Constituent PRs:
#1637 by @Michaelyklam — protect raw pre from glued-bold lift (closes#1451)
#1639 by @bergeouss — macOS auto-scroll race + custom:* provider list (closes#1360, #1619)
#1642 by @nesquena-hermes — YAML/JSON/diff code block newlines (closes#1618, #1463)
Opus advisor SHIP verdict on stage-295. One observation absorbed:
- api/config.py:2533 dead-code comment per Opus (defensive belt-and-braces
for #1619 fallback; load-bearing fix is in routes.py /api/models/live)
PR #1641 (Michaelyklam parallel-discovery duplicate of #1642) closed as
superseded; UI media adopted with co-author trailer.
4245 → 4255 tests passing (+10).
Adopt the UI media from @Michaelyklam's parallel-discovery PR #1641 which
shipped the same one-character regex relax fix for #1618. PR #1641 is
being closed as superseded by #1642 (which carries nesquena APPROVED +
322 LOC test suite); preserving Michael's UI evidence here so the visual
proof of the fix lives in-tree alongside the canonical PR.
Co-authored-by: Michael Lam <Michaelyklam1@gmail.com>
test_scroll_listener_hides_button_when_pinned checked 300 chars after
el.addEventListener('scroll', but the rAF-debounce fix moved the
scrollToBottomBtn logic into the requestAnimationFrame callback,
beyond the 300-char window. Extended to 600 to cover the full block.
#1360 — On macOS WKWebView, trackpad momentum scrolling fires scroll
events that interleave with the _programmaticScroll setTimeout(0) guard.
A mid-momentum scroll event either gets swallowed (_programmaticScroll
still true) or falsely reports nearBottom (momentum hasn't settled),
keeping _scrollPinned=true and snapping the viewport back down.
Fix: rAF-debounce the scroll listener so the nearBottom check runs at
the next paint frame when the browser's scroll position has settled.
Added a hysteresis counter requiring 2 consecutive near-bottom samples
before re-pinning, preventing accidental re-pin during deceleration.
#1619 — When a custom:* provider (e.g. custom:relay via custom_providers)
has models that overlap with auto-detected models from base_url /v1/models,
the dedup logic at config.py:2263 skipped them all. The named custom
group ended up empty, and the continue at line 2334 silently discarded
the auto-detected models. Result: only the default model appeared.
Fix 1 (config.py): When custom:* named group has 0 models after dedup,
fall back to auto_detected_models_by_provider instead of dropping them.
Fix 2 (routes.py): Extended /api/models/live fallback to handle
custom:* slugs (not just bare "custom") for both custom_providers
config lookup and base_url live fetch.
Closes#1618 (reported by @Zixim) and corrects #1463's previous fix.
Bug: YAML, JSON, and diff/patch fenced code blocks render flattened to a
single line. Reporter noted the bug persisted v0.50.279 -> v0.50.291 ->
v0.50.292 despite PR #1516's CSS-only "fix".
Root cause: PR #484 (v0.50.237) added a JSON/YAML tree-viewer that routes
those languages through <div class="code-tree-wrap">...<pre class="tree-raw-view">
instead of bare <pre>. Same release added the diff/patch coloring path
that emits <pre class="diff-block">. The _pre_stash regex at
static/ui.js:1914 matched only literal <pre> with no attributes:
<pre>[\s\S]*?<\/pre>
Both new shapes failed to match, fell through to the paragraph-wrap pass,
and \n characters inside the code blocks got replaced with <br> tags
inside <code>. By the time Prism ran, no newlines remained for the CSS
rule (PR #1516, language-yaml .token { white-space: pre !important }) to
preserve.
Fix: relax the regex to accept any attribute on <pre>:
<pre>[\s\S]*?<\/pre> -> <pre[^>]*>[\s\S]*?<\/pre>
One regex character. Pulls JSON, YAML, and diff/patch blocks into the
stash so paragraph-wrap can't mangle them. Bash, Python, Go, etc. were
never affected because they emit bare <pre>.
Tests: 9 new (2 source-string invariants + 7 behavioural via node-driver
against the actual static/ui.js renderMd()). 6 of the 7 behavioural tests
fail on master and pass with the fix; the 3 sanity checks (yml-alias,
bash, mermaid) pass on both.
Plus widened source-scan window in 3 pre-existing test_745 assertions
from 400 to 1500 chars. The new comment block above the fixed regex
pushed it past the previous scan window. Pure window-narrowness bug,
not a behavior regression.
4245 -> 4254 passing.
Closes#1633. STATE_DIR/models_cache.json was persisted across server
restarts without any version stamp, so a Docker container update from
version A to B read the cache file written by version A — users saw
stale picker contents (missing models, phantom provider groups) for
up to 24 hours until either the TTL expired, an unrelated provider
edit triggered invalidate_models_cache(), or they manually deleted
the file.
Reporter Deor (Discord) updated to v0.50.292 — which contained fixes
for #1538, #1539, and #1568 — did a hard refresh and cleared site
data, and still saw byte-for-byte identical picker contents because
the server kept reading the v0.50.281 cache file off the host-mounted
state volume.
Fix:
* _save_models_cache_to_disk() stamps payloads with _webui_version
(resolved lazily from api.updates.WEBUI_VERSION via sys.modules
lookup to avoid the api.config <-> api.updates circular import)
and _schema_version = 2.
* New _is_loadable_disk_cache() validator checks both stamps in
addition to shape. Mismatch on either field rejects the load.
* _load_models_cache_from_disk() calls the new validator and
strips the disk-only metadata before returning, so the rest of
the code sees the same shape it always did.
* _is_valid_models_cache() kept loose (shape-only) so in-memory
cache writes that never touch disk don't fail validation.
Schema version is independent of the WebUI version stamp so future
cache-shape changes can invalidate older releases without relying
on a tag bump alone.
Early-init edge case (api.updates not yet loaded) skips the version
check rather than wedging the boot — at worst an unstamped file is
written once and rejected on the next call.
Updated existing tests/test_model_cache_metadata.py to use subset/
round-trip semantics rather than byte-for-byte equality, since the
disk payload now has additional stamps. The four response-shape
fields still round-trip verbatim; the load result is unchanged
(stamps stripped). 19 new regression tests.
4180 -> 4199 tests pass.
#1430 — renderSessionList() had no staleness guard. Multiple concurrent
callers (message send, rename, session switch) could race, allowing a
slower older API response to overwrite _allSessions with stale data.
Added a generation counter that increments on each call and discards
responses from superseded generations.
#1470 — docker_init.bash unconditionally called groupmod/usermod even
on read-only root filesystems (podman with read_only=true). Added a
writability check for /etc/group and /etc/passwd. If read-only and
UID/GID already match, the mod is skipped gracefully. If they don't
match, a clear error message suggests setting matching IDs or disabling
read_only mode.
SHOULD-FIX: rate-limit _repair_stale_pending repair-firing telemetry. Switch
from unconditional logger.warning to age-keyed: WARNING when pending_age <
5min (the diagnostically valuable race window — actual leak-path candidates
that slipped past the grace guard) and DEBUG for the long-tail (orphaned
sidecars from prior process lifetimes). Prevents reconnect loops on stuck
sessions from flooding the log while preserving the diagnostic signal we
want for tuning _REPAIR_STALE_PENDING_GRACE_SECONDS empirically.
NIT: _LOCAL_SERVER_PROVIDERS expanded with lm-studio (hyphenated alias used
in some custom_providers configs and already recognized at api/config.py:2189
for SSRF host trust) and localai (LocalAI project). Test parametrize expanded
from 7 to 11 names, also covering pre-existing koboldcpp and textgen for
symmetry. +4 regression tests.
NIT (docs): CHANGELOG callout for the RFC1918 behavior change. Internal-
network OpenAI-compatible proxies now preserve the model prefix on private-IP
base_urls. Documented the migration path: configure as a custom_providers
entry to bypass the local-server detection.
NIT (deferred, optional): narrowing the heuristic to is_loopback only is
left as future work; the broader scope was an explicit goal in the bug
body and Opus flagged it as SHOULD-DISCUSS-but-not-block.
4184 -> 4188 passing. 0 regressions. ~10 LOC absorbed total.
Closes#1623 — Lower SSE app heartbeat from 30s to 5s at every long-lived
handler (main agent, terminal, gateway-watcher, approval-poller, clarify-poller).
Kernel TCP keepalive declares peer dead at 25s worst-case (10s KEEPIDLE +
5s KEEPINTVL * 3 KEEPCNT, added v0.50.289 #1581). 30s app heartbeat let the
kernel tear sockets down on flaky networks before the app sent its first
keepalive byte — drops at ~10s during long thinking phases. New named
constant _SSE_HEARTBEAT_INTERVAL_SECONDS=5; regression test pins the
inequality (app_heartbeat * 2 <= kernel_window) so future tuning can't
re-introduce the misalignment.
Closes#1624 — Add 30s grace period to _repair_stale_pending() trigger.
Without it, any narrow race between the streaming thread clearing
pending_user_message and STREAMS.pop(stream_id) produces a false-positive
'Previous turn did not complete.' marker on a turn that finished correctly
(reproducible after every command-approval turn). Defense-in-depth, not
the root-cause fix — the actual streaming-thread leak path is tracked
separately. Falsy pending_started_at (legacy sidecars) treated as
'old enough' so legitimate legacy-data recovery still works. Plus
logger.warning telemetry on every legitimate repair so the next batch of
user reports tells us whether the underlying race still fires.
Closes#1625 — Local model servers (LM Studio, Ollama, llama.cpp, vLLM,
TabbyAPI, koboldcpp, textgen-webui) now keep the full HuggingFace-style
model id (e.g. 'qwen/qwen3.6-27b' instead of stripped 'qwen3.6-27b'). New
_LOCAL_SERVER_PROVIDERS set + _base_url_points_at_local_server() loopback/
RFC1918 heuristic — either signal triggers no-strip. Backward compat
preserved for OpenAI-compatible proxies on public hosts (LiteLLM at
litellm.example.com still strips openai/gpt-5.4 -> gpt-5.4). Updated the
existing #230/#433 test to reflect that #1625 supersedes the strip-on-custom
rule for loopback hosts (see api/config.py and test_model_resolver.py
docstring update). Reported by @akarichan8231 in Discord on 2026-05-04.
42 regression tests across:
tests/test_issue1623_sse_heartbeat_alignment.py (3)
tests/test_issue1624_repair_stale_pending_grace.py (9)
tests/test_issue1625_local_server_model_id_preservation.py (30)
4142 -> 4184 passing. 0 regressions.
SHOULD-FIX #1 (renamed-root client cross-alias): drop strict-equality client
filter at static/sessions.js:1853. Server-side _profiles_match cross-aliases
'default'-tagged rows to a renamed root 'kinni'; the strict-equality client
would reject them, dropping every legacy session for renamed-root users. The
server is now solely authoritative for profile scoping.
SHOULD-FIX #2 (messaging-source dedupe ordering): _keep_latest_messaging_session_per_source
now runs AFTER the profile filter at api/routes.py:2078. Before, it ran on
the merged-cross-profile list with profile-blind keys, discarding the older
profile's row across profiles before the scope filter — leaving zero rows for
any messaging identity the active profile shared with another profile.
NIT #3: _projects_migrated flag now set only AFTER successful save_projects.
NIT #4: cleaned dead test code in test_is_root_profile_invalidation_drops_stale.
NIT #5: _create_profile_fallback's clone_from=='default' literal now routes
through _is_root_profile() for parity with the 5 other callsites.
+2 regression tests pin the SHOULD-FIX shapes:
- test_keep_latest_messaging_runs_after_profile_filter (source-string ordering)
- test_static_sessions_js_trusts_server_profile_scoping (no client re-filter)
4173 -> 4175 tests pass. 0 regressions.
Closes#1611 — /api/sessions filters by active profile by default; ?all_profiles=1
opt-in for aggregate views; new _profiles_match() helper honours renamed-root
cross-aliasing; static/sessions.js drops the s.is_cli_session bypass; toggle-on
re-fetches with all_profiles=1 instead of slicing client-cached rows.
Closes#1612 — new _is_root_profile() central helper consults list_profiles_api()
for is_default=True matches alongside the legacy 'default' alias. Replaces five
literal-default callsites in api/profiles.py. Memoized with explicit invalidation
hooks at create + delete. Sticky active_profile file write now stores '' for
renamed root, consistent with the legacy empty==root contract.
Closes#1614 — projects carry a profile field stamped at create-time;
/api/projects filters by active profile; /api/projects/{create,rename,delete}
and /api/session/move reject ops on cross-profile projects with 404; new
_PROJECTS_MIGRATION migration in load_projects() back-tags untagged projects
from any session that uses them, fall back to 'default'; ensure_cron_project
keys lookup by (name, profile) so each profile gets its own Cron Jobs project.
31 regression tests (9+11+11) pin the renamed-root resolution, server-side
profile scoping shape, helper invariants, cross-alias matching, migration
behavior, and active-profile guards on every project mutation endpoint.
4148 tests pass.
Reporter: @stefanpieter
Co-authored-by: stefanpieter <noreply@github.com>
When the clipboard carries both text and an image (rich-text sources like
Notes, Word, Slack, browser selection attach a rendered preview alongside
the plain text), the paste handler in static/boot.js unconditionally
called e.preventDefault() and routed the image into addFiles(), silently
discarding the text payload.
Fix:
- Detect text in the clipboard via items[].kind === 'string' &&
(type === 'text/plain' || type === 'text/html'). When present, return
early so the browser's default text-paste runs.
- Tighten the image filter to kind === 'file' && type.startsWith('image/')
so string items advertising an image MIME (e.g. text/html with an
embedded data URI) are not misclassified as a true screenshot paste.
Pure-screenshot paste (image-only clipboard, e.g. Cmd+Shift+Ctrl+4 on macOS)
is unchanged.
Adds tests/test_1620_paste_text_with_image.py with 6 static-analysis checks
on the handler shape, matching the pattern of test_issue1095_pasted_images.py.
- Delete QuietHTTPServer.server_bind() override entirely:
TCP_KEEP* setsockopts on the listening socket are no-ops without
SO_KEEPALIVE, and SO_REUSEADDR=1 is already set by the parent class.
The actual fix lives entirely in Handler.setup().
- Restructure Handler.setup() with per-platform branches so
SO_KEEPALIVE=1 is always applied before timing params, and macOS
(TCP_KEEPALIVE) gets keepalive instead of aborting on TCP_KEEPIDLE.
Remove the try/except Exception wrapper around
cron_profile_context_for_home(...).__enter__() in _run_cron_tracked.
A silent fallback to ctx=None would leave the worker thread unpinned
against process-global HERMES_HOME, silently corrupting cross-profile
state — the same class of bug as #1573.
Add regression test to catch any future re-introduction.
nesquena-hermes