempt
to use a missing image; ensure the code that reads slim can find the new
attachment handle/thumbnail key and that existing callers check the runnable
flag before sending images.
- Around line 402-403: The complete event handler in executeStep closes over a
stale workflowSteps causing appendCommandResult to mis-handle batch results; fix
by introducing a ref (e.g., workflowStepsRef) that is updated whenever
workflowSteps changes and read workflowStepsRef.current inside the "complete"
handler (instead of direct workflowSteps) so the closure always sees the latest
state; update the ref in the same component where workflowSteps is defined (use
useRef and a useEffect or assignment on render) and leave executeStep's
dependency array as-is while replacing reads of workflowSteps with
workflowStepsRef.current (also update any related checks such as hasLiveResults
to use the ref).
In `@components/projects/studio-workspace.tsx`:
- Around line 457-465: Add file size validation and FileReader error handling to
the image upload handler: before creating the FileReader, check
e.target.files?.[0].size against a defined MAX_IMAGE_BYTES (e.g., 2MB) and
early-return with a user-visible error (set state or call a provided error
handler) if too large; attach reader.onerror to handle read failures (set the
same error state and avoid calling setInputs) and keep reader.onload to
setInputs({ ...inputs, [input.key]: reader.result as string }) only on
successful read; ensure any error messages reference the input key or context
for clarity.
- Around line 403-469: The image file input isn't being reset after removal so
re-selecting the same file may not fire onChange; update the hidden <input> (the
one with type="file" and onChange that uses FileReader and setInputs) to clear
its value on click (e.g., add an onClick handler that sets e.currentTarget.value
= ''), and ensure the remove button that calls setInputs({ ...inputs,
[input.key]: "" }) remains as-is so the input is truly cleared both in state and
DOM before the next selection.
In `@data/tool-guides/spatial-positioning-guide.md`:
- Around line 145-164: The markdown around the "CRITICAL: Scaled Dimensions"
section and following code blocks needs whitespace and fenced-block fixes to
satisfy the linter: add a blank line before and after the "### CRITICAL: Scaled
Dimensions" and the "### \"Leaning Against\" / \"Beside\" (Surface Contact)"
headings, ensure the fenced code blocks containing the lines "actual_extent_x =
R × sx (NOT just R)" and "half_x = (S / 2) × sx" are fenced with ```python
(language specifier) and have a blank line both before the opening fence and
after the closing fence, and add blank lines around any inline tables or other
block elements (e.g., the "ALWAYS multiply dimensions..." paragraph) so the
section meets MD022, MD031/MD040, and MD058 requirements.
In `@desktop/assets/vipermesh-addon.py`:
- Around line 21-30: The bl_info metadata still points doc_url to the old
ModelForge repo; update bl_info["doc_url"] (in the bl_info dict) to the renamed
project's URL (the ViperMesh repo/homepage) so the addon details link in Blender
points to the correct, current project documentation. Ensure the string value is
replaced with the new canonical project URL.
In `@docs/test-prompts.md`:
- Around line 411-435: The test references
docs/test18-reading-nook-reference.png but never attaches it, so upload or embed
that reference image and update the prompt to point to the included asset;
specifically, add the image file into the repo (or specify an uploaded tester
artifact) and change the prompt block to include the image inline (e.g.,
markdown image tag referencing docs/test18-reading-nook-reference.png) and add a
short sentence that the tester will supply the image as input so the agent can
call get_viewport_screenshot/use vision — update the test verification text
accordingly to state the agent will receive the image and therefore exercise the
vision path.
- Around line 370-425: The markdown for the new "### Test 17: Multi-Object Scene
with Complex Spatial Relationships & Lighting" and "### Test 18: Image Reference
Recreation (Vision + Spatial)" sections needs lint fixes: add a blank line
before and after each heading and before/after every fenced code block, and
annotate the fences with a language (use ```text for the prompt blocks) so
MD022/MD031/MD040 are resolved; update the fenced blocks in both Test 17 and
Test 18 (the prompt/code fences shown in the diff) to include the language spec
and ensure there's an empty line separating the fences from surrounding
paragraphs/headings.
In `@gemini.md`:
- Around line 138-139: Move the two completed items ("Added glassmorphism
floating pill navbar with magnetic hover pills and spring CTA animations." and
"Added LineShadowText component (Magic UI) to '3D Models' hero text with teal
shadow.") out of the "Future Implementation Plans" section and append them to
the correct session log entry above; locate the "Future Implementation Plans"
heading and the nearest preceding session log entry in gemini.md and place these
lines under that session log so they reflect completed work rather than future
plans.
- Around line 3-10: Fix the Markdown heading spacing by ensuring there is a
blank line before and after each top-level or subheading mentioned (e.g., the
"Last Session: 2026-03-24 (Recent Changes)" heading, the "What Was Done"
subheading, and other headings that follow like the sections around "README
Updates", "What Was Done" items, and subsequent section headers); insert a
single empty line above each heading and a single empty line between the heading
and the paragraph or list that follows so the document conforms to markdownlint
MD022 and MD024 style rules.
In `@lib/ai/agents.ts`:
- Around line 1102-1128: The current logic marks any short lastContent as a
retry and then picks the longest prior human message (RETRY_PATTERNS,
isShortFollowUp, allHumanMessages, bestQuery), which can pull unrelated context;
change isShortFollowUp so it only triggers for explicit retry patterns or truly
negligible edits (e.g. match RETRY_PATTERNS OR lastContent.length < a very small
threshold) rather than every short message, and when isShortFollowUp is true,
iterate messages in reverse (or reverse allHumanMessages) to select the most
recent prior human message whose content length > 30 and that does not itself
match RETRY_PATTERNS, then set content to that recent substantive message
instead of the longest one.
In `@lib/orchestration/prompts/blender-agent-system.md`:
- Around line 13-17: The few-shot examples in blender-agent-system.md contradict
the "Visual Confirmation — MANDATORY" rule; update each example sequence so
every example calls get_viewport_screenshot after creating main
geometry/objects, after setting up lighting, and immediately before calling
render_image, removing any flows that skip those checkpoints and ensuring the
examples mirror the exact phrasing and ordering of the rule (referencing
get_viewport_screenshot and render_image) so no counterexamples remain.
In `@public/downloads/vipermesh-addon.py`:
- Around line 21-30: Update the bl_info dictionary entry doc_url in
vipermesh-addon.py so it no longer points to the old ModelForge repo; locate the
bl_info symbol and change the "doc_url" value to the correct ViperMesh
documentation or repo URL (e.g., the ViperMesh GitHub or docs page) to keep
branding consistent.
- Line 1438: The requests.get() calls to the Poly Haven API currently have no
timeout and can hang Blender; update each call to include an appropriate
timeout: add timeout=30 for lightweight API endpoints (e.g., the call in
response = requests.get(f"https://api.polyhaven.com/categories/{asset_type}",
...) and the functions search_polyhaven_assets and related API queries) and use
timeout=60 for larger file downloads (e.g., download_polyhaven_asset, the
texture download, model download, and include file download functions) so
network requests fail fast and don’t block the UI.
In `@README.md`:
- Around line 52-58: The new markdown headings (e.g., "### 🌐 Premium Web
Dashboard UI") lack surrounding blank lines causing markdownlint MD022; add a
single blank line before and after each heading and the same for the other
flagged headings mentioned (lines containing "Dual Modes", "Teal Design System",
"Supabase Auth" and the sections at 66-76 and 94-99) so each heading is
separated from preceding and following content and the MD022 violations are
resolved.
---
Outside diff comments:
In `@app/api/ai/chat/route.ts`:
- Around line 1068-1080: The empty-fallback branch currently replaces any
already-streamed assistant text by doing assistantText = emptyFallback; instead,
append the fallback like the non-empty branch to preserve previously streamed
content: build the appended text using (assistantText || "").trimEnd() + "\n\n"
+ emptyFallback and assign that to assistantText, and continue sending the
followup_delta via send({ type: "followup_delta", delta: emptyFallback }); refer
to assistantText, followUpText, groupedLabels and overallSuccess in the existing
branch to locate and update the logic.
- Around line 799-809: Validate and sanitize image attachments before building
messageContent: before the mapping that uses attachments and a.data (the
hasImageAttachments branch using agentPrompt and attachments.map), enforce a max
attachments count (e.g., 5), validate each a.data matches the data URL MIME
pattern /^data:image\/[a-zA-Z0-9.+-]+;base64,/, decode and check each image's
decoded size does not exceed the per-image limit (e.g., 5–20MB) and track a
running total to enforce a global payload cap to avoid memory exhaustion; if any
check fails, short-circuit with a clear 4xx error response instead of
forwarding, and only build messageContent (the array used for provider
forwarding) after all attachments pass validation.
In `@app/globals.css`:
- Around line 28-44: Update the CSS custom property names to match the new brand
by renaming each --forge-* variable (e.g., --forge-accent, --forge-accent-hover,
--forge-surface, --forge-border, --forge-text, --forge-glass, --forge-shadow,
etc.) to a consistent ViperMesh prefix such as --viper-* or --vm-* and then
update every reference across the codebase to use the new names to avoid
breakage (search for occurrences of the specific symbols --forge-accent,
--forge-accent-hover, --forge-accent-subtle, --forge-accent-muted,
--forge-surface, --forge-surface-dim, --forge-surface-raised, --forge-border,
--forge-border-strong, --forge-text, --forge-text-muted, --forge-text-subtle,
--forge-glass, --forge-glass-border, --forge-shadow, --forge-shadow-lg and
replace them with the chosen --viper- or --vm- equivalents).
In `@desktop/package.json`:
- Around line 106-110: Update the "publish" metadata in the package.json publish
block so the GitHub target points to the new repo: change the "repo" value in
the "publish" object (the keys "publish", "provider", "owner", "repo") from
"ModelForge" to "ViperMesh" so Electron Builder and auto-update metadata are
generated against Ker102/ViperMesh.
In `@lib/orchestration/tool-catalog.ts`:
- Around line 1-6: Update the file-level JSDoc at the top of
lib/orchestration/tool-catalog.ts to use the new branding: change the phrase
"ModelForge tools" to "ViperMesh tools" (and any other "ModelForge" mentions in
that same JSDoc block) so the comment reads "Tool Catalog — Central knowledge
base for all ViperMesh tools & categories." Locate the top-of-file JSDoc comment
(the block describing "Tool Catalog") and make the textual replacement to
reflect the rebranding.
In `@README.md`:
- Around line 162-177: Update the README clone/install steps to use the new
ViperMesh branding: change the git clone URL and the cd target from "ModelForge"
to "ViperMesh" in the clone step shown around the current block (the lines with
"git clone https://github.com/Ker102/ModelForge.git" and "cd ModelForge") and
also update the identical occurrence referenced at lines ~186-188 so both places
consistently point to "https://github.com/Ker102/ViperMesh.git" and "cd
ViperMesh".
In `@SECURITY.md`:
- Around line 73-74: Replace the outdated reference to NextAuth.js in
SECURITY.md with Supabase Auth: change the bullet "CSRF protection via
NextAuth.js" to something like "CSRF protection via Supabase Auth (session and
token handling provided by Supabase)" and ensure the document references the
current auth implementation (see docs/architecture.md and Supabase imports in
the auth layer) so the security entry accurately reflects the migration.
In `@SETUP.md`:
- Line 26: Update the example PostgreSQL connection strings that still use the
database name "modelforge" (specifically the DATABASE_URL example and the other
examples referenced on lines 26, 91-92, 220, 223) to use the new
brand-consistent database name "vipermesh" or alternatively add a short note
that the database name is configurable; locate and update the DATABASE_URL
example and any other explicit "modelforge" occurrences in SETUP.md to either
replace them with "vipermesh" or append a parenthetical like "(replace with your
database name)" to make it clear the name is configurable.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: 0d282faf-dcc4-4b36-8203-e2fe38cd5673
⛔ Files ignored due to path filters (9)
Modify_wireframe_opacity_202603230424.jpegis excluded by!**/*.jpegModify_wireframe_opacity_202603230425.jpegis excluded by!**/*.jpegdocs/test18-reading-nook-reference.pngis excluded by!**/*.pngpublic/images/c815e7cf-ee13-48ff-a8a9-2189ac48a424.pngis excluded by!**/*.pngpublic/vipermesh-logo-image.jpegis excluded by!**/*.jpegpublic/vipermesh-logo-transparent.pngis excluded by!**/*.pngpublic/vipermesh-logo-v1.svgis excluded by!**/*.svgpublic/vipermesh-logo-v2.svgis excluded by!**/*.svgpublic/vipermesh-logo.svgis excluded by!**/*.svg
📒 Files selected for processing (58)
README.mdSECURITY.mdSETUP.mdapp/addons/page.tsxapp/api/ai/chat/route.tsapp/auth/electron-callback/page.tsxapp/auth/start-oauth/page.tsxapp/dashboard/projects/[id]/page.tsxapp/docs/page.tsxapp/download/page.tsxapp/globals.cssapp/layout.tsxapp/login/page.tsxapp/setup/page.tsxapp/signup/page.tsxcomponents/auth/electron-auth-listener.tsxcomponents/auth/login-form.tsxcomponents/dashboard/dashboard-nav.tsxcomponents/dashboard/local-llm-settings-card.tsxcomponents/dashboard/mcp-connection-card.tsxcomponents/dashboard/quick-start-card.tsxcomponents/landing/features.tsxcomponents/landing/footer.tsxcomponents/landing/hero.tsxcomponents/landing/navbar.tsxcomponents/projects/agent-activity.tsxcomponents/projects/project-chat.tsxcomponents/projects/step-session-drawer.tsxcomponents/projects/studio-advisor.tsxcomponents/projects/studio-layout.tsxcomponents/projects/studio-workspace.tsxcomponents/ui/animated-logo.tsxcomponents/ui/line-shadow-text.tsxdata/tool-guides/lighting-guide.mddata/tool-guides/render-guide.mddata/tool-guides/spatial-positioning-guide.mddeploy/runpod/README.mddesktop/assets/vipermesh-addon.pydesktop/main.jsdesktop/package.jsondesktop/preload.jsdocs/3d-pipeline-integration.mddocs/3d-pipeline-strategy.mddocs/HANDOFF.mddocs/addon-integration-roadmap.mddocs/architecture-notes.mddocs/architecture.mddocs/future-plans.mddocs/research-pipeline-techniques.mddocs/test-prompts.mdgemini.mdlib/ai/agents.tslib/orchestration/prompts/blender-agent-system.mdlib/orchestration/prompts/index.tslib/orchestration/tool-catalog.tslib/types/electron.d.tspublic/downloads/vipermesh-addon.pytailwind.config.ts
Comment on lines
+54
to
+57
| <div className="w-16 h-16 mx-auto flex items-center justify-center"> | ||
| <img src="/vipermesh-logo-transparent.png" alt="ViperMesh" className="w-16 h-16 object-contain" /> | ||
| </div> | ||
| <h1 className="mt-4 text-xl font-semibold text-white">ModelForge</h1> | ||
| <h1 className="mt-4 text-xl font-semibold text-white">ViperMesh</h1> |
There was a problem hiding this comment.
🧹 Nitpick | 🔵 Trivial
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Description: Check that the ViperMesh logo file exists in public directory
fd -t f "vipermesh-logo-transparent.png" publicRepository: Ker102/ViperMesh
Length of output: 97
🏁 Script executed:
head -60 app/auth/start-oauth/page.tsxRepository: Ker102/ViperMesh
Length of output: 2490
Consider using Next.js Image component for optimized loading.
The native <img> tag works, but doesn't leverage Next.js image optimization. The file is already a client component, so using the Image component from next/image is straightforward. Update the import and replace the img tag:
import Image from "next/image"
// ...
<Image src="/vipermesh-logo-transparent.png" alt="ViperMesh" width={64} height={64} className="object-contain" />🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@app/auth/start-oauth/page.tsx` around lines 54 - 57, Replace the native img
tag in the start-oauth page component with Next.js's optimized Image: add an
import for Image from "next/image" at the top of the component file and swap the
<img ... /> element inside the div (the element rendering
"/vipermesh-logo-transparent.png") for an <Image> usage that supplies width and
height (64 each) and retains the "object-contain" styling and alt text; ensure
the file remains a client component so the Image import is allowed.
Comment on lines
184
to
+205
| <div className="grid grid-cols-6 gap-1.5 mb-4"> | ||
| {tools.map((tool) => ( | ||
| <div key={tool.name} | ||
| className="aspect-square rounded-lg border flex flex-col items-center justify-center gap-1" | ||
| {tools.map((tool, idx) => { | ||
| const isActive = tool.name === activeTool | ||
| return ( | ||
| <div key={tool.name || `empty-${idx}`} | ||
| className={`aspect-square rounded-lg border flex flex-col items-center justify-center gap-1 ${tool.desc ? 'cursor-pointer transition-all duration-200 hover:scale-105 hover:shadow-md' : ''}`} | ||
| style={{ | ||
| borderColor: tool.active ? "hsl(var(--forge-accent))" : "hsl(var(--forge-border))", | ||
| backgroundColor: tool.active ? "hsl(var(--forge-accent-subtle))" : "hsl(var(--forge-surface-dim))", | ||
| }}> | ||
| <svg width="16" height="16" viewBox="0 0 24 24" fill="none" | ||
| stroke={tool.active ? "hsl(var(--forge-accent))" : "hsl(var(--forge-text-subtle))"} strokeWidth="1.5"> | ||
| <path d="M12 3l8 4.5v9L12 21l-8-4.5v-9L12 3z" /> | ||
| </svg> | ||
| <span className="text-[8px]" style={{ | ||
| color: tool.active ? "hsl(var(--forge-accent))" : "hsl(var(--forge-text-subtle))", | ||
| }}>{tool.name}</span> | ||
| borderColor: isActive ? "hsl(var(--forge-accent))" : "hsl(var(--forge-border))", | ||
| backgroundColor: isActive ? "hsl(var(--forge-accent-subtle))" : "hsl(var(--forge-surface-dim))", | ||
| }} | ||
| onClick={() => tool.name && setActiveTool(tool.name)} | ||
| onMouseMove={(e) => handleMouseMove(e, tool.desc)} | ||
| onMouseLeave={() => setTooltip(null)} | ||
| > | ||
| {tool.icon(isActive ? "hsl(var(--forge-accent))" : "hsl(var(--forge-text-subtle))")} | ||
| {tool.name && ( | ||
| <span className="text-[8px]" style={{ | ||
| color: isActive ? "hsl(var(--forge-accent))" : "hsl(var(--forge-text-subtle))", | ||
| }}>{tool.name}</span> | ||
| )} | ||
| </div> | ||
| ))} | ||
| ) |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
# First, let's examine the file and the specific lines mentioned
wc -l components/landing/hero.tsxRepository: Ker102/ViperMesh
Length of output: 91
🏁 Script executed:
# Read the relevant section with context
sed -n '175,210p' components/landing/hero.tsx | cat -nRepository: Ker102/ViperMesh
Length of output: 2271
🏁 Script executed:
# Check the entire tool rendering and state management
sed -n '180,220p' components/landing/hero.tsx | cat -nRepository: Ker102/ViperMesh
Length of output: 2670
🏁 Script executed:
# Look for any keyboard event handlers in the component
rg -A 5 -B 5 "onKey" components/landing/hero.tsxRepository: Ker102/ViperMesh
Length of output: 42
🏁 Script executed:
# Check for any aria attributes being used in the file
rg "aria-" components/landing/hero.tsxRepository: Ker102/ViperMesh
Length of output: 121
🏁 Script executed:
# Look at the parent grid and any keyboard handling context
sed -n '160,230p' components/landing/hero.tsx | cat -nRepository: Ker102/ViperMesh
Length of output: 4389
🏁 Script executed:
# Search for any keyboard event handlers that might apply to this section
rg -B 15 "tools.map" components/landing/hero.tsx | head -50Repository: Ker102/ViperMesh
Length of output: 689
🏁 Script executed:
# Check if the parent div has any role or keyboard delegation
rg "role=" components/landing/hero.tsxRepository: Ker102/ViperMesh
Length of output: 42
🏁 Script executed:
# Check for tabindex in the file
rg "tabIndex" components/landing/hero.tsxRepository: Ker102/ViperMesh
Length of output: 42
The selectable tool tiles need button semantics and keyboard support.
These tiles are interactive controls (clickable with onClick, managing activeTool state), but they're rendered as plain divs with no keyboard handlers or accessibility attributes. Keyboard users cannot focus, activate, or navigate these controls, and assistive tech gets no indication of the selected state. Convert them to button type="button" elements, add aria-pressed={isActive} to indicate selection, and ensure focus/blur styles mirror the hover behavior for visual consistency across input methods.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@components/landing/hero.tsx` around lines 184 - 205, The tool tiles rendered
in tools.map should be real buttons to provide keyboard focus and activation:
replace the outer <div> used in the mapping with a <button type="button"> while
preserving existing behavior from setActiveTool, handleMouseMove, setTooltip,
and tool.icon; add aria-pressed={isActive} to expose selected state to assistive
tech; keep the conditional cursor/hover classes but also add matching
focus/focus-visible styles so keyboard focus mirrors hover (e.g., same
scale/shadow/border/background changes when focused); ensure the key prop
(tool.name || `empty-${idx}`) remains and that click activation continues to
call setActiveTool(tool.name).
Comment on lines
66
to
94
| <Link href="/login"> | ||
| <Button | ||
| variant="ghost" | ||
| className="text-sm font-medium rounded-full px-5" | ||
| style={{ color: "hsl(var(--forge-text-muted))" }} | ||
| <motion.button | ||
| whileHover={{ scale: 1.05, backgroundColor: "rgba(0,0,0,0.05)" }} | ||
| whileTap={{ scale: 0.95 }} | ||
| className="text-sm font-medium rounded-full px-5 py-2.5 text-muted-foreground hover:text-foreground transition-colors" | ||
| > | ||
| Log in | ||
| </Button> | ||
| </motion.button> | ||
| </Link> | ||
| <Link href="/signup"> | ||
| <Button | ||
| className="text-sm font-semibold rounded-full px-6" | ||
| <motion.button | ||
| whileHover={{ scale: 1.05, filter: "brightness(1.1)" }} | ||
| whileTap={{ scale: 0.95 }} | ||
| className="text-sm font-semibold rounded-full px-6 py-2.5 relative overflow-hidden group" | ||
| style={{ | ||
| backgroundColor: "hsl(var(--forge-accent))", | ||
| color: "white", | ||
| boxShadow: "0 2px 8px hsl(168 75% 32% / 0.25)", | ||
| boxShadow: "0 4px 14px hsl(168 75% 32% / 0.3)", | ||
| }} | ||
| > | ||
| Get Started | ||
| </Button> | ||
| <motion.div | ||
| className="absolute inset-0 z-0 bg-white/20" | ||
| initial={{ x: "-100%" }} | ||
| whileHover={{ x: "100%" }} | ||
| transition={{ duration: 0.6, ease: "easeInOut" }} | ||
| /> | ||
| <span className="relative z-10">Get Started</span> | ||
| </motion.button> | ||
| </Link> |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
# First, let's check if the file exists and examine its content
fd "navbar.tsx" --type fRepository: Ker102/ViperMesh
Length of output: 89
🏁 Script executed:
# Get context around lines 66-94 in the navbar component
cat -n components/landing/navbar.tsx | sed -n '60,100p'Repository: Ker102/ViperMesh
Length of output: 1769
🌐 Web query:
Next.js Link component renders anchor tag button inside
💡 Result:
No, the Next.js Link component does not render an anchor tag containing a button element by default. The Link component renders as an <a> tag when its direct child is text or when explicitly structured that way, but its behavior depends on the children provided. Key points from official documentation: - Link "extends the HTML <a> element" and "renders to an <a> element" (App Router docs, Pages Router docs). - Example: <Link href="/dashboard">Dashboard</Link> outputs <a href="/dashboard">Dashboard</a>. - If the child is a non-string React element like <button>, Link returns the child directly without wrapping in <a> (confirmed in Next.js source code references and Stack Overflow analyses). Putting a <button> inside Link creates nested interactive elements (<a><button></a>), which is invalid HTML, harms accessibility (e.g., double focus on tab navigation), and causes unexpected behavior like both onClick and navigation firing. Solutions include: - Use e.preventDefault and e.stopPropagation on the button's onClick to block Link navigation. - Avoid nesting: Use Link for navigation semantics or router.push with prefetch for buttons. - For styled buttons that navigate, use libraries like shadcn/ui with asChild prop to compose properly without nesting. This applies to both App Router (Next.js 13+) and Pages Router, with App Router being current as of 2026.
Citations:
- 1: https://nextjs.org/docs/pages/api-reference/components/link
- 2: https://stackoverflow.com/questions/67911391/nextjs-link-isnt-rendering-an-anchor-tag
- 3: https://nextjs.org/docs/14/pages/api-reference/components/link
- 4: https://stackoverflow.com/questions/71422463/how-to-fix-unexpected-behavior-of-next-link-when-it-is-a-child-of-a-button
- 5: https://stackoverflow.com/questions/74015358/nextjs-button-inside-link-component
- 6: https://www.reddit.com/r/nextjs/comments/1ch018x/is_this_good_approach_to_wrap_button_with_link/
- 7: https://nextjs.org/docs/app/api-reference/components/link
- 8: https://nextjs.org/docs/13/app/building-your-application/routing/linking-and-navigating
🌐 Web query:
HTML5 interactive content model button inside anchor tag accessibility
💡 Result:
No, placing a button element inside an anchor (<a>) tag is not valid under the HTML5 content model and poses significant accessibility issues. The HTML5 specification defines the <a> element's content model as transparent but explicitly prohibits interactive content descendants, including <button> elements. This is confirmed in the official W3C HTML Standard: "Content model: Transparent, but there must be no interactive content descendant." Button is classified as interactive content. Accessibility problems arise because nesting interactive elements confuses screen readers and assistive technologies. Screen readers may fail to announce the inner button properly, create empty tab stops, or misinterpret roles and names, violating WCAG 2.1/2.2 criteria such as 4.1.2 Name, Role, Value (Level A) and potentially 2.1.1 Keyboard (A). Sources describe this as creating focus problems, interference in event handling, and poor support across browsers and AT. Workarounds like preventDefault on clicks exist but do not resolve validity or full a11y compliance; they are discouraged. Instead, use semantic alternatives: style links as buttons (role="button") for navigation or buttons with JS navigation for actions, ensuring no nesting.
Citations:
- 1: https://stackoverflow.com/questions/6393827/can-i-nest-a-button-element-inside-an-a-using-html5
- 2: https://www.w3.org/TR/html5/the-a-element.html
- 3: https://stackoverflow.com/questions/62092211/html-validation-why-is-it-not-valid-to-put-an-interactive-element-inside-an-int
- 4: https://accessibilitychecker.org/wcag-guides/ensure-interactive-controls-are-not-nested-as-they-are-not-always-announced-by-screen-readers-or-can-cause-focus-problems-for-assistive-technologies
- 5: https://lumar.io/wiki/accessibility/keyboard-a/nested-interactive
- 6: https://theodorusclarence.com/blog/nesting-button-inside-link
Don't nest button elements inside Link.
Both CTAs render as <a><button /></a>, which violates the HTML5 content model—the <a> element must not contain interactive content descendants. This breaks keyboard navigation and screen-reader behavior, creating focus problems and WCAG 2.1 compliance violations (4.1.2 Name, Role, Value; 2.1.1 Keyboard). Render these as styled links instead, or put the Link inside a non-interactive motion wrapper.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@components/landing/navbar.tsx` around lines 66 - 94, The CTAs use
motion.button inside Link which creates nested interactive elements; change both
motion.button usages (the "Log in" and "Get Started" CTA blocks) to
non-interactive anchors by rendering motion.a (or using Link with an <a> child)
instead of motion.button, keep the same className, motion props
(whileHover/whileTap) and the inner motion.div for the animated sheen, and add
an appropriate aria-label if the text isn't sufficient; alternatively wrap Link
around a non-interactive motion.div and put the anchor inside, but do not nest a
<button> inside Link to resolve the accessibility/focus issues.
Comment on lines
+146
to
+208
| // ── Collapsed summary when agent is done ── | ||
| if (!isActive && completedCount > 0) { | ||
| return ( | ||
| <details | ||
| className="rounded-xl border px-4 py-3 mb-3 transition-all duration-300" | ||
| style={{ | ||
| backgroundColor: "var(--forge-glass)", | ||
| borderColor: "hsl(var(--forge-border))", | ||
| backdropFilter: "blur(12px)", | ||
| boxShadow: "var(--forge-shadow)", | ||
| }} | ||
| > | ||
| <summary | ||
| className="cursor-pointer flex items-center gap-2 text-xs font-semibold uppercase tracking-wider select-none" | ||
| style={{ color: "hsl(var(--forge-text-subtle))" }} | ||
| > | ||
| <svg | ||
| width="14" | ||
| height="14" | ||
| viewBox="0 0 24 24" | ||
| fill="none" | ||
| stroke="hsl(150 60% 45%)" | ||
| strokeWidth="2.5" | ||
| strokeLinecap="round" | ||
| strokeLinejoin="round" | ||
| className="shrink-0" | ||
| > | ||
| <polyline points="20 6 9 17 4 12" /> | ||
| </svg> | ||
| Done — {completedCount} tool{completedCount !== 1 ? "s" : ""} used | ||
| </summary> | ||
| <div className="mt-2 space-y-0.5"> | ||
| {allCompleted.map((e, i) => ( | ||
| <div | ||
| key={`${e.toolName}-${e.timestamp}-${i}`} | ||
| className="flex items-center gap-2 py-1 px-2 rounded-lg opacity-60" | ||
| > | ||
| <ToolIcon status="completed" /> | ||
| <span | ||
| className="text-xs" | ||
| style={{ color: "hsl(var(--forge-text-muted))" }} | ||
| > | ||
| {e.label} | ||
| </span> | ||
| <svg | ||
| width="12" | ||
| height="12" | ||
| viewBox="0 0 24 24" | ||
| fill="none" | ||
| stroke="hsl(150 60% 45%)" | ||
| strokeWidth="3" | ||
| strokeLinecap="round" | ||
| strokeLinejoin="round" | ||
| className="ml-auto shrink-0" | ||
| > | ||
| <polyline points="20 6 9 17 4 12" /> | ||
| </svg> | ||
| </div> | ||
| ))} | ||
| </div> | ||
| </details> | ||
| ) | ||
| } |
There was a problem hiding this comment.
Handle terminal failures separately from the “done” path.
When isActive is false and completedCount === 0, this falls through to the live panel and still renders the pulsing “Agent working” header. Mixed success/failure runs also collapse to a success-only summary, so failed tools disappear once streaming stops.
Suggested fix
+ const failedCount = useMemo(
+ () => toolEvents.filter((e) => e.status === "failed").length,
+ [toolEvents]
+ )
+
- if (!isActive && completedCount > 0) {
+ if (!isActive && (completedCount > 0 || failedCount > 0)) {
return (
<details
className="rounded-xl border px-4 py-3 mb-3 transition-all duration-300"
style={{
backgroundColor: "var(--forge-glass)",
@@
<summary
className="cursor-pointer flex items-center gap-2 text-xs font-semibold uppercase tracking-wider select-none"
style={{ color: "hsl(var(--forge-text-subtle))" }}
>
@@
- Done — {completedCount} tool{completedCount !== 1 ? "s" : ""} used
+ {failedCount > 0
+ ? `Stopped — ${completedCount} completed, ${failedCount} failed`
+ : `Done — ${completedCount} tool${completedCount !== 1 ? "s" : ""} used`}
</summary>
@@
- <PulsingDot />
+ {isActive && <PulsingDot />}
<span
className="text-xs font-semibold uppercase tracking-wider"
style={{ color: "hsl(var(--forge-text-subtle))" }}
>
- Agent working
+ {isActive ? "Agent working" : "Agent stopped"}
</span>Also applies to: 222-229
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@components/projects/agent-activity.tsx` around lines 146 - 208, The collapsed
“Done” branch currently treats any terminal non-active state the same and hides
failures; update the conditional rendering around isActive so terminal failure
cases are handled separately: keep the existing Done summary only when !isActive
&& completedCount > 0, add a new branch for !isActive && completedCount === 0 &&
failedCount > 0 (or derive failed entries from allCompleted/allFailed) that
renders a failure summary and lists failed tools (use a different ToolIcon
status like "failed" and list from allFailed), and for mixed results when
!isActive && failedCount > 0 && completedCount > 0 render a combined summary
that shows both success and failure lists; ensure these branches short-circuit
so the live “Agent working” header never renders after terminal failure.
Comment on lines
+1481
to
+1493
| {(() => { | ||
| // Derive whether a tool call is currently in-flight | ||
| const startedTools = new Set<string>() | ||
| for (const evt of agentEvents) { | ||
| if (evt.type === "agent:tool_call") { | ||
| const tc = evt as unknown as { toolName: string; status: string } | ||
| if (tc.status === "started") startedTools.add(tc.toolName) | ||
| else startedTools.delete(tc.toolName) | ||
| } | ||
| } | ||
| const hasActiveTool = startedTools.size > 0 | ||
| const isThinking = agentActive && !hasActiveTool | ||
|
|
There was a problem hiding this comment.
Count active tool calls instead of toggling by tool name.
Lines 1483-1489 track in-flight work with a Set<string>. If two agent:tool_call events for the same tool overlap, the first completion removes the name and flips the UI back to “Thinking…” even though another call is still running.
Suggested fix
- const startedTools = new Set<string>()
+ const activeToolCounts = new Map<string, number>()
for (const evt of agentEvents) {
if (evt.type === "agent:tool_call") {
const tc = evt as unknown as { toolName: string; status: string }
- if (tc.status === "started") startedTools.add(tc.toolName)
- else startedTools.delete(tc.toolName)
+ if (tc.status === "started") {
+ activeToolCounts.set(tc.toolName, (activeToolCounts.get(tc.toolName) ?? 0) + 1)
+ } else {
+ const next = (activeToolCounts.get(tc.toolName) ?? 0) - 1
+ if (next > 0) activeToolCounts.set(tc.toolName, next)
+ else activeToolCounts.delete(tc.toolName)
+ }
}
}
- const hasActiveTool = startedTools.size > 0
+ const hasActiveTool = activeToolCounts.size > 0🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@components/projects/project-chat.tsx` around lines 1481 - 1493, The current
logic builds a Set<string> (startedTools) from agentEvents and toggles presence
by tool name, which fails when multiple overlapping calls for the same tool
exist; replace the Set with a Map<string, number> (or a per-tool counter) and
iterate agentEvents to increment the count for a tool on a "started" event and
decrement on completion/failure/finished events, removing the entry when the
count reaches zero; then compute hasActiveTool by checking if any map value > 0
(or the total sum > 0) and keep isThinking = agentActive && !hasActiveTool,
using the existing agentEvents, startedTools (renamed), hasActiveTool and
isThinking identifiers to locate where to change the logic.
Comment on lines
+1102
to
+1128
| // Detect short "retry" / follow-up messages that are too vague for RAG. | ||
| // If the last message is very short or matches retry patterns, fall back to | ||
| // the longest previous human message for a more meaningful search query. | ||
| const RETRY_PATTERNS = /^(try\s*again|redo|retry|again|re-?run|go\s*again|one\s*more|repeat|do\s*it\s*again)/i | ||
| const isShortFollowUp = lastContent.length < 30 || RETRY_PATTERNS.test(lastContent.trim()) | ||
|
|
||
| let content = lastContent | ||
| if (isShortFollowUp) { | ||
| // Find the longest previous human message as a better RAG query | ||
| const allHumanMessages = messages.filter((m) => isHumanMessage(m)) | ||
| let bestQuery = "" | ||
| for (const hm of allHumanMessages) { | ||
| const hmRec = hm as unknown as Record<string, unknown> | ||
| const hmContent = typeof hmRec.content === "string" | ||
| ? hmRec.content | ||
| : Array.isArray(hmRec.content) | ||
| ? (hmRec.content as Array<{ text?: string }>).map((c) => c.text ?? "").join("") | ||
| : "" | ||
| // Pick the longest human message that isn't itself a short follow-up | ||
| if (hmContent.length > bestQuery.length && hmContent.length > 30) { | ||
| bestQuery = hmContent | ||
| } | ||
| } | ||
| if (bestQuery) { | ||
| console.log(`[RAG] Follow-up detected ("${lastContent.slice(0, 40)}") — using original prompt for search (${bestQuery.length} chars)`) | ||
| content = bestQuery | ||
| } |
There was a problem hiding this comment.
Use the latest substantive prompt for retry fallback, not the longest one.
Line 1106 treats every short message as a retry, so legitimate edits like “add bevel” or “make it blue” lose their current intent. Then Lines 1111-1123 pick the longest prior human message, which can pull RAG context from an unrelated earlier task.
Suggested fix
- const isShortFollowUp = lastContent.length < 30 || RETRY_PATTERNS.test(lastContent.trim())
+ const isExplicitRetry = RETRY_PATTERNS.test(lastContent.trim())
let content = lastContent
- if (isShortFollowUp) {
- // Find the longest previous human message as a better RAG query
- const allHumanMessages = messages.filter((m) => isHumanMessage(m))
+ if (isExplicitRetry) {
+ // Walk backward to the nearest substantive human prompt
+ const allHumanMessages = messages.filter((m) => isHumanMessage(m)).slice(0, -1).reverse()
let bestQuery = ""
for (const hm of allHumanMessages) {
const hmRec = hm as unknown as Record<string, unknown>
const hmContent = typeof hmRec.content === "string"
? hmRec.content
: Array.isArray(hmRec.content)
? (hmRec.content as Array<{ text?: string }>).map((c) => c.text ?? "").join("")
: ""
- // Pick the longest human message that isn't itself a short follow-up
- if (hmContent.length > bestQuery.length && hmContent.length > 30) {
+ if (hmContent.length > 30 && !RETRY_PATTERNS.test(hmContent.trim())) {
bestQuery = hmContent
+ break
}
}
if (bestQuery) {
- content = bestQuery
+ content = `${bestQuery}\n\nFollow-up: ${lastContent}`
}
}🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@lib/ai/agents.ts` around lines 1102 - 1128, The current logic marks any short
lastContent as a retry and then picks the longest prior human message
(RETRY_PATTERNS, isShortFollowUp, allHumanMessages, bestQuery), which can pull
unrelated context; change isShortFollowUp so it only triggers for explicit retry
patterns or truly negligible edits (e.g. match RETRY_PATTERNS OR
lastContent.length < a very small threshold) rather than every short message,
and when isShortFollowUp is true, iterate messages in reverse (or reverse
allHumanMessages) to select the most recent prior human message whose content
length > 30 and that does not itself match RETRY_PATTERNS, then set content to
that recent substantive message instead of the longest one.
Comment on lines
+13
to
+17
| 5. **Visual Confirmation — MANDATORY**: You MUST call `get_viewport_screenshot` at these checkpoints: | ||
| - After creating the main geometry/objects (to verify placement, scale, and proportions) | ||
| - After setting up lighting (to verify the scene is properly illuminated) | ||
| - Before calling `render_image` (final visual check) | ||
| Skipping visual verification is a serious error — `get_viewport_screenshot` is your ONLY way to see the scene. The `render_image` tool does NOT return visual data you can analyze. |
There was a problem hiding this comment.
Keep the few-shot examples consistent with the new screenshot rule.
This block makes get_viewport_screenshot mandatory after lighting and before render_image, but the examples below still show flows that skip those checkpoints. That gives the model a counterexample inside the same prompt and weakens the new rule.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@lib/orchestration/prompts/blender-agent-system.md` around lines 13 - 17, The
few-shot examples in blender-agent-system.md contradict the "Visual Confirmation
— MANDATORY" rule; update each example sequence so every example calls
get_viewport_screenshot after creating main geometry/objects, after setting up
lighting, and immediately before calling render_image, removing any flows that
skip those checkpoints and ensuring the examples mirror the exact phrasing and
ordering of the rule (referencing get_viewport_screenshot and render_image) so
no counterexamples remain.
Comment on lines
21
to
30
| bl_info = { | ||
| "name": "ModelForge Blender", | ||
| "author": "ModelForge Team", | ||
| "name": "ViperMesh Blender", | ||
| "author": "ViperMesh Team", | ||
| "version": (1, 1, 0), | ||
| "blender": (3, 0, 0), | ||
| "location": "View3D > Sidebar > ModelForge", | ||
| "description": "Connect Blender to ModelForge AI Assistant", | ||
| "location": "View3D > Sidebar > ViperMesh", | ||
| "description": "Connect Blender to ViperMesh AI Assistant", | ||
| "category": "Interface", | ||
| "doc_url": "https://github.com/Ker102/ModelForge", | ||
| } |
There was a problem hiding this comment.
Inconsistent doc_url still references old ModelForge repo.
The doc_url in bl_info still points to https://github.com/Ker102/ModelForge while the rest of the addon has been rebranded to ViperMesh. This should be updated for consistency.
📝 Suggested fix
- "doc_url": "https://github.com/Ker102/ModelForge",
+ "doc_url": "https://github.com/Ker102/ViperMesh",📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| bl_info = { | |
| "name": "ModelForge Blender", | |
| "author": "ModelForge Team", | |
| "name": "ViperMesh Blender", | |
| "author": "ViperMesh Team", | |
| "version": (1, 1, 0), | |
| "blender": (3, 0, 0), | |
| "location": "View3D > Sidebar > ModelForge", | |
| "description": "Connect Blender to ModelForge AI Assistant", | |
| "location": "View3D > Sidebar > ViperMesh", | |
| "description": "Connect Blender to ViperMesh AI Assistant", | |
| "category": "Interface", | |
| "doc_url": "https://github.com/Ker102/ModelForge", | |
| } | |
| bl_info = { | |
| "name": "ViperMesh Blender", | |
| "author": "ViperMesh Team", | |
| "version": (1, 1, 0), | |
| "blender": (3, 0, 0), | |
| "location": "View3D > Sidebar > ViperMesh", | |
| "description": "Connect Blender to ViperMesh AI Assistant", | |
| "category": "Interface", | |
| "doc_url": "https://github.com/Ker102/ViperMesh", | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@public/downloads/vipermesh-addon.py` around lines 21 - 30, Update the bl_info
dictionary entry doc_url in vipermesh-addon.py so it no longer points to the old
ModelForge repo; locate the bl_info symbol and change the "doc_url" value to the
correct ViperMesh documentation or repo URL (e.g., the ViperMesh GitHub or docs
page) to keep branding consistent.
| return {"error": f"Invalid asset type: {asset_type}. Must be one of: hdris, textures, models, all"} | ||
|
|
||
| response = requests.get(f"https://api.polyhaven.com/categories/{asset_type}", headers=REQ_HEADERS, timeout=30) | ||
| response = requests.get(f"https://api.polyhaven.com/categories/{asset_type}", headers=REQ_HEADERS) |
There was a problem hiding this comment.
Missing timeout on HTTP request can hang Blender indefinitely.
This requests.get() call (and several others flagged by static analysis at lines 1460, 1480, 1604, 1741, 1759) lacks a timeout parameter. If the Poly Haven API is slow or unresponsive, Blender will freeze with no way to recover.
🐛 Suggested fix for all affected lines
def get_polyhaven_categories(self, asset_type):
...
- response = requests.get(f"https://api.polyhaven.com/categories/{asset_type}", headers=REQ_HEADERS)
+ response = requests.get(f"https://api.polyhaven.com/categories/{asset_type}", headers=REQ_HEADERS, timeout=30)Apply the same fix to:
- Line 1460:
search_polyhaven_assets - Line 1480:
download_polyhaven_asset(files endpoint) - Line 1604: texture download
- Line 1741: model download
- Line 1759: include file download
A 30-second timeout is reasonable for API calls; consider 60 seconds for file downloads.
📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| response = requests.get(f"https://api.polyhaven.com/categories/{asset_type}", headers=REQ_HEADERS) | |
| response = requests.get(f"https://api.polyhaven.com/categories/{asset_type}", headers=REQ_HEADERS, timeout=30) |
🧰 Tools
🪛 Ruff (0.15.6)
[error] 1438-1438: Probable use of requests call without timeout
(S113)
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@public/downloads/vipermesh-addon.py` at line 1438, The requests.get() calls
to the Poly Haven API currently have no timeout and can hang Blender; update
each call to include an appropriate timeout: add timeout=30 for lightweight API
endpoints (e.g., the call in response =
requests.get(f"https://api.polyhaven.com/categories/{asset_type}", ...) and the
functions search_polyhaven_assets and related API queries) and use timeout=60
for larger file downloads (e.g., download_polyhaven_asset, the texture download,
model download, and include file download functions) so network requests fail
fast and don’t block the UI.
Comment on lines
+52
to
58
| ### 🌐 Premium Web Dashboard UI | ||
| - **Dual Modes**: | ||
| - **Autopilot**: Conversational AI assistant for procedural generation. | ||
| - **Studio**: Full-page, immersive workspace with an icon-sidebar and interactive tool-card grid. | ||
| - **Teal Design System**: Professional, high-end visual aesthetics (#0d9488 focus) with interactive 3D hero sections and animated SVGs. | ||
| - **Supabase Auth**: Secure authentication with Google and GitHub OAuth. | ||
|
|
There was a problem hiding this comment.
Add blank lines around the new headings.
markdownlint is already flagging MD022 on these sections. The new headings run directly into the following paragraph/list/rule, so the README stays lint-failing until blank lines are added before and after them.
Also applies to: 66-76, 94-99
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@README.md` around lines 52 - 58, The new markdown headings (e.g., "### 🌐
Premium Web Dashboard UI") lack surrounding blank lines causing markdownlint
MD022; add a single blank line before and after each heading and the same for
the other flagged headings mentioned (lines containing "Dual Modes", "Teal
Design System", "Supabase Auth" and the sections at 66-76 and 94-99) so each
heading is separated from preceding and following content and the MD022
violations are resolved.
Batch 1 (Quick Wins): - Fix doc_url in both addon copies (ModelForge -> ViperMesh) - Add timeout=30/60 to all requests.get() calls in addons - Replace native img with Next.js Image in start-oauth page - Fix markdown lint (MD040) in spatial-positioning-guide and test-prompts - Add 'attach image' instruction to Test 18 Batch 2 (Component Fixes): - Fix nested button inside Link a11y issue in navbar.tsx - Add file input reset, 10MB validation, FileReader error handling - Add failed tool count/display in agent-activity collapsed summary Batch 3 (Logic Fixes): - Fix RAG retry to pick most-recent substantive prompt (not longest) - Fix stale closure via workflowStepsRef in studio-layout.tsx - Add 'skipped' to StepCommandResult status union type
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Type of Change
Related Issues
Closes #
Motivation and Context
How Has This Been Tested?
Screenshots (if applicable)
Checklist
Additional Notes
Summary by CodeRabbit
New Features
UI/UX Improvements
Updates