Summary

Pulled the branch and read the full diff against master. The persistence shape here is right — localStorage['hermes-activity-disclosure:<sid>:<turn_key>'] keyed by session id and either assistant:<index> (settled) or live:<stream_id> (in-flight), with a _copyActivityDisclosureState() handoff when a live turn settles. That is the smallest reasonable durable representation for a UI-only toggle.

There is one thing to flag, though.

Merge collision with #1725

#1725 (open right now, also touching static/ui.js and the same ensureActivityGroup() template in particular) deletes the .tool-call-group-list and .tool-call-group-count spans entirely from the summary. This PR keeps them. After both merge in either order, whoever lands second will need to rebase, and the obvious mechanical resolution leaves dead spans in this PR's template.

Reading static/ui.js:3819 on this branch, the new template still includes:

group.innerHTML=`<button type="button" class="tool-call-group-summary" aria-expanded="${collapsed?'false':'true'}" onclick="_toggleActivityGroup(this)"><span class="tool-call-group-chevron">${li('chevron-right',12)}</span><span class="tool-call-group-label">Activity</span><span class="tool-call-group-list">tools / thinking</span><span class="tool-call-group-duration"></span><span class="tool-call-group-count">0</span></button><div class="tool-call-group-body"></div>`;

vs #1725's simplified version which drops .tool-call-group-list and .tool-call-group-count (and also strips the corresponding CSS rules + sync logic in _syncToolCallGroupSummary). The persistence work here is orthogonal to the summary-cleanup work there, so a coordinated rebase should be straightforward — the _toggleActivityGroup(this) onclick rewrite and the data-activity-disclosure-key attribute don't depend on those spans being present.

Code reference

The persistence path on static/ui.js:3766-3784:

const _activityDisclosureStoragePrefix='hermes-activity-disclosure:';
function _activityDisclosureStorageKey(activityKey){
  if(!activityKey||!S.session||!S.session.session_id) return null;
  return _activityDisclosureStoragePrefix+S.session.session_id+':'+activityKey;
}
function _readActivityDisclosureState(activityKey){
  const key=_activityDisclosureStorageKey(activityKey);
  if(!key) return null;
  try{
    const saved=localStorage.getItem(key);
    return saved==='open'||saved==='closed'?saved:null;
  }catch(_){return null;}
}

And the live-to-settled transfer in static/messages.js:929-932:

if(typeof _copyActivityDisclosureState==='function'&&lastAsst){
  const assistantIdx=S.messages.indexOf(lastAsst);
  if(assistantIdx>=0) _copyActivityDisclosureState('live:'+streamId, 'assistant:'+assistantIdx);
}

Diagnosis

Two minor observations beyond the #1725 collision:

1. The _writeActivityDisclosureState writes lazily on user toggle. There is no proactive cleanup of orphan keys when sessions are deleted. Per the body, this is intentional ("no pruning is added in this PR") and the entries are tiny, but if someone runs Hermes for a year with hundreds of sessions, the localStorage namespace will grow unbounded. Worth noting in a follow-up tracking issue rather than blocking here.
2. _activityKeyForLiveTurn() returns null when there is no S.activeStreamId. ensureActivityGroup() falls back to live:<stream_id> only via opts.activityKey. The two ensureActivityGroup({live:true, ...}) callers in appendLiveToolCard and appendThinking both already pass activityKey:_activityKeyForLiveTurn(), so no path silently drops the key — that's good.

Verification

The static source-shape tests in tests/test_ui_tool_call_cleanup.py:150-191 cover the right invariants (storage prefix presence, session-id scoping, key attribute, hydrate-before-default, click-persists, settled key shape, live key shape, and the _copyActivityDisclosureState call in messages.js). All three CI checks are green.

Recommendation: rebase against #1725 once one of the two lands (this PR or that one); the persistence logic and the summary cleanup are independent and should not need any logic changes when reconciled — only the template literal and the contents of _syncToolCallGroupSummary.

Thanks @Michaelyklam — this shipped in v0.51.8 (commit 85d0279) as part of a 7-PR full-sweep batch release. Stage rebased your branch onto current master, ran the full pre-release gate (4584 pytest, browser tests, Opus advisor verdict SHIP), and merged via release PR #1737.

GitHub didn't auto-close because the merge commit only references the squash-merged stage branch, not your fork's commit directly — closing manually for hygiene.

Live now on https://get-hermes.ai/ and on existing installs after git pull + restart.

Release notes: https://github.com/nesquena/hermes-webui/releases/tag/v0.51.8