Add transcript view modes (Summary / Normal / Verbose) to the VS Code extension to hide tool-call noise
Preflight Checklist
- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)
Problem Statement
In the VS Code extension, the conversation panel renders every tool call as its own full-size block — Read, Grep, Glob, each Playwright/MCP call (browser_navigate, browser_click, browser_take_screenshot, plus their "Ran code" / "Result" boxes), web fetches, and so on. On a long agentic session this produces a wall of low-signal noise that I have to scroll past to find the parts I actually care about: the reasoning, the todo updates, the edits, and the final answer.
The desktop app already solves this with transcript view modes (Summary / Normal / Verbose), where Summary collapses the tool churn. The VS Code extension has no equivalent — no view mode, no setting, and no way to collapse or hide tool-call blocks (only extended-thinking blocks collapse, via Ctrl+O). This is the single biggest readability gap between the extension and the desktop app.
Proposed Solution
Bring the desktop app's Summary / Normal / Verbose view modes to the VS Code extension, with a toggle in the panel and/or a setting.
At minimum, add a setting to hide or collapse-by-default specific tool-call block types in the transcript (e.g. Read, Grep, Glob, WebFetch, WebSearch, and all mcp__* tools), while keeping reasoning text, todo updates, Edit diffs, and Bash commands visible.
Alternative Solutions
- Click-to-collapse on individual tool blocks (like the existing thinking-block chevron), plus a collapse-all shortcut.
- A per-tool-type visibility config in
settings.json.
Priority
High - Significantly impacts my daily workflow
Feature Category
Configuration and settings
Use Case Example
Writing/research/agentic sessions that make many Read and Playwright calls become very hard to follow in the VS Code panel — the meaningful output (an "Update Todos" block, a paragraph of reasoning, a final summary) is buried between dozens of "Ran Playwright code" / "Result" boxes.
Additional Context
Related but narrower: #72848 (hide/collapse subagent tool activity). This request covers all tool activity in the main conversation, not just subagents — the two could be unified under one "transcript verbosity / view mode" setting.
3 Comments
I want this option but for the opposite reason: to allow me to enable _more_ verbose transcript; so I can see the inner monologue of its thinking like I can with vscode copilot chat, which can help me spot misconceptions or "wait actually..." loops sooner
+1 to this request. This is currently one of the biggest UX pain points for me in the VS Code extension.
The main issue is not just that tool calls take up a lot of space, but that the assistant's actual prose gets interleaved with command output, file reads, diffs, and other tool-call blocks. When Claude finishes a turn and control comes back to me, I often have to scroll through the entire transcript just to find and re-read what Claude actually said.
This becomes especially annoying in longer turns, where useful assistant messages may be scattered across different points in the transcript rather than appearing only at the very end.
I would strongly prefer a UI where, once a turn is complete:
The current transcript is useful for debugging and inspection, but it should feel secondary to the actual conversation, not compete with it for attention.
Additional proposal: Conversation-only feed with persistent questions requiring user attention
I use the Claude Code extension in Cursor for long-running development tasks, often involving orchestration, multiple agents, and dozens of tool calls.
The biggest usability problem isn't simply that tool output occupies too much space. It's that operational logging and communication intended for the user are mixed into a single chronological feed.
For example, the transcript contains Bash commands, file operations, GitHub PR creation, and agent activity interspersed with important decisions and questions. An agent might ask me three questions, continue doing other work for twenty minutes, and eventually say, "Still waiting on you: question 2."
At that point I have to scroll back through potentially hundreds of tool operations to locate the original question and its context.
Proposed solution
Introduce three optional views of the same underlying session:
Importantly, this would be a presentation-layer feature. The full underlying transcript and agent context would remain unchanged.
Outstanding questions
Questions should remain easy to find until answered or explicitly withdrawn, even if the agent continues working.
Ideally:
Why this matters
In increasingly autonomous workflows, the developer is often supervising rather than directly performing every action. The interface should distinguish between what the agent is doing and what the agent needs from the human.
A conversation-only view would make Claude Code substantially easier to supervise without sacrificing access to detailed operational logs.
Related: #73413 (transcript view modes), #80576 (question context disappearing in the VS Code extension).