[loganj]

🤖

Summary

• Problem: main's browser CI fails in solo-agent-tagging.spec.mjs on Chromium and WebKit. The journey waits for a "Retry agents" notice under the Solo message box, and that notice no longer exists.
• Why: two PRs crossed. test(solo): cover agent tagging behind a background read-state snapshot #2941 added this journey, and its last CI run was on a base from before fix(agents): drop the Solo composer's inventory notice and reread partial failures #2948. fix(agents): drop the Solo composer's inventory notice and reread partial failures #2948 then removed the Solo composer's standing "Retry agents" notice: a failed agent list now shows in the @mention suggestions ("Could not load agents. Retry to refresh." with a Retry suggestions option) and in the agent picker. When test(solo): cover agent tagging behind a background read-state snapshot #2941 merged on top, the journey waited for a notice the app no longer draws.
• Now: the journey recovers the same way a user does today. It types @ in the Solo chat, sees Retry suggestions in the mention suggestions, clicks Retry suggestions while the background read-state snapshot is still held, and checks that the composer addresses the one agent in the chat. Retry picks nobody, so the draft is still the bare @.

Details

• Test-only. No production code changes.
• The journey keeps its purpose from test(solo): cover agent tagging behind a background read-state snapshot #2941: the agent list read must not wait behind a slow background read-state snapshot (the fix(solo): restore agent selection during slow background reads #2751 fix). With that fix reverted locally (agent list read on background priority), the journey fails: the agent is never addressed after Retry.
• Retry suggestions calls the same agent list refresh as the removed notice's button.
• After the connection gap, the journey refuses the first agent list read and holds any later one until the viewer has clicked Retry. Whether the reconnect rereads the list again (added in fix(agents): drop the Solo composer's inventory notice and reread partial failures #2948) depends on whether the chat is open when the connection returns, and that order differs between machines. A read that is held can't succeed or fail before Retry, so the journey passes or fails the same way in both orders.
• While a reread is running, the mention menu shows "Loading recipients…" in place of the failure text, but it still offers Retry suggestions. The journey uses that option as its failure signal.
• The journey also checks that every agent list read it lets through goes out at foreground priority, next to its existing check that the held snapshot is a background read.

🤖 Generated with Claude Code

[loganj]

🤖 Superseded by #3015, which fixed this spec on main first (Retry through the Choose agent picker). Closing this one.

One note for whoever touches this spec next: the version on main still uses a single refuse-until-flip flag. The number of agent list reads after the gap is 1 or 2. It depends on whether the Solo chat is open when the connection returns, because the #2948 reconnect reread only fires when something holds the list. So a reread can either be refused but not yet settled when Retry is clicked (Retry joins it and no new read goes out), or arrive after the flip and clear Retry before the click. This PR's last commit (81af325) avoids both: it refuses one read and holds later ones until after the click. Its CI passed, including the WebKit flake lane.