Custom providers that have a curated models: list in config.yaml
(e.g. ZenMux gateways) should show ONLY those configured models in
the picker dropdown, not the full /v1/models catalog.
Before this fix, _named_custom_groups unconditionally called
_read_custom_endpoint_models() which would pull hundreds of models
from aggregator gateways and overwrite the user's curated list.
Now the build checks if the custom_provider entry has a non-empty
models dict/list in config.yaml — if so, it skips the live fetch
and uses only the configured models (same behavior as hermes-agent
model_switch.py Section 4 patch).
Closes: configure-model-list-should-be-authoritative
The WebUI clarification popup had a response-delivery failure: users
submitted answers in the popup, but the agent still fell through to the
timeout fallback message. Three bugs conspired:
1. No stable clarify_id — _ClarifyEntry had no unique identifier, so
the frontend could not reference a specific pending prompt. The
backend used FIFO resolution which silently failed for stale/late
responses.
2. Frontend hid the card before confirmation — respondClarify() called
hideClarifyCard(true, 'sent') BEFORE the API call completed. If the
backend rejected the response, the card was already gone and the
user's draft was discarded.
3. Backend lied about success — _resolve_clarify_legacy() returned
bool(resolved) or not bool(clarify_id). Since the frontend never
sent clarify_id, the backend always reported ok:true even when
nothing was resolved.
Changes:
api/clarify.py:
- _ClarifyEntry now auto-generates a stable clarify_id (uuid4.hex[:12])
- submit_pending() injects clarify_id into the data dict visible to the
frontend via SSE and polling
- New resolve_clarify_by_id() for O(1) lookup by id instead of FIFO pop
api/routes.py:
- _resolve_clarify_legacy() uses resolve_clarify_by_id when clarify_id
is provided; returns actual bool result (no more unconditional True)
- _handle_clarify_respond() returns HTTP 409 + {ok:false, stale:true}
when resolution fails
static/messages.js:
- respondClarify() now sends clarify_id in the POST body
- Waits for a positive backend acknowledgement before hiding the card
- Saves a draft copy before POST and restores it on failure
- On 409/network error: re-enables controls, shows error toast
- Guards against parallel-SSE race where clearing the cache after a
successful response could erase a newly queued next prompt (codex P1)
tests:
- Updated test_sprint30.py for new ack-before-hide behaviour
- Updated test_clarify_unblock.py for 409 on stale responses
Closes#2639.
Compact tool activity regressed into separate Activity rows and standalone Thinking blurbs when interim assistant text retired the current live activity group and Thinking rendered outside the disclosure.
Render Compact-mode Thinking inside the shared Activity body for live and settled turns, keep interim assistant text from splitting the current Activity group, and remove the now-unused stream-local activity-close path. This restores the intended single compact disclosure without adding new functionality.
The _normalized_message_timestamp_for_key helper was preserving
microsecond precision (%.6f). When the same message is persisted by
both the WebUI sidecar JSON writer and the Hermes agent state.db
writer, their timestamps can differ by a few microseconds, causing
_session_message_merge_key to produce different keys for the same
logical message and letting both copies survive the dedup pass in
merge_session_messages_append_only.
Truncating to second-level granularity collapses sub-second drift to
the same key, so the duplicate is suppressed correctly.
Fixes#2616
Two non-functional cleanups from the second Copilot pass:
1. The inline comment in `test_error_marker_no_preserved_as_draft`
said the legacy "user message above was preserved" wording was used
for the post-retry-give-up case. The actual implementation demotes
give-up markers to a different neutral wording ("Partial output may
have been lost."). Comment rewritten to match the contract.
2. The regression test `test_lost_response_recovered_on_second_read`
declared a `monkeypatch` parameter it never used. Dropped.
Four code-review comments from the automated Copilot reviewer on this PR:
1. `_journal_tool_already_present` dedupe was session-wide, so a
legitimately-repeated tool (e.g. a second `terminal: ls` in an
earlier turn) could cause the retry path to falsely skip
materializing the recovered tool card. The helper now takes a
keyword `stream_id` argument; when supplied, a tool card whose
`_recovered_stream_id` is set AND differs from the candidate is no
longer treated as a duplicate. Untagged tool cards (live tools, or
tool cards carried over from a pre-tagging core transcript) still
match, preserving the existing 'core transcript already has this
tool, don't duplicate' invariant. Two new tests in
`TestJournalToolDedupeScoping` cover both legs of the rule.
2./3. The troubleshooting FAQ pointed at `~/.hermes/webui/sessions/session_<sid>.json`
and `~/.hermes/_run_journal/...`. The actual sidecar filename has
no `session_` prefix and the run-journal lives under the WebUI
sessions dir (`~/.hermes/webui/sessions/_run_journal/<sid>/<stream>.jsonl`,
default). Both paths fixed and an explicit note added about
`HERMES_WEBUI_STATE_DIR` overriding the state root.
4. Drop unused `json` / `queue` / `Path` imports from
`tests/test_session_lost_response_regression.py` so the file stops
carrying noise that future linting would flag.
CHANGELOG: append an Unreleased / Fixed entry describing the user-visible
behaviour change (interrupted-turn marker now self-heals on the next
session read; gives up gracefully after 12 retries or 24h).
docs/troubleshooting.md: add a 'Symptom → Why → Diagnostic → Fix →
Caps → When to file a bug' entry for the
'no agent output was recovered' marker so users who hit the lost-response
shape on WSL2 / network FS can recognise it, verify the run-journal on
disk, and know that reloading the session is enough.
nesquena-hermes