phase-10f-run-console-ui — Claude Code prompt: agent run console UI
phase-10f-run-console-ui — a prompt for building a run console UI in Phantom on Murph: displays live tool activity, counts, and session state with durable rehydration after refresh. Keeps the chat responsive during long operations.
Apache-2.0 ★ 1,497 updated 2026-06-16
.claude/commands/phase-10f-run-console-ui.mdWhat phase-10f-run-console-ui does
The prompt frames Claude Code as principal engineer, architect, and product manager combined. The goal is to evolve the run activity row into a proper console showing tool history, subagents, compaction, and MCP status. Operators see what the agent is doing—tool parameters, results, activity timeline—while credential-shaped values remain masked.
The scope is focused: upgrade the compact live and replayed run console, keep tool card disclosure accessible, improve visual density and spacing without a full redesign. The runtime contract stays generic; no Phantom-specific hooks are added to Murph, and model-facing tool result semantics don't change.
Acceptance includes passing existing tests, a clean build, typecheck and lint passing, and the live browser proof on Phantom-on-Murph still working. The UI renders tool progress as an activity timeline with expandable details, not an isolated row, and rehydrates correctly on refresh or replay.
What it helps with
- build a run console UI with live activity and tool counts
- display tool parameters and output without exposing credentials
- rehydrate session state after browser refresh
- keep tool card disclosure accessible and inspectable
- integrate with Murph through the Agent SDK-compatible boundary
Example prompts for the agent
- Place
phase-10f-run-console-uiin.claude/commandsand use it to design Phantom's run console interface with Murph - Apply the prompt when implementing live tool activity tracking in the chat UI
- Reference the accessibility and credential redaction requirements from the prompt during implementation
How to install phase-10f-run-console-ui
Manually: put the files in place
.claude/commands/phase-10f-run-console-ui.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about phase-10f-run-console-ui
How do I install phase-10f-run-console-ui in Claude Code?
Copy the prompt file to .claude/commands/phase-10f-run-console-ui.md. Claude Code will load it automatically from that directory.
What does phase-10f-run-console-ui do?
It frames Claude Code to design Phantom's run console: live tool activity, counts, expandable parameters and output, and durable rehydration after refresh, without exposing credentials.
Why build a console UI instead of just showing run activity?
A timeline-based console keeps the chat from feeling dead during long tool operations. Operators get a structured view of tools, subagents, and MCP status—the way serious agent products work—not just an event list.
Source text
phase-10f-run-console-ui.md · Apache-2.0
ultrathink. ultrathink. ultrathink. # Phase 10F: Phantom Run Console UI You are a principal engineer, principal architect, and principal product manager at Anthropic, all three at once. ## Grounding Phantom is now running on Murph through the Agent SDK-compatible boundary. Murph emits enriched tool progress based on Pi runtime lifecycle events. Phantom already persists durable run timelines and rehydrates them after refresh. Phase 10E proved: - OpenAI, Anthropic, and Z.AI provider proof passed. - A live Phantom browser session on Murph showed a Bash card while running. - Safe Bash parameters appeared inside the expanded card. - Final output and durable session detail contained the tool evidence. The next goal is product quality. The chat should not feel dead during long tool runs. It should feel like a durable, inspectable run console. ## User Direction The operator owns the agent. We should not hide normal tool activity from the operator. Keep credential-shaped values protected, but otherwise show what the agent is doing. Avoid hacks. Build this the way serious agent products do: structured live activity, clear states, readable tool details, and durable rehydration. ## Scope Implement a focused next slice: 1. Improve the live and replayed run activity row into a compact run console. 2. Show counts and useful facts from tools, subagents, compaction, and MCP status. 3. Preserve accessible disclosure behavior for tool cards. 4. Improve visual density, borders, and spacing without redesigning the whole app at once. 5. Keep the runtime contract generic. Do not add Phantom-specific shortcuts to Murph. ## Non-Goals - Do not change model-facing tool result semantics. - Do not remove credential-shaped redaction. - Do not build a new TUI, SaaS dashboard, or separate runtime server. - Do not turn this into a broad redesign that makes verification harder. ## Acceptance Criteria - Existing unit tests pass. - Chat UI build passes. - Phantom typecheck and lint pass. - Live browser proof still passes on Phantom running on Murph. - The UI shows tool progress as an activity timeline, not just an isolated row. - Expanded tool details show parameters and output when present. - Refresh or replay still rehydrates the durable run activity.
© ghostwright · Apache-2.0 · ghostwright/phantom