[Bug]: Bash command shown twice in tool-call activity — dedup check doesn't strip the "Bash: " prefix baked into detail #7711

Before submitting

• I searched existing issues and did not find a duplicate. Closest is [Bug]: Bash toolcall does not show the command being ran, just the output twice. #7307, but that one is Desktop+OpenCode+WSL and describes the output rendering twice with the command missing. This is a different, more specific symptom: the command itself renders twice (once plain, once prefixed Bash: ), output isn't affected.

Area

apps/web

Steps to reproduce

1. Run any Bash tool call in a Claude-provider thread (desktop or web).
2. Expand the tool-call activity in the chat timeline.

Expected behavior

The command is shown once.

Actual behavior

The command is shown twice: once as the plain command preview, and once again below it prefixed with Bash: .

Root cause (traced against the live event log + source)

The canonical item.completed activity carries the same command in two places on one payload:

• payload.data.input.command — bare command, e.g. cd ... && grep ...
• payload.detail — the same command with a tool-name prefix baked in, e.g. Bash: cd ... && grep ...

extractToolDetail (apps/web/src/session-logic.ts) is designed to suppress detail when it's redundant with the already-rendered command preview — it normalizes both strings via normalizePreviewForComparison → normalizeCompactToolLabel and only returns detail if the normalized strings differ.

normalizeCompactToolLabel only strips a trailing complete/completed suffix:

function normalizeCompactToolLabel(value: string): string {
  return value.replace(/\s+(?:complete|completed)\s*$/i, "").trim();
}

It never strips the leading "<ToolName>: " prefix that's already baked into payload.detail. So normalizedCommand ("cd ... && grep ...") and normalizedDetail ("bash: cd ... && grep ...") never match, the dedup check in extractToolDetail treats them as different content, and the redundant detail line renders anyway — directly under the command that was already shown.

Example payload from a live session (Claude provider)

{
  "type": "item.completed",
  "payload": {
    "itemType": "command_execution",
    "title": "Command run",
    "detail": "Bash: cd /Users/.../eksaq-platform/knowledge-base && grep -n -E \"F0[0-9][0-9]\" TRACKER.md | grep -i -E \"hpc|F0\" | sed -n '1,120p'",
    "data": {
      "toolName": "Bash",
      "input": {
        "command": "cd /Users/.../eksaq-platform/knowledge-base && grep -n -E \"F0[0-9][0-9]\" TRACKER.md | grep -i -E \"hpc|F0\" | sed -n '1,120p'"
      }
    }
  }
}

Impact

Cosmetic, but consistently reproducible on every Bash tool call — clutters the timeline for any thread with meaningful shell usage.

Version or commit

Nightly, macOS desktop, Claude provider.

Suggested fix

Strip the "<ToolName>: " prefix (or any leading toolName + ": ") from detail before the redundancy comparison in extractToolDetail/normalizePreviewForComparison, so it's correctly recognized as duplicate content when it only differs from the command preview by that prefix.