A Next.js portfolio website with voice-controlled site navigation, project management, settings, and chat integration. Exposes site tools for voice agents to control the portfolio interface.
27 tools with consistent naming (verb_noun pattern) and complete input schemas. Descriptions are present but brief (avg ~80 chars, below 194-char baseline). All tools have enums for constrained inputs, which is excellent. However, parameter descriptions are sparse or missing entirely, most params lack explanations of their purpose or constraints. Output schemas are not documented. Error handling guidance is absent. The server is a portfolio voice agent UI controller, not a backend API, which explains the simpler structure, but LLM-facing descriptions need expansion to guide tool selection and parameter usage.
Go back or forward in browser history when the visitor says go back or go forward.
Close the open project modal when the visitor asks to close, dismiss, or hide the current project, modal, note, or preview.
Control the open project preview video when the visitor asks to play, pause, mute, or unmute it. Fails if no modal or video is open.
End the voice agent session with an optional reason (user, health, or error).
Type into a visible site field when the visitor dictates guestbook, feedback, terminal, or chat text. Never submit the form.
Fetch recent user action context (current page, open modals, chat history snippet) to ground voice agent responses.
Parameter descriptions missing or minimal. Most enum parameters (e.g., 'action', 'direction', 'category') lack explanations of what each value does or when to use them. LLMs cannot infer intent from enum names alone.
Output schemas not documented. Tools return results but no schema is provided describing the response structure, fields, or types. LLMs cannot plan downstream calls or extract data reliably.
No error handling guidance. Tools lack descriptions of failure modes, recovery steps, or actionable error messages. LLMs cannot self-correct or know when to retry vs. ask the user.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Fetch a tiny fact snippet about Dhruv or the site before answering biographical, project, or site questions. Keep queries short.
Open an internal page when the visitor asks to go somewhere on this site.
Skip to the next disco mode track when the visitor asks to change the music.
Navigate to or focus the full /chat experience when the visitor wants to type or send a chat note.
Open the feedback note when the visitor wants to report a bug or leave feedback.
Open an approved external link when the visitor asks for GitHub, LinkedIn, resume PDF, a repo, email, or phone.
Open a project modal after navigating to /projects when the visitor asks to show or open a specific project. After success, you can offer video play/pause/mute if that project has a preview.
Open the keyboard shortcuts overlay when the visitor asks for hotkeys or keyboard help.
Run one allowlisted bare terminal command with no arguments, including hint. Never sudo, admin, puzzle, matrix, clear, or commands with arguments.
Scroll the current page container when the visitor asks to scroll up, down, to the top, or to the bottom. Amount is viewport heights.
Send a chat note through the real chat composer when the visitor dictates a message and wants it sent now. Fails if chat is not mounted or busy. Does not clear history.
Set one audio category volume when the visitor asks to change voice-agent, website-effects, or chat-read-aloud volume. Independent of master volume and the sound-effects mute toggle. Never unmutes.
Set sitewide master volume when the visitor asks to turn volume up, down, mute-level quiet, or a specific percent. Independent of the sound-effects mute toggle.
Set motion preference when the visitor asks to follow the device, reduce motion, or always animate.
Flip a boolean settings toggle when the visitor asks to change sound, haptics, stickers, speak-by-default, or voice extras. Does not clear chat, enable experimental features, or activate staging.
Change appearance when the visitor asks for dark, light, toggle, disco, or disco off.
Choose chat mic transcription when the visitor asks for native browser speech or on-device Whisper. Whisper may download a model.
Choose spoken-reply output when the visitor asks for device TTS or server custom speech.
Begin a voice agent session, initializing voice mode and context.
Send the open feedback note only after the visitor dictated the note and confirmed they want it sent.
Pin a guestbook note only after the visitor dictated the note and confirmed they want it posted.
open_project has empty enum for 'slug' parameter. The enum constraint is present but contains no values, making the parameter unusable. Either populate the enum or remove the constraint and document expected format.
Tool descriptions are brief (avg ~80 chars vs. 194-char baseline). Many lack context on WHEN to use the tool vs. similar ones, or what the tool returns. Descriptions should answer: What? When? What does it return?