* feat: add GLM-5.3 Flash to Z.AI model list
Add glm-5.3-flash next to glm-5.3 in _PROVIDER_MODELS["zai"] and mirror
it in _FALLBACK_MODELS, following the existing glm-4.5 / glm-4.5-flash
sibling pattern. Reasoning-effort gating needs no change: the >= 5.2
version gate already classifies the id at the full effort tier.
Adds a regression suite asserting catalog presence, label, ordering,
fallback entry, and end-to-end propagation through
get_available_models().
* test: pin glm-5.3-flash sibling adjacency in zai provider catalog
Per PR review feedback (#7323): the provider-catalog ordering test now
asserts glm-5.3-flash immediately after glm-5.3 and glm-5.2 immediately
after the flash sibling, matching the adjacency the fallback test
already pins. The base<flash relative check is subsumed; flash<glm-5.1
is kept to pin 5.1 after the whole 5.3 generation.
---------
Co-authored-by: nesquena-hermes <nesquena+hermes@gmail.com>
* refactor(routes): extract the 585-line GET /api/session handler to _handle_session_get
Pure move out of handle_get's if/elif chain into a named module-level handler, matching the pattern the other ~200 routes already use. Body is verbatim (one-level dedent only); AST identity of the moved statements verified; no behavior change — the GET /api/session test slice passes identically before and after.
* test: repoint four source-window tests to _handle_session_get
The moved block's source-window extractors (chat_start synthesiser/read-only
pins, regressions runtime-streaming-state window, issue1436 load-path
markers, security-redaction inspect) now anchor on the extracted handler;
assertions unchanged.
* test: repoint messaging-session loader pin to _handle_session_get
---------
Co-authored-by: nesquena-hermes <nesquena+hermes@gmail.com>
* Fix geometric duplication of recovered reasoning-only rows at turn settle
_restore_display_reasoning_metadata() re-inserted every display-only
reasoning row at its API-safe position in the new result. After context
compaction that position is past the end of result_messages, so the row
was appended after the new reply; _message_identity() is None for an
empty assistant row, so the display merge could not dedup it. Each
settled turn therefore re-added every existing empty
_recovered_from_run_journal anchor once more (1,1,2,4,8,... blocks).
Only restore a reasoning-only row in front of its own API-safe
successor; a compacted result with no aligned slot is skipped.
Refs #7388, #7416 (independent: that PR fixes linear journal replay).
* tests: keep docstrings within the 5-line block budget
* Prefer stable ids when aligning a reasoning row's successor at restore
The successor-alignment guard compared the historical successor with the
aligned current row via _message_identity(), which ignores stable ids and
returns equal keys for repeated prompts ("continue"). A distinct current
turn with the same text false-matched the historical successor, every
recovered anchor before it was re-inserted, and 64 anchors became 128
after one restore->merge settle (nesquena-hermes gate on ed2076eda).
Alignment is now stable-id-first: when either row carries an id, both
must carry the same id; _message_identity() is the fallback only when
neither row has one. Adds a production-order regression through
_settle_result_messages() with 64 anchors, stable ids and repeated
"continue" prompts (RED on ed2076eda and origin/master: 128 != 64), plus
a unit test pinning all four id-presence branches.
* Fail closed on stable-id identity: positive-int only, unique, no content fallback
_restore_display_reasoning_metadata() compared ids with raw equality, so a
persisted historical successor with id=True (or 1.0) matched a distinct
current row minted as id=1 and restored the reasoning card before the
wrong turn. Ids are now accepted only when type(value) is int and > 0,
both rows must carry a valid equal id whenever either has an id field,
and the id must be the sole owner in both API-safe projections.
_assign_stable_message_ids() re-mints invalid ids so they cannot survive
into a collision with minted integers.
* tests: explicit zip strict= for the ruff B905 gate
* Resolve the active-turn boundary before restoration; never carry historical ids/rows into it
A historical assistant successor with the same content as the distinct
current assistant ("Done." / "Done.") let _restore_reasoning_metadata copy
the historical stable id onto the current row; the uniqueness guard then
accepted that forged identity and 64 recovered cards settled to 128.
_active_turn_boundary() is decided once in _prepare_marker_clean_writeback
before any restore (token > strict context prefix > last matching user row
> 0) and threaded through _settle_result_messages into both restores.
Rows at/after it receive no historical id/metadata and no historical
reasoning-only row. Assistant-only results are boundary zero.
* Keep the pinned _restore_reasoning_metadata signature; boundary goes through _restore_reasoning_metadata_before_boundary
tests/test_sprint49.py pins the two-arg signature as source text, so the
boundary-aware core lives in _restore_reasoning_metadata_before_boundary
and the original name stays a thin no-boundary wrapper.
* Content-only prefix is not turn ownership: fail closed to boundary 0; sync route passes result/Agent turn authority
- _active_turn_boundary: a previous-context prefix matched by _message_identity
(ids/timestamps ignored) only narrows where the current user may be; the
boundary is the prompt-matching user AT OR AFTER that prefix, else 0.
- _handle_chat_sync: resolve authority via _resolve_active_turn_authority
(result, agent) instead of hard-coding None.
- Regressions: three-row current-only collision (boundary 0, one 64 block),
exact-authority + append-only controls, ID-less-successor settle/save/load
pinned at the persisted-display boundary, sync-route authority pass-through.
---------
Co-authored-by: carlotestor <carlotestor@users.noreply.github.com>
Co-authored-by: nesquena-hermes <nesquena+hermes@gmail.com>
* rebase #7351 onto master (bind turn workspace as agent session cwd)
Co-authored-by: samfoy <samfoy@users.noreply.github.com>
* fix test: use deployed agent.runtime_cwd API (_SESSION_CWD.get) not nonexistent _session_cwd_override
The two failing assertions referenced agent.runtime_cwd._session_cwd_override(), a
helper not present in the installed Agent (which exposes _SESSION_CWD, resolve_agent_cwd,
set/clear_session_cwd). The production binder is correct and verified: binding workspace=/tmp
sets _SESSION_CWD.get()=='/tmp' and resolve_agent_cwd()=='/tmp', reset restores _UNSET.
Line 343 was redundant with the surrounding assertions; line 386 now reads _SESSION_CWD.get().
Co-authored-by: samfoy <samfoy@users.noreply.github.com>
* docs(changelog): stamp #7351 turn workspace cwd binding
---------
Co-authored-by: n <a@n>
Co-authored-by: samfoy <samfoy@users.noreply.github.com>
* rebase #7160 onto master (report unverified Agent updates, no auto-restart)
Co-authored-by: hrdwdmrbl <hrdwdmrbl@users.noreply.github.com>
* harden agent-update marker read: no-follow, fstat regular-file, bounded (fixes gate CORE)
The Agent update marker is attacker-adjacent shared state. _read_live_agent_update()
used Path.read_text(), which follows symlinks, blocks forever on a FIFO, and reads an
unbounded regular file — a rejected/stale-runtime request could hang or OOM. Now opens
O_RDONLY|O_NONBLOCK|O_NOFOLLOW, fstat-verifies a small regular file, and reads a bounded
max, classifying anything else as 'unknown' (fail-closed). Adds FIFO/oversized/symlink/
happy-path regression tests.
Co-authored-by: hrdwdmrbl <hrdwdmrbl@users.noreply.github.com>
* fix marker-read portability: gate os.open fast-path on O_NONBLOCK+O_NOFOLLOW
os.O_NONBLOCK is Unix-only; accessing it unconditionally raised AttributeError
on native Windows, breaking every stale-revision barrier (chat/compression/etc).
Now both flags resolve via getattr and the os.open() fast path is taken only
when BOTH are available (_MARKER_SAFE_OPEN_AVAILABLE); otherwise fall back to an
lstat-only classification (absent vs unknown) so a fallback platform can never
symlink-traverse or crash. Adds a Windows-fallback regression test.
Co-authored-by: hrdwdmrbl <hrdwdmrbl@users.noreply.github.com>
* docs(changelog): stamp #7160 stale-runtime report + hardened marker read
---------
Co-authored-by: n <a@n>
Co-authored-by: hrdwdmrbl <hrdwdmrbl@users.noreply.github.com>
Preserve nested request isolation, make the response allowlist single-source,
and strip heavy fields from existing index rows after reconciliation.
Co-authored-by: Martin Dell <martindell@users.noreply.github.com>
* fix(streaming): lock-disciplined STREAMS reads via peek_stream()
Writers pop STREAMS under STREAMS_LOCK, but four hot reads used bare
STREAMS.get(): the streaming worker start, the gateway run start, and the two
SSE attach paths in routes.py. Each is individually saved today by GIL-atomic
dict.get plus a None-guard, but the check-then-act windows (get → register,
get → subscribe) race teardown pops, and the bare pattern invites unprotected
copies. peek_stream() centralizes the locked read; all four sites migrate with
no behavioral change.
* test: dual-patch config.STREAMS alongside routes.STREAMS in SSE seam tests
peek_stream() reads the canonical api.config.STREAMS registry; tests that
inject a fake registry by patching only the routes-level re-export left the
helper reading the real (empty) registry and hung the SSE loop to the
pytest-timeout ceiling. Mirror the test_issue6623 pattern: patch both
bindings.
* docs(changelog): stamp #7339 lock-disciplined STREAMS reads
---------
Co-authored-by: Sebastian B <sb@wmc.sh>
Co-authored-by: n <a@n>
Gate finding on #7336: read_body() rejected whitespace-only bodies that
worked on master (a body-optional DELETE/PATCH/PUT can send b' \n'), returning
400 Invalid JSON body. Normalize a fully-whitespace body to {} before json.loads,
matching the pre-#7336 lenient behavior for that shape. Adds a parametrized
regression test.
Co-authored-by: zicochaos <zicochaos@users.noreply.github.com>
Observed: on a multi-profile install where one profile's state.db predates
the agent's `source` column, errors.log gained one identical line every
~15 s (276 lines in ~70 min):
api.models WARNING agent session listing skipped: state.db at
/data/hermes/profiles/planner/state.db has no 'source' column (older
hermes-agent?). Agent sessions unavailable. Upgrade hermes-agent to
fix this.
Mechanism: api/agent_sessions.py read_importable_agent_session_rows()
runs `PRAGMA table_info(sessions)` and warns unconditionally when the
column is absent (api/agent_sessions.py:564 before this change).
api/models.py get_cli_sessions(all_profiles=True) calls it for every
profile on every sidebar poll, behind only the 5 s
_CLI_SESSIONS_CACHE_TTL_SECONDS cache, so the same DB re-emits the same
warning on every cache miss. The condition is a property of the DB
file, not of the poll.
Fix: a module-level `_SOURCE_COLUMN_WARNED_DB_PATHS: set[str]` keyed by
the resolved db_path. The warning fires the first time a given path is
seen in the process and is suppressed after that; a different path
warns on its own. Message text is unchanged. Process-lifetime only: a
restart warns again, which is intended (the line remains the operator's
cue that the agent still needs upgrading). Plain set mutation under the
GIL is sufficient; a duplicate line from two racing first calls is
harmless.
Tests (tests/test_issue634.py, real pre-`source` sqlite DBs under
tmp_path, asserting on captured log records):
- same db_path called twice -> exactly one WARNING (failed before:
`assert 2 == 1`)
- two db_paths, then the first again -> exactly two WARNINGs, one per
path (failed before: `assert 3 == 2`)
The seven existing #634 source-shape tests still pass; neighbours
test_issue3762/3930/5455/1494 pass.
Deliberately not changed: the exception-path warning in
get_cli_sessions() (api/models.py), the read-only-open fallback warning,
and the 5 s cache TTL. The operator-side root cause (the agent
reconciler creating a pre-`source` state.db) is a separate upstream fix;
this only stops the repetition.
Title generation via the aux client ran without the Agent's ambient
conversation context, so relay targets that key on it (OpenCode Go)
rejected the completion call with HTTP 400.
Publish the webui session id as the Agent conversation context around
the aux completion (restoring it in a finally block) and pass it from
all three title entry paths: background update, background refresh and
the on-demand regenerate helper.
Refs #7470
Two Codex adversarial-security-gate findings on the batch:
1. #7331 CORE (backward-compat break): making '|' an unconditional trusted-groups
separator would re-split an existing deployment whose group NAME legitimately
contains a literal '|' (e.g. 'unprivileged|admins' read as one group), silently
changing its profile binding. Pipe splitting is now OPT-IN via
HERMES_WEBUI_TRUSTED_GROUPS_PIPE_SEPARATOR; comma/newline stay the default.
Tests: literal-pipe preserved by default, pipe splits only with the flag set.
2. #6772 follow-up (my own hardening introduced it): rebuilding the origin from
parsed parts dropped the brackets around an IPv6 host, so 'http://[::1]:8787'
became 'http://::1:8787' and never matched the browser's bracketed
clientDataJSON.origin -> passkey login over IPv6 would fail. Re-bracket IPv6
literals. Added rp_context regression tests (Origin-first, port preserved,
IPv6 bracketed, malformed->Host fallback, userinfo stripped).
Follow-up on #6772. Preferring Origin over Host for the WebAuthn RPID is correct
(the RPID must match the page origin, not the backend's), and it is not a trust
decision: the authenticator only releases a credential whose RPID is a registrable
suffix of the real page origin, clientDataJSON.origin must equal the stored origin,
rpIdHash must match the stored RPID, and the assertion signature is verified against
the stored credential public key.
Still, only accept a syntactically valid http(s) Origin carrying a hostname, and
rebuild the stored origin from the parsed scheme/host/port instead of echoing the raw
header — so a malformed, non-http (javascript:/file:), or userinfo-bearing Origin
falls through to the Host-derived path rather than being stored verbatim.
The Hermes Agent's state-module refactor (agent commit 53db597201, released
v2026.9.7) renamed SessionDB.set_auto_title_if_empty -> set_auto_title(
session_id, title, *, source) and (agent commit 14791b4d4e) dropped the
tools.approval._ApprovalEntry back-compat facade (class moved, unchanged, to
tools.approval_gateway_wait).
Production impact (real, silent): api.state_sync.sync_session_title called the
removed set_auto_title_if_empty, whose AttributeError was swallowed by the
surrounding `except Exception: logger.debug(...)`, so WebUI auto-generated
session titles stopped persisting to state.db on agent >= v2026.9.7 — hermes
sessions list went blank again (the exact #6892 regression). Now calls
set_auto_title with LLM provenance, which preserves the identical semantics
(only fills NULL/lower-authority title, returns False untouched on a manual
rename, ValueError -> get_next_title_in_lineage #N de-dup).
Test fixes (agent-coupled; skip on GitHub CI which installs no agent, so they
only surface on an agent-equipped box like our gate):
1. tools.approval._ApprovalEntry -> idempotent autouse conftest compat shim
backfilling the class from its post-split location only when the installed
agent lacks the dropped facade. No production code imports it.
2. test_goal_command_webui::test_profile_goal_falls_back_when_session_db_path_is_frozen
was missing the monkeypatch.syspath_prepend(...) that every sibling test in
the file has, so its bare `import hermes_state` failed under full-suite
ordering after a predecessor stripped the agent dir from sys.path. Added it.
Verified: authoritative 5-shard CI-equivalent split all green (0 failed);
each failure proven pre-existing on clean master (not this diff).
The not-found reply capped available_skills at 20 with no total and no
truncation marker, so a caller with more skills than the cap was handed a
partial list that reads as the complete set. A skill sorting past the cut
looked uninstalled.
Keep the bound; stop it lying. The payload now carries total_skills and
available_skills_truncated, and the hint names how many of how many were
shown, matching the (list, truncated) shape already used for capped commit
subject lists in api/updates.py.
hermes-agent v0.21 removed agent.anthropic_adapter.normalize_anthropic_response and the AIAgent._normalize_codex_response method; the sanctioned out-of-band contract is now agent._get_transport(<mode>).normalize_response(...) returning NormalizedResponse (.content). Port the anthropic_messages and codex_responses branches of generate_title_raw_via_agent (api/streaming.py) and _agent_text_completion (api/routes.py); on agent v0.21 those branches currently die on import/attribute errors and silently fall back, degrading title generation and handoff summaries. build_anthropic_kwargs, _anthropic_messages_create, and _run_codex_stream are unchanged in v0.21 and stay as-is.
Named custom providers are only routable when the slug resolves to a real,
unique custom_providers[] entry. A stale session provider such as
'custom:missing' must not be minted into an @custom:missing:... route -
resolve_model_provider() would take the named-provider lane and find no
matching endpoint. _unique_custom_provider_entry() returns None for unknown
slugs and raises AmbiguousCustomProviderError for collisions, matching the
point-of-return guard used by resolve_model_provider.
Addresses maintainer review on #7356. Adds two regression tests: unknown
named custom slug stays bare, ambiguous normalized slug fails closed.
model_with_provider_context() passed any slash-bearing model ID through
bare when the session provider was not OpenRouter/ACP/plugin and not
listed under config.yaml -> providers:. That dropped the hint for portal
providers (nous, nvidia, opencode-zen, opencode-go) and named custom
providers (custom:<slug>), so resolve_model_provider() fell back to the
profile default provider + base_url and sent a vendor-scoped id to the
wrong endpoint (HTTP 404, e.g. upstage/solar-pro4:free to api.x.ai).
Emit an explicit @provider:model hint for any known routable provider
that differs from the configured default; keep the bare passthrough for
unknown/ambiguous provider slugs so custom/proxy base_url routing stays
in charge. Closes#7333.
`_strip_compact_echo_suffix` probed every candidate cut index and re-folded
the whole remaining tail for each probe, so the cost grew with
`window x suffix` rather than with the data actually inspected. It runs up to
four times per interim assistant message, on the reasoning buffer, the shared
stream mirror and up to two per-message segments.
The practical effect is worst on the longest message of a turn: 297ms for a
200-character message but 5.6s for a 6000-character one, measured on the live
process. That time is spent holding the GIL, so a single long final message
also stalls every other stream served by the same worker. Profiling a live
turn attributed ~5% of total CPU to this one helper.
Fold the window once, then walk backwards across exactly as many non-space
characters as the folded suffix holds. `str.isspace` and the `\s` pattern used
by `_compact_for_echo_compare` agree on all 1114112 Unicode code points, so
the folded view and the walk stay consistent.
Behaviour is unchanged, including the leftmost-cut tie-break on repeated text
and the bounded search window. The new test pins equivalence against a naive
reference oracle over ~2020 cases (exact, partial and absent echoes, mixed
whitespace, unicode, empty and None inputs, repetitions, out-of-window
echoes), asserts the fold no longer runs once per cut index, and caps the
runtime of a long-conclusion call.
Measured: 200 chars 297ms -> 0.149ms, 6000 chars 5645ms -> 0.720ms.
read_body() silently turned malformed JSON into {} (clients then got a
misleading 'Missing field' 400) and passed non-dict JSON through to handlers
where body.get() raised AttributeError and surfaced as a generic 500.
Raise ValueError instead — handle_post already maps it to a clean 400 — and
give the PATCH/DELETE/PUT dispatchers the same mapping they lacked. Empty
bodies keep the documented {} default.
Authentik's proxy provider on this deployment joins multiple group
names with "|" instead of "," (e.g. "admins|developpeur"). The
comma-only split in _trusted_groups_header_value() treated that whole
string as a single unmatched group name, silently dropping the
session to the unbound "default" profile even for identities that
are legitimate members of a mapped group. Single-group identities were
never affected since there was no separator to parse.
Accept comma, pipe, and newline as equivalent separators so the parser
works regardless of which the outpost happens to send.
* perf(session-load): memoize the compression-lineage display stitch
_webui_sidecar_lineage_messages_for_display re-merged multi-thousand-row
parent+child transcripts on EVERY GET /api/session (measured 5.1s per
request on a real 5.2k-message lineage, py-spy: merge_session_messages_
append_only + _matching_visible_duplicate dominate). The merge result is
now cached per session id, validity keyed by the exact stat signature of
the child sidecar AND every snapshot parent involved, bounded LRU(16).
Any write to any involved sidecar invalidates naturally; cache hits and
the populating call hand out shallow-copied rows so caller-side display
metadata cannot corrupt cached state.
Measured: 5.17s cold -> 0.002s warm on the real session; invalidation
after touch re-merges. Tests: tests/test_lineage_display_cache.py
(correctness, copy isolation, child/parent invalidation, no pollution
for lineage-less sessions).
* perf(api): memoize inactive-session display merges
* perf(api): skip state load on cached transcript reopen
* test(api): remove unused lineage-cache import
* fix(api): reject streaming key without state database
* test(api): isolate display cache fixture
* fix(api): bind cache publication to read signature
* fix(api): harden display cache invalidation
* fix(api): bind cache hits to lineage provenance
* fix(api): close remaining display cache scope races
* fix(api): scope display cache revision to target
* fix(api): evict unverifiable lineage provenance
* fix(api): keep incomplete lineage provenance uncached
---------
Co-authored-by: Anthony Ruiz <ruizanthony@users.noreply.github.com>
_split_webui_provider_model_value() and _strip_webui_provider_prefix()
split provider-qualified model values at the last colon. That amputates
any model name carrying its own tag: @ollama:qwen3.8:27b-mtp-q8_0
resolved to model 27b-mtp-q8_0 with provider ollama:qwen3.8, and the
truncated name was sent upstream, which 404'd. The splitter also cannot
distinguish a tagged model from a multi-segment custom provider ID, so
@custom:mykey:model-a:free was affected the same way.
Both sites now delegate to config._parse_provider_qualified_model_id(),
the grammar routes.py and the gateway adopted in #6723. The import is
function-local, matching the existing api.config imports in this module.
_split_webui_provider_model_value() persists its result into profile
config, so the mangled name outlived the session that produced it.
Fixes#7182.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: nesquena-hermes <nesquena+hermes@gmail.com>
webtecnica