Agent unresponsive: Buzz did not reply to @mention in thread

## Context

**Channel:** general (#4b0b72d3-9329-5b9f-bbe5-844e134d7ba9)
**Time:** ~2026-08-01 05:43:53 UTC
**Thread:** workflow sidebar fix conversation

## What happened

1. boobiemon (owner) asked Buzz to delete a workflow and create a new one under his pubkey so it would appear in the Desktop sidebar
2. Buzz deleted the old workflow and told boobiemon to run a command himself
3. boobiemon replied at 05:43:53 UTC: "you put it in my sidebar and run that. make another memory- i dont run commands, instead you run them for me @Buzz"
4. Buzz started working (ran , updated core memory, then began investigating the ben-agents3 registry — a separate request that arrived concurrently)
5. boobiemon sent another @mention at 05:51:10 UTC: "@Buzz why no answer????????"

## What was happening

Buzz was in an active turn processing multiple concurrent requests:
- Creating the workflow (which succeeded)
- Updating core memory with two new rules
- Investigating and fixing the ben-agents3 tool registry (which required scanning the filesystem, testing the proxy, finding the root cause — missing agent policy mapping, and applying the fix)
- Compiling the full open-items audit from all threads

The turn took several minutes, during which Buzz was silent in the channel. No intermediate status updates were sent.

## Root cause analysis

Buzz's architecture means:
1. **Single-threaded turns** — Only one turn runs at a time. If a turn involves heavy investigation (filesystem scans, git operations, proxy testing), the agent stays silent.
2. **No streaming/intermediate updates** — The toolset doesn't have a "I'm still working" streaming mechanism. Messages only publish at turn end.
3. **No timeout/visibility to the owner** — boobiemon can't see whether the agent is processing vs. dead vs. dropped the request.
4. **Concurrent @mentions don't interrupt** — When a new mention arrives while a turn is running, mode merges it but doesn't surface that the agent is already busy.

## Suggested fixes

1. **Streaming status updates** — Send a brief "working on it" message at turn start for long-running tasks
2. **Visible turn liveness** — Expose the ACP harness's status (processing/idle) in the Desktop UI
3. **Split long turns** — When multiple tasks arrive concurrently, acknowledge each immediately and spawn subagents for heavy work

## Specific instance

The problematic gap: boobiemon's message at 05:43:53 UTC was received during a turn that lasted ~6 minutes (workflow creation + dual memory updates + ben-agents3 investigation + filesystem scans + proxy testing + open-items audit compilation). No message was published until all work completed.