[Bug] Live Block occasionally fails to mark tools as completed

[randomm]

Problem

In the Live Block tool display, some tools occasionally remain showing the "running" status (◇) even after they've completed, while other tools on the same screen correctly show the "completed" checkmark (✓).

Screenshot Evidence

In the screenshot:

• ✓ task @git-agent — correctly marked done
• ✓ bash remory search ... — correctly marked done
• ✓ read ratcher/README.md — correctly marked done
• ✓ read ratcher/AGENTS.md — correctly marked done
• ◇ read ratcher/CONTRIBUTING.md — STILL SHOWS RUNNING (bug)
• ✓ bash remory add ... — correctly marked done (tools after it completed fine)

Expected

All completed tools should show ✓ checkmark.

Actual

Occasional tools get "stuck" showing ◇ running status even though they completed.

Likely Causes

1. Race condition in liveblock.ts — tool_end event arrives but the render doesn't update that specific tool
2. ID mismatch — the tool_end callID doesn't match the tool_start callID in the idMap
3. Dedup logic interference — the deduplication logic (lastToolKey) might be causing some tool_end events to update the wrong tool
4. log-update timing — the rewrite might be happening before the state is fully updated

Key Files

• packages/opencode/src/cli/lite/liveblock.ts — toolStart(), toolEnd(), render()
• packages/opencode/src/cli/lite/index.ts — tool event handling with idMap and dedup

Acceptance Criteria

• All completed tools show ✓ checkmark
• No race conditions in tool state updates
• Consistent rendering regardless of event timing

[randomm]

Additional hypothesis from user:

The stuck tool (read ratcher/CONTRIBUTING.md) might be a file that doesn't exist. The tool would error/fail.

If tool errors emit a different event type (e.g., tool_error) instead of tool_end, and we only listen for tool_end, the tool would stay stuck showing ◇.

Check:

1. What event is emitted when a tool fails?
2. Does liveblock.ts handle error status?
3. Does index.ts handle tool error events?

[randomm]

Improved by using lastToolId as fallback instead of toolCounter. Note: Root fix requires ensuring all tool events have callIDs from backend. This change reduces the failure rate significantly.