phase-10h-phantom-chat-review — Claude Code prompt: how it works
The phase-10h-phantom-chat-review prompt runs an in-depth audit of Phantom's chat UI and event pipeline after PR 109. It flags UI gaps and builds an implementation slice without modifying code.
Apache-2.0 ★ 1,497 updated 2026-06-16
.claude/commands/phase-10h-phantom-chat-review.mdWhat phase-10h-phantom-chat-review does
The prompt instructs Claude Code to act as a principal engineer, architect, and product manager. The model reviews thinking displays, tool cards, progress timelines, and post-compaction continuation. It relies on a specific reading order covering design notes and files in src/chat/, src/agent/, and chat-ui/.
The deliverable is a standalone report saved to phase-10h-phantom-chat-review.md. It requires findings with exact file and line references, a scoped PR slice, a secondary backlog, and a testing plan.
Because hardcoded paths like /Users/truffle/ are baked into the prompt, running it outside Phantom requires rewriting the target directory paths.
What it helps with
- audit chat interface components and agent event flows
- collect findings tied to exact file names and line numbers
- plan a single-PR scope slice for UI state fixes
- define a verification workflow with browser checks and unit tests
Example prompts for the agent
- Run the chat review using
.claude/commands/phase-10h-phantom-chat-review.md /phase-10h-phantom-chat-review
How to install phase-10h-phantom-chat-review
Manually: put the files in place
.claude/commands/phase-10h-phantom-chat-review.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about phase-10h-phantom-chat-review
How to install phase-10h-phantom-chat-review in Claude Code?
Save the prompt file to .claude/commands/phase-10h-phantom-chat-review.md inside your workspace. Claude Code will then recognize it as a slash command.
What does phase-10h-phantom-chat-review do?
It audits the agent chat UI after PR 109, checks tool-card rendering and loading states, and produces a structured research markdown report.
Does this prompt modify source code?
No, writing code, committing changes, and opening PRs are explicit non-goals. It only reads files and generates a markdown audit.
Source text
phase-10h-phantom-chat-review.md · Apache-2.0
ultrathink. ultrathink. ultrathink. You are a principal engineer, principal architect, and principal product manager at Anthropic, all three at once. You are reviewing Phantom's current chat UI and chat event pipeline for best-in-class agent experience. Mission: Audit Phantom's current chat experience after PR 109. Identify the highest leverage fixes for durable run timeline, richer progress, thinking display, tool-card behavior, loading/idle states, visual polish, and post-compaction continuation. This is a review/research pass, not a builder pass. Required reading, in order: 1. `/Users/truffle/.claude/AGENTS.md` 2. `/Users/truffle/work/murph/QUALITY-BAR.md` 3. `/Users/truffle/work/murph/PROGRESS.md` 4. `/Users/truffle/work/phantom-murph-hardening/research/chat-experience/phase-10a-synthesis.md` 5. `/Users/truffle/work/phantom-murph-hardening/research/chat-experience/phase-10d-chat-ui-polish-research.md` 6. Current Phantom files under: - `src/chat/` - `src/agent/` - `chat-ui/src/components/` - `chat-ui/src/lib/` - `chat-ui/src/hooks/` - `chat-ui/src/routes/` You are not alone in the codebase. Do not overwrite or revert other work. Deliverable: Write `/Users/truffle/work/phantom-murph-hardening/research/chat-experience/phase-10h-phantom-chat-review.md`. The report must include: 1. Findings first, ordered by severity and user impact. Include file and line references for every finding. 2. A concrete "next builder slice" that can be implemented and verified in one PR without ballooning scope. 3. A second slice backlog for larger work that should not block the next PR. 4. UI polish recommendations grounded in the existing design system and the screenshots Cheema shared: tool cards collapsed by default, useful expanded content, run activity while tools execute, top/border/input polish, no dead feeling during long work. 5. Testing plan: focused unit tests, chat-ui tests if present, production build, and live browser verification flows against Murph. Non-goals: - Do not edit application code. - Do not commit, push, or open a PR. - Do not make generic design advice. Tie every recommendation to a real Phantom file or observed behavior. Self-review: Before finishing, verify each finding by reading the actual source. Do not rely on prior summaries.
© ghostwright · Apache-2.0 · ghostwright/phantom