Managed starter agents show RUNNING but never respond to DMs or mentions in local `just dev` environment

# Managed starter agents show RUNNING but never respond to DMs or mentions in local `just dev` environment

## Summary

Starter agents (Fizz, Honey, Bumble) appear healthy and report `RUNNING`, but they never respond to either direct messages or channel mentions when running Buzz locally via `just dev`.

The relay continues to accept and ingest events normally, but no replies are ever produced.

---

# Environment

- macOS (Apple Silicon)
- Docker Desktop
- Buzz running from source
- Branch: `main`

Commit tested:

```text
dd222a509b156ba52ed3219e895d7bf1cf322c92
```

Working tree:

```text
On branch main
Your branch is up to date with 'origin/main'.

nothing to commit, working tree clean
```

---

# Setup

Repository cloned locally.

Ran:

```bash
just setup
just dev
```

Completed onboarding by:

- Importing an existing Nostr identity
- Allowing macOS Keychain access ("Always Allow")
- Joining the local development community
- Accepting the default starter agents (Fizz, Honey, Bumble)

---

# Reproduction

1. Clone the repository.
2. Run:

```bash
just setup
just dev
```

3. Complete onboarding.
4. Import an existing Nostr identity.
5. Verify the starter agents appear.
6. Verify each starter agent reports:
- Runtime: Codex
- Status: RUNNING
- Green status indicator
7. Open either:
- a direct message with Fizz, or
- the `#Welcome` channel.
8. Send:

```text
Reply with exactly: hello
```

(or any other prompt)

---

# Expected behavior

The starter agent should receive the message, invoke the configured Codex runtime, and reply.

The Harness Log should show activity corresponding to the request.

---

# Actual behavior

The message is accepted by Buzz but no reply is ever generated.

Observed behavior:

- Starter agents remain `RUNNING`
- Direct messages never receive a response
- Channel mentions never receive a response
- Harness Log initially remained empty
- No reply is ever posted back into Buzz

---

# Relay observations

The relay appears healthy.

Observed during testing:

- Events are accepted successfully
- Events are ingested successfully
- Relay continues normal operation

Representative relay output:

```text
HTTP POST /events
accepted=true

Event ingested via pipeline
```

The relay continues processing requests normally after the event is accepted.

---

# Additional observations

While reproducing the issue:

- The Codex runtime reports as configured.
- Starter agents show a green RUNNING indicator.
- The local relay continues servicing requests normally.
- I did not observe any obvious relay errors or crashes.
- The issue reproduces with both direct messages and channel mentions.

---

# What I already verified

- Docker Desktop running
- Local relay starts successfully
- Local development community created
- Existing Nostr identity imported successfully
- macOS Keychain permissions granted
- Codex runtime selected
- Starter agents show RUNNING
- Direct messages tested
- Channel mentions tested
- Clean checkout on current `main`
- No local modifications

---

# Attachments

Screenshot of dm and agent status

---

# Notes

I'm happy to collect additional diagnostics or run any suggested debugging commands if they would help narrow this down.