[Castor-o7]

Why

scripts/test-restart-file-e2e.sh failed intermittently: on #185 at 255b343, on #194 at 18a1dd7 (neither diff touches it), and in 2 of 15 local runs at #185's head. With a copy of the script that prints the error the script normally throws away, the daemon's answer to the member's muster://messages?inbox=true&unread=true read was:

[unregistered] caller (client …) has no session on the roster — re-SET your Session and retry

A fresh shim read before its Session SET landed. The test's readiness loop passed early because is_self matched the session's row from the previous shim run, which has the same id and a dead connection. Tool calls already hold for registration and retry once on this refusal (MUS-72, MUS-79). The messages read did neither. A real agent reading its inbox right after a restart could hit the same refusal.

What

• The hold and the retry-once are moved out of call_tool into as_registered, and the messages resource uses it too. The tool path behaves the same.
• Roadmap entry for MUS-114 (ledger record 1046).

Checked

• cargo test -p muster-shim --bins: 151 passed.
• cargo clippy -p muster-shim --all-targets -D warnings: clean.
• The restart-file e2e, run 30 times locally with the fix: 0 failed (before: 2 of 15).

Not covered

• No unit test drives the refusal through as_registered, because it needs a ToolHost and a live client. The e2e is the regression test.
• The conformance failures on docs: Rally is the board's source of truth, and the hand-kept table is archived (RAL-9) #194 are a different cause, MUS-93 (succession never re-checks an unobservable predecessor). plush-tortoise is taking that one.

🤖 Generated with Claude Code