Dispatch channel stops generating responses after interacting with child sessions — messages marked Read but no reply

$(cat <<ENDOFBODY

Bug Description

The Cowork Dispatch channel receives user messages (marked "Read") but generates zero responses. The dispatch thread is permanently broken — no restart, cache clear, or data wipe recovers it.

Environment

• OS: macOS (Apple Silicon)
• Claude Desktop: Latest (as of March 28, 2026)
• Plan: Max

Steps to Reproduce

1. Use Dispatch normally — send tasks, receive responses
2. Open dispatch-spawned child sessions from the sidebar and interact with them directly (type messages into them)
3. Return to Dispatch and send a message
4. Message is marked "Read" but no response is generated
5. All subsequent messages to Dispatch fail the same way

Root Cause (from logs)

Three compounding errors in ~/Library/Logs/Claude/claude.ai-web.log and claude.ai-web1.log:

1. Orphaned subagent message routing

[LOCAL_SESSION] Subagent parent not found for user message - possible out of order processing [object Object]

70+ of these fire in a single second. Manually interacting with dispatch-spawned child sessions breaks the parent-child message routing. The dispatch thread tries to route messages to subagent parents that no longer exist.

2. TypeError crash on every dispatch message

Uncaught (in promise) TypeError: Cannot read properties of undefined (reading includes)

This fires every time a user sends a message to Dispatch. The response pipeline crashes before generating a reply. Messages get marked "Read" (input pipeline works) but no response is ever produced.

3. EventEmitter memory leak

MaxListenersExceededWarning: Possible EventEmitter memory leak detected. 11 listeners added.

Listener accumulation on multiple IPC channels from stacked dispatch sessions degrades event handling.

Additional errors

[REACT_QUERY_CLIENT] QueryClient error: Error: Not found
[REACT_QUERY_CLIENT] QueryClient error: Error: ["account_profile"] data is undefined
CSP violation: Connecting to https://a-api.anthropic.com/v1/m blocked by Content Security Policy

Recovery Attempts (all failed)

• Quit and relaunch Claude Desktop
• Delete Session Storage directory and relaunch
• Delete IndexedDB directory and relaunch
• Full app restart from dock
• Sending test messages from both desktop and mobile

The corrupted state appears to be server-side — local cache clearing does not resolve it. The dispatch conversation on the backend has broken routing data that persists across all local recovery attempts.

Expected Behavior

Dispatch should either:

1. Handle orphaned subagent parents gracefully instead of crashing the response pipeline
2. Allow users to interact with child sessions without corrupting the parent dispatch thread
3. Provide a user-facing reset mechanism for the dispatch channel

Related Issues

• #36157 — Dispatch causes black screen freeze, requires full data wipe to recover (same CSP and QueryClient errors)
• #37473 — Stale local-agent-mode-sessions directory
• #38049 — Dispatch stuck on "Brewing..." (same symptom: messages received, no response)
• #37713 — Dispatch sessions fully isolated from Projects

ENDOFBODY
)