phase-10a-product-experience-research — Claude Code prompt
The phase-10a-product-experience-research prompt guides research into Phantom’s long-running agent chat, from run status to tool details and error states. Use it to shape an interface proposal and a phased build plan.
Apache-2.0 ★ 1,497 updated 2026-06-16
.claude/commands/phase-10a-product-experience-research.mdWhat phase-10a-product-experience-research does
The prompt asks Claude Code to inspect Phantom’s existing chat and research public OpenAI and Anthropic references. It covers what users see before the first response, quiet tool waits, subagents, errors, mobile layouts, and completed actions. Claims about external products should include source links.
The requested report includes product principles, an active-run UI model, disclosure rules for tool details, and staged implementation guidance. It also calls for a test plan using long-running Phantom tasks on Murph and browser verification. Local notes go under /Users/truffle/work/phantom-chat-experience/research/chat-experience/.
The prompt relies on project files at absolute local paths, so those paths need adapting outside that environment. It rules out exposing hidden reasoning or turning the chat into a raw log viewer.
What it helps with
- design status updates for long agent runs
- choose which tool-call details appear by default
- plan a mobile layout for agent chat
- split interface work into shippable phases
- prepare long-run and browser checks
Example prompts for the agent
- Run
/phase-10a-product-experience-researchto plan the long-run chat experience for Phantom. - Review Phantom’s chat and propose how it should show waiting, tool activity, and run completion.
- Use this prompt to draft implementation phases and a Murph test plan.
How to install phase-10a-product-experience-research
Manually: put the files in place
.claude/commands/phase-10a-product-experience-research.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about phase-10a-product-experience-research
How do I install phase-10a-product-experience-research in Claude Code?
Save the prompt at .claude/commands/phase-10a-product-experience-research.md in your project. Then invoke it with /phase-10a-product-experience-research.
What does phase-10a-product-experience-research do?
It guides a study of Phantom’s interface for long agent tasks and requests an active-run model, tool-detail disclosure rules, and a phased implementation plan.
Which Phantom files does the prompt ask Claude Code to read?
It lists the project README and CLAUDE.md, global styles, and components for the message list, assistant message, tool-call card, and thinking block, using absolute local paths.
Source text
phase-10a-product-experience-research.md · Apache-2.0
ultrathink. ultrathink. ultrathink. You are a principal engineer, principal architect, and principal product manager at Anthropic, all three at once. Your mission is to define what a best-in-class long-running agent chat experience should feel like for Phantom. No context anxiety. No token limits. No time pressure. No cost anxiety. No "v2" thinking. There is no v2. Build it right. Take as long as you need. ## Context Phantom is not a generic chatbot. It is an autonomous AI co-worker with its own computer. It runs long tasks, creates files, uses tools, spawns subagents, compacts context, and sometimes works for minutes. The chat UI must make that feel alive, trustworthy, and inspectable without flooding the user with raw logs. Cheema wants the experience to compete with the best agent products, not just to show a spinner. Research the current product patterns, but filter hard for official docs, primary sources, and directly observable product behavior. Avoid SEO listicles. ## Required Reading Read Phantom first: 1. `/Users/truffle/work/phantom-chat-experience/README.md` 2. `/Users/truffle/work/phantom-chat-experience/CLAUDE.md` 3. `/Users/truffle/work/phantom-chat-experience/chat-ui/src/styles/globals.css` 4. `/Users/truffle/work/phantom-chat-experience/chat-ui/src/components/message-list.tsx` 5. `/Users/truffle/work/phantom-chat-experience/chat-ui/src/components/assistant-message.tsx` 6. `/Users/truffle/work/phantom-chat-experience/chat-ui/src/components/tool-call-card.tsx` 7. `/Users/truffle/work/phantom-chat-experience/chat-ui/src/components/thinking-block.tsx` Then research current public references. Prioritize: 1. Official OpenAI docs for tool calls, Responses API streaming, and reasoning summaries if available. 2. Official Anthropic docs for tool use, extended thinking, and streaming. 3. Claude Code, Codex, Cursor, ChatGPT, and similar products only where there are primary docs, screenshots, or direct observable behavior. 4. Local reference repos only for implementation patterns, not UI taste: `/Users/truffle/work/opencode`, `/Users/truffle/work/openclaw`, `/Users/truffle/work/claw-code`, `/Users/truffle/work/pi-mono`. ## Questions To Answer 1. What should the user see immediately after sending a message, before the first assistant token or tool card exists? 2. How should Phantom communicate "the agent is working" during silent tool waits, subagent work, long shell commands, web fetches, browser checks, and compaction? 3. How much tool detail should be shown by default versus behind disclosure? 4. How should completed tool calls be represented in the transcript? Should they disappear, collapse into a timeline, or stay inline? 5. How should errors, rate limits, context compaction, and retries be framed so users understand the system is recovering? 6. What should be visible on mobile, where space is tight? 7. What is the design language that fits Phantom specifically: autonomous co-worker, own computer, shareable artifacts, long-running real work? ## Acceptance Criteria Produce: 1. A product principle list, no more than 10 principles. 2. A concrete UI model for active runs: global run status, inline activity rail, tool cards, thinking/activity summaries, completion metadata. 3. A default disclosure policy for tool inputs, outputs, sensitive values, thinking, screenshots, and large payloads. 4. A phased build recommendation: first shippable slice, second slice, later polish. 5. A test plan that includes real Phantom-on-Murph long-running tasks and browser verification. ## Anti-Patterns Do not recommend a marketing-style redesign. Phantom is an operational tool. Do not expose raw hidden reasoning. Do not turn the chat into a log viewer. Do not bury the actual answer under tool chrome. Do not add cards inside cards. Do not use huge hero-style UI inside the chat surface. ## Handoff Return a report with source links for external claims. If you create local notes, write under `/Users/truffle/work/phantom-chat-experience/research/chat-experience/`. ultrathink. The bar is "I trust this agent is working" within one second.
© ghostwright · Apache-2.0 · ghostwright/phantom