Direct tool HTTP calls already identified as Hermes-Agent, but the main
OpenAI-SDK chat path still sent OpenAI/Python. Set Hermes-Agent/<ver> for
api.x.ai clients (xai + xai-oauth) so normal text traffic is attributed correctly.
Port from Kilo-Org/kilocode#12508: a stray read-only state.db / -wal /
-shm (sudo run, restored backup, copied dotfiles) previously killed
SessionDB init with an opaque 'sqlite3.OperationalError: attempt to
write a readonly database' raised from deep inside _init_schema —
naming no file and no fix — and the obvious wrong 'fix' (deleting the
-wal) silently loses committed transactions.
New preflight_db_writability() runs before the first connection on both
DB open paths (SessionDB.__init__ and hermes_cli.kanban_db.connect):
- files inside the Hermes home tree are repaired with chmod u+rw (the
safe scope: Hermes owns them, and the OS makes chmod fail on files
the user doesn't own, which bounds the repair exactly);
- anything else (root-owned files, read-only mounts, custom paths)
fails fast with an error naming the exact file and the exact chmod
command, plus an explicit 'do NOT delete the -wal' warning;
- WAL sidecars are never deleted or truncated — once writable, the
normal open path checkpoints committed frames into the DB.
Proven live on main first: chmod 444 state.db -> SessionDB() raises the
opaque readonly error. With the fix: in-home DBs self-heal; out-of-home
DBs get the actionable message. Sabotage run confirms the integration
tests fail without the wiring (2 failed / 10 passed).
A /steer redirect during a thinking phase serialized the streamed
reasoning into the persisted assistant checkpoint ('Reasoning shown
before the interruption: ...'). An assistant turn exposing its own
chain-of-thought reads to Anthropic's output classifier as
reasoning-injection/prefill jailbreak, so every subsequent call on the
session deterministically returned 'Provider returned an empty
response' — and because the checkpoint is persisted and replayed, no
retry, nudge, or empty-recovery branch could ever escape it. Four
sessions were permanently bricked this way in the week of Jul 21-27
(42+ blocked calls; every reasoning-free checkpoint that week was
untouched — same mechanism as the prefill.json incident).
Class fix: streamed reasoning is now display-only state. The
_current_streamed_reasoning_text accumulator is removed entirely
(producer in _fire_reasoning_delta, resets, and init), so no future
path can serialize CoT into replayable content. The checkpoint keeps
only the visible response text; the model regenerates its reasoning on
the retried turn. Invariant documented in _apply_active_turn_redirect.
Regression tests: CoT never appears in either checkpoint shape,
reasoning-only interrupts produce a bare checkpoint, reasoning deltas
stay display-only.
Tool-schema assembly at CLI/Desktop startup runs the browser-family
check_fns (browser, browser_cdp, browser_dialog, browser_vision). Each
of those gates called _get_cdp_override(), which resolves the configured
endpoint over HTTP (GET /json/version, timeout=10) — so a *stale*
browser.cdp_url pointing at a dead debug browser cost ~7 serial blocking
socket connects before the banner rendered. Measured on a real Windows
install with a dead http://[::1]:9222 config: 15.1s of an 18s launch,
with no warning or error — just mystery slowness. The value is easy to
leave behind: /browser connect writes a session-scoped env override, but
'hermes config set browser.cdp_url' persists forever while the debug
Chrome it pointed at dies on the next browser restart.
Split the helper:
- _get_cdp_override_raw() — returns the configured value (env var or
config.yaml) with zero network I/O. Used by every is-it-configured
gate: check_browser_requirements, _browser_cdp_check, _is_local_mode,
_is_local_backend, _navigation_session_key, _should_inject_engine
(via _is_local_mode), and the hermes doctor chromium-skip check.
- _get_cdp_override() — unchanged contract (raw + /json/version
resolution), now only called on paths that are about to connect:
session creation and the dialog-supervisor attach.
This follows the existing rule in check_browser_requirements ('do not
execute agent-browser --version here') and the browser.manage status
path, which already banned _get_cdp_override for exactly this reason
(test_browser_manage_status_does_not_call_get_cdp_override): schema
assembly must not perform blocking I/O.
A/B on the same machine, same dead endpoint: get_tool_definitions()
15.08s unpatched -> 1.89s patched, with browser_cdp/browser_dialog
still advertised (gate now keys off configuration, not reachability —
matching the documented lazy-supervisor contract in
_browser_dialog_check).
Adds a regression test asserting the browser_cdp check_fn never touches
the network.
Sessions with no cwd — or whose folder can't be promoted to a project (the
bare home dir, a deleted workspace, HERMES state) — were dropped from the
project tree entirely, so the grouped sidebar silently showed fewer chats
than flat Recents. Collect them into a synthetic `__no_project__` node at
the head of the list. It carries one lane purely to hold the rows, and is
omitted when empty so a project-less install stays blank.
The fuzzy branch of `complete.path` ranked basenames from
`_list_repo_files`, which lists files only, so a directory was only ever
reachable by typing a `/` — `@Desktop` returned nothing at all. Rank each
ancestor directory alongside the files, and break same-tier ties toward
the folder so `@docs` leads with `docs/` rather than `docs.md`.
Outside a git repo the fallback `os.walk` compounded this: it can spend
the whole `_FUZZY_CACHE_MAX_FILES` budget inside one deep subtree before
reaching a sibling, hiding top-level folders entirely. Seed the scan with
a `listdir` of the root so immediate children are always candidates.
_peel_moa_guidance hand-implemented the inverse of moa_loop's
_attach_reference_guidance from a different module — a drifting separator
or shape would make the peel silently no-op and put the last cache
breakpoint on the turn-varying guidance block (the #72626 bug class).
Move the inverse into moa_loop.peel_reference_guidance directly adjacent
to the attach, keep a thin wrapper in conversation_loop, and pin the
contract with a round-trip test over all three attach shapes.
Also fix the empty-list residue: peeling a guidance-only content part now
drops the whole message (mirroring the appended-user-message shape)
instead of leaving an empty-content user turn behind.
guidance=None is a real prepared shape (all references failed / silent
degraded policy builds prepared_request without attaching guidance), and
the MoA facade sends prepared['messages'] — not api_kwargs['messages'].
Gating the rebase on 'and guidance' left the stale decoration in the
prepared object for the no-guidance MoA sub-path, so #72626 persisted
there. rebase_prepared_request already handles falsy guidance (copies
messages, skips the attach).
The static-prefix reconstruction pattern (build_system_prompt_parts ->
['stable'] -> startswith gate -> fail-open) existed in three copies:
session restore (conversation_loop), compression keep-prompt path
(conversation_compression), and the new failover redecoration helper.
Hoist it into agent/system_prompt.reconstruct_static_prefix and call it
from all three sites.
Also memoize failed rebuilds per stored prompt (_static_rebuild_failed_for):
the redecoration chokepoint runs at the top of every retry attempt, and a
persistent stable-tier mismatch (restored session whose SOUL.md/skills
changed since save) would otherwise re-run the full prompt build — SOUL.md,
context files, memory I/O — on every attempt of every API call for the
life of the session. A legitimately changed stored prompt retries once.
strip_anthropic_cache_control flattened ANY pure-text multi-part content
list with a separator-less join. Decoration only ever produces a single
text part or the 2-part [static, volatile] system split; organic
multi-part text (merged user turns, imported transcripts) got word-jammed
and parts carrying extra keys (citations) were silently dropped — on the
common no-failover path, since redecoration runs on every attempt.
Restrict the flatten to the exact decoration-produced shapes and make
marker removal copy-on-write on part dicts (the per-call message copy is
shallow, so parts alias the persistent history).
Add strip_anthropic_cache_control coverage and policy-change cases
(cache-off→on, on→off, native→envelope, MoA guidance outside marker)
that TestSyncFailoverPreservesCacheDecoration did not exercise.
Three code-reuse fixes applied during salvage:
1. Reuse _relative_time from hermes_cli/main.py instead of duplicating
the relative-time formatting logic in hermes_cli/status.py.
2. Extract _stamp_hygiene_compression_provenance helper in gateway/run.py
to deduplicate the two nearly-identical try/except blocks that stamp
compression timeout/abort provenance in the hygiene path.
3. Add ContextCompressor.record_timeout_failure() method and use it from
the in-agent compress_context timeout callback instead of re-implementing
the (60, 300, 900) cooldown ladder inline. The existing summary-LLM
exception handler already has this ladder — now both paths share one
method.
Three mechanisms to detect and notify when gateway sessions stall silently:
1. Mid-turn activity heartbeats stamped to SessionDB so hermes sessions list
and hermes status show progress during long turns without new message rows.
2. Stall watchdog: when a busy session has pending inbound and the shared
activity clock is idle past agent.session_stall_timeout (default 300),
log a WARNING and notify the user once to try /new. Notify-only; does
not kill the turn.
3. Compaction timeout: fenceless compress_context callers get a progress-aware
host budget (compression.context_timeout_seconds default 120 idle,
compression.context_total_ceiling_seconds default 600 ceiling). On timeout,
cancel via commit fence, skip compaction without dropping messages, and
continue the turn.
Closes#72016 (slices 1-3; slice 4 cumulative SSE stream-retry deadline
remains a follow-up).
Cherry-picked from PR #72424 by @fangliquanflq.
PR #72425 added getattr(agent, '_incremental_persistence_failed', False)
checks at the top of execute_tool_calls_{sequential,concurrent,segmented}.
A bare MagicMock auto-creates a truthy value for any attribute access,
so the interrupt-skip test's MagicMock agent short-circuited before
appending cancelled-tool messages — assert len(messages)==3 got 0.
Production is unaffected: run_conversation resets the flag to False
explicitly at turn start (conversation_loop.py:~1028).
- codex app-server sibling path: surface a WARNING (was silent debug) when
the projected-message flush fails — same bug class as the main fix, but
codex output has already streamed so fail-closed and agent_persisted=False
are both wrong here (#860/#42039 duplicate-write hazard); loud durability
gap logging instead.
- map session_persistence_failed in _format_turn_completion_explanation so
the user sees an actionable reason instead of 'The request failed:
unknown error' + explainer test.
- contributors/emails mapping for elco@thedaoist.gg (attribution CI).
The change-detector test asserted 'hasattr(signal, "SIGPIPE")' in source.
PR #72677 replaced inline hasattr+signal.signal with _install_signal()
which does the same guard via getattr internally. Update the test to
accept either form.
Importing tui_gateway.entry from a worker thread raised
'ValueError: signal only works in main thread of the main interpreter'
because signal.signal() was called unconditionally at import time.
On the Desktop/WebSocket path, server._build() runs in a daemon thread and
does 'from tui_gateway.entry import ensure_mcp_discovery_started' as the
first import of entry (entry.main() is never run there), which crashed and
aborted MCP discovery startup — every session then lost its MCP servers
(e.g. Dart-mcp) with ClosedResourceError.
signal handlers are process-global, so installing them only when the module
is first imported in the main thread is sufficient; importing from a worker
thread becomes a safe no-op.
Fixes#72667
The stale-.pyc bug class (#6207, #60242, live WhatsApp report: gateway
ImportError 'cannot import name parse_model_flags_detailed' after /update)
has one shared shape: the checkout's .py files change while __pycache__
retains bytecode from the previous revision, and a later process trusts
the stale .pyc.
Update-time clears can never fully close this class: 'hermes update'
always executes the PRE-pull updater code, so hardening added to it takes
effect one update late — and a manual 'git pull' never runs the updater
at all.
Class fix:
- Launch-time guard in main(): compare the checkout fingerprint (cheap
file reads via _read_git_revision_fingerprint, no git subprocess)
against a .bytecode-fingerprint stamp; sweep __pycache__ once when they
diverge. Covers manual pulls, old updaters, ZIP restores — every entry
point (CLI, gateway service, desktop backend) passes through main().
- Record the stamp at all three update-time clear sites (git path pre-
install, git path post-install, ZIP path) so a normal update never
triggers a redundant launch sweep.
- Sibling site: 'hermes plugins update' + dashboard plugin update now
clear __pycache__ under the plugin dir after git pull (plugin trees
live outside the repo guard).
E2E validated: reproduced the stale-pyc shadowing (same-size same-mtime
source swap → old symbol wins in a fresh process), confirmed the launch
sweep restores the new symbol, no-ops when the checkout is unchanged,
and re-sweeps on the next pull. Sabotage-tested: reverting the sweep
fails 4 of the new tests.
Refresh update-sensitive modules before lazy backend refresh so an in-place git update does not keep using pre-pull module objects when newly pulled code imports fresh helpers.
Constraint: Issue #60242 reports post-update lazy backend refresh importing stale hermes_constants after a large Windows update.
Rejected: Only clearing __pycache__ earlier | the update process can still hold old modules in sys.modules.
Confidence: high
Scope-risk: narrow
Directive: Keep update-time lazy refresh guarded against in-process code skew after git pull.
Tested: ./.venv/bin/python -m pytest tests/hermes_cli/test_update_autostash.py::test_cmd_update_reloads_runtime_modules_before_lazy_refresh tests/hermes_cli/test_update_autostash.py::test_reload_updated_runtime_modules_restores_new_hermes_constants_symbol -q
Tested: ./.venv/bin/python -m pytest tests/hermes_cli/test_update_autostash.py -q
Tested: ./.venv/bin/python -m ruff check hermes_cli/main.py tests/hermes_cli/test_update_autostash.py
Tested: ./.venv/bin/python -m py_compile hermes_cli/main.py tests/hermes_cli/test_update_autostash.py
Tested: git diff --check
Not-tested: Windows v0.14.0 to v0.18.0 end-to-end update.
claude-fable-5 is a Mythos-class reasoning model (1M context, 128K output,
adaptive thinking per anthropic_adapter.py) but was missing from the
_REASONING_STALE_TIMEOUT_FLOORS table. Without a floor entry it got the
default 180s stale timeout (300s with context scaling), which is too short
for fable-5's thinking phase on large contexts.
Each stale kill bumped the cross-turn circuit breaker streak; after 5
consecutive kills _check_stale_giveup() fired immediately (elapsed: 0.00s),
aborting all calls with "Provider has been unresponsive for 5 consecutive
stale attempts." Users with 191K-token contexts hit this reliably.
Add ("claude-fable", 600) — deep-reasoning tier alongside o1/deepseek-r1/
nemotron-3-ultra. The claude-fable slug matches claude-fable-5 and future
variants via the existing right-anchor regex.
Reshapes the salvaged fix so it recovers the `tsc: not found` build without
mis-diagnosing healthy trees.
The npm config forcing is dropped. It rested on the premise that an inherited
`npm_config_omit=dev` can beat the `--include=dev` CLI flag; npm resolves
command-line flags above environment config and filters `omit` by `include`,
so the flag already wins. Verified on npm 10 and 11.5.1: with
`npm_config_omit=dev` set and `--include=dev` passed, devDependencies install.
`npm_config_production` is worse than redundant — npm 9 removed it, so setting
it prints `npm warn config production Use --omit=dev instead.` on every
install.
The readiness probe now reads every root `npm run build` searches, not just
the workspace root. npm links a package's bin shims under the package itself
when it owns its lockfile (#42973), so a root-only check called a working tree
broken: it forced a redundant install, skipped the build entirely, and made
`hermes web` exit 1 on a layout that builds fine today.
The pre-build probe is gone with it. A build that works is never second-guessed
and no filesystem introspection gates it; recovery is driven by the failure
instead. When the build cannot resolve tsc or vite, we reinstall (visibly) and
retry before the generic delayed retry, which otherwise just reruns the same
command and leaves the stale dist in place forever. The lockfile-hash skip
still invalidates on an incomplete toolchain so the next update repairs itself.
Also drops the branches that changed behavior based on whether a test mock was
installed, and replaces the mock-call-count tests with real temp trees covering
both hoisting layouts, the Windows shim extensions, and each shell's wording of
an unresolvable binary.
Co-authored-by: Gerardo Camorlinga Jr. <gercamjr.dev@gmail.com>
hermes update could report a successful workspace npm ci while
devDependencies were omitted (production/omit-dev env leakage), leaving
tsc/vite unlinked. The subsequent web UI build then failed with
`tsc: not found` and fell back to a stale dashboard dist.
Force npm_config_include=dev on install, verify toolchain shims after
workspace install (and refuse to hash-cache a half tree), repair/reinstall
when node_modules exists without tsc/vite, prepend workspace .bin dirs to
PATH, and retry the build after a tsc/vite-not-found failure.
* fix: Branch in new chat loses the question and the branched session (#issue)
- session.branch on the backend now accepts a count param to truncate
the parent history to the clicked message, instead of always forking
the entire transcript. Also returns stored_session_id/messages/info
so the frontend has parity with session.create's response shape.
- branchCurrentSession (open live chat) now slices history from 0
instead of from the clicked message index, so the question preceding
an assistant reply is no longer dropped when branching.
- forkBranch now calls session.branch (not session.create) when
branching an open live chat, since session.create only persists a DB
row lazily on first prompt - a branched chat that nobody types into
never got saved, and vanished as 'session not found' on the next
app restart. branchStoredSession (branching from the sidebar, no
live runtime) keeps using session.create as before.
* test: cover session.branch count truncation and open-chat branching
- backend: assert session.branch with a count param only persists the
first N messages of the live history to the new session.
- frontend: BranchHarness now exposes branchCurrentSession; assert
branching an open chat from a middle message calls session.branch
with the parent session id and the correct trimmed count, instead of
session.create.
Update _pre_msg_count after adopting the durable transcript so the
post-compression log reflects the correct pre-adoption message count.
Also use the existing _live_child_id() helper in the updated test
instead of hand-rolled child-id extraction.
Follow-up to #72631.
Busy sessions (memory review / shared session writers) kept outrunning the
in-memory snapshot, so rotation-mode compress aborted every attempt with
"changed before lease acquisition" and surfaced as a fake "No changes from
compression". Adopt the durable transcript and continue compressing instead
of returning the stale snapshot unchanged.
Per hermes-sweeper review on #56671: the fallback exclusion matched any
canonical string starting with "custom" (e.g. "customproxy"), not just
the durable named-custom-provider syntax ("custom" bucket or
"custom:<name>" slugs). Narrow it to an exact/prefix match on that
syntax so unrelated vendor names aren't accidentally exempted from the
openrouter fallback.
Also clarifies the test suite: a properly configured custom:<name>
provider now resolves via resolve_custom_provider before this fallback
is ever reached (added upstream in 9a15fad0d6), so the existing test
was mislabeled as exercising that primary path when it was actually
exercising the fallback-safety-net case (missing/unresolved config
entry). Split into explicit primary-path and fallback-safety-net tests,
plus a regression test for the "customproxy"-style false positive.
_normalize_main_model_assignment() (POST /api/model/set, the endpoint
Desktop's Settings -> Model page uses to persist the main model slot) has
a fallback for a specific analytics bug: an older session row with no
billing_provider sends the model's bare vendor prefix as "provider"
(e.g. "anthropic" from "anthropic/claude-opus-4.6"), so the code detects
an unrecognized provider paired with a slash-bearing model and treats it
as that stray-vendor-prefix case.
Named custom providers are represented as "custom:<name>" slugs
everywhere else in the codebase (runtime_provider.py, model_switch.py),
but _KNOWN_PROVIDER_NAMES only lists the bare "custom" bucket. So picking
a named custom provider (e.g. "custom:litellm", a LiteLLM proxy fronting
Ollama) together with a slash-bearing model ("ollama/glm-5.2") looked
identical to the stray-vendor-prefix case and got silently rewritten to
provider: openrouter in config.yaml on save -- reassigning the provider
entirely, not just mangling the model id.
Exclude anything starting with "custom" from the fallback, matching the
guard the same function already applies later for the actual
normalize_model_for_provider call.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
_MATCHING_PREFIX_STRIP_PROVIDERS includes "custom" so that manually typed
config values like "zai/glm-5.1" repair themselves for their matching
native provider. But "custom" is a generic bucket for arbitrary
user-defined endpoints, not a vendor identity -- unlike zai/gemini/xai,
where a matching alias really does mean "this prefix names the same
backend as the target provider."
_PROVIDER_ALIASES maps "ollama" -> "custom", so a model configured as
"ollama/glm-5.2" against a named custom provider (e.g. a LiteLLM proxy
fronting Ollama, which registers its routes as "ollama/<model>") had its
prefix stripped to bare "glm-5.2" -- a name the proxy doesn't recognize.
_strip_matching_provider_prefix now only strips a literal "custom/"
prefix when the target resolves to "custom"; an alias that merely
resolves to custom (ollama) no longer qualifies, since custom has no
vendor identity for it to redundantly repeat.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Root-context rm -f "$log_dir/lock" could follow a raced directory
symlink and unlink a foreign lock outside HERMES_HOME. Clear the
stale lock via s6-setuidgid hermes alongside mkdir, and assert
victim/lock survives the swap-race test.
Without restartable log/run chown, warm volumes that keep a hermes-owned
HERMES_HOME but root-owned logs/gateways would again deny hermes mkdir.
Add a non-recursive stage2 parent heal and cover the poisoned-parent reboot path.
Root-context log/run used to pathname-chown hermes-writable log paths,
which a hermes user can race through a symlink swap via the writable
log control FIFO. Create the leaf with s6-setuidgid hermes mkdir instead;
parent logs/gateways ownership stays a stage2 boot concern (#45258).
Every fallback/dedup/skip decision asks one question — 'is this candidate
the same backend as the one that failed, along the axis that failure
invalidated?' — but it was re-implemented inline at six sites across four
subsystems, each comparing whatever string was locally convenient. Each
incident fixed one site while the others kept the bug: #22548, #70893,
#59561, #72468, #62984/#54250/#57584.
agent/backend_identity.py now owns the concept: BackendIdentity (provider /
model / base_url axes), FailureScope (MODEL / CREDENTIAL / ENDPOINT — each
failure class invalidates a different axis), and should_skip_candidate().
Unknown axes never manufacture a skip (over-skipping strands failover; a
wrong try costs one RTT).
Migrated sites:
- chat_completion_helpers.try_activate_fallback: replaces the provider+model
early-exit (the #62984 bug: ignored base_url, stranding multi-endpoint
pools) AND _fallback_entry_is_same_backend_by_base_url (deleted)
- auxiliary_client._try_configured_fallback_chain +
_try_main_agent_model_fallback: replace label/model comparisons; auth and
payment map to CREDENTIAL scope, keeping the #59561 carve-out
- hermes_cli/fallback_cmd add: primary-match + duplicate checks now identity-
aware (#54250/#57584): same provider+model on a different explicit
base_url is a pool entry, not a duplicate
_mark_provider_unhealthy stays label-keyed deliberately: its only triggers
are confirmed 402s, which ARE credential-scoped.
Owner-level tests pin each incident's semantics by number; sabotage-verified
(removing the base_url axis fails the #62984 test).
Follow-ups on the salvaged #63359:
- _finalize_child_results carries tool_call_history on subagent_stop
(the #62011/#72403 field landed after the PR branched; the shared
pipeline must emit it for both delegate_task and plugin-launched
children). Lifecycle test updated for the new payload field.
- The lifecycle executor uses DaemonThreadPoolExecutor — a wedged or
abandoned child must never block interpreter exit at atexit-join time
(same rationale as _run_single_child's timeout executor and the
async-delegation pool).
- delegate_task's batch path keeps live-transcript wiring while routing
child construction through the shared
_build_child_preserving_parent_tools helper.
`hermes -c`/`--resume` (continue last session) resolved the globally
most-recently-used session, then cd'd into *its* recorded cwd. So running
`hermes -c` from repo A could land you in repo B's session — the session
you last touched anywhere, not the last one *here*.
Now `_resolve_last_session` scopes to the current workspace first: the git
repo root when CWD is inside a repo (so all sessions across its
subdirs/worktrees group together), else the CWD itself — matching the
`workspace_key` identity `hermes sessions list --workspace` already groups
on. It falls back to the unscoped global MRU when no session matches the
current workspace, preserving the old behaviour for fresh directories.
Adds `workspace_key` param to `SessionDB.search_sessions` and a
`_workspace_key_clause` SQL helper that mirrors `workspace_key()`: a row
matches when its `git_repo_root` equals the key, or (legacy rows without
git metadata) when its `cwd` is at or under it.
Widen okalentiev's failed_model narrowing (#59561) to
_try_main_agent_model_fallback. The safety-net layer still skipped on a
provider-label match alone, so single-provider users whose aux compression
model and main model share one custom endpoint had ZERO fallbacks: the aux
model timing out exhausted the chain in one hop and compression aborted.
Real incident (0.19.0 debug dump): aux zai-org/glm-5.2 hung 324s and timed
out while main mindai/macaron-v1-venti on the SAME endpoint was serving
448K-token turns — the label-only skip discarded the one viable summarizer,
the session wedged over threshold, and the anti-thrash breaker tripped.
Same convention as the chain fix: model-specific failures (timeout,
connection, rate limit) pass failed_model so only the exact failed model is
skipped; provider-wide failures (auth 401 / payment 402) pass None and keep
the whole-provider skip. Both sync and async call_llm sites pass it.
Sabotage-verified: the new regression test fails on the provider-only skip.
Follow-up to the same-provider fallback fix: narrowing the configured-chain
skip to the exact failed model is only correct for model-specific failures.
Auth (401) and payment (402) errors are provider-wide — every model on the
provider shares the same broken credentials/account — so trying a sibling
model can't recover and merely burns another doomed request before the
aux task fails. Worse, returning that sibling client bypasses the
main-agent-model safety net that a provider-wide skip would have reached.
Only forward failed_model to _try_configured_fallback_chain for
model-specific failures (timeout, connection, rate limit, model-incompatible,
invalid response). Auth/payment keep failed_model=None (whole-provider skip),
preserving the pre-existing safety-net behaviour for credential/billing
failures.
Adds an integration test that a timeout forwards the failed model, and
updates the payment-error test to assert failed_model=None.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
_try_configured_fallback_chain skipped every fallback_chain entry whose
provider matched the one that just failed. A chain that intentionally lists
several models under the same provider (e.g. two more NVIDIA NIM models
after the primary NIM model times out) was therefore skipped wholesale,
falling straight through to the main-agent-model safety net instead of
trying the other configured models on that provider.
Add failed_model so the skip narrows to the exact (provider, model) pair
that failed. Callers that only know the provider (client-build failures,
where the whole provider is unreachable regardless of model) keep the old
provider-wide skip; the two runtime request-error call sites (call_llm,
async_call_llm) now pass the model that just failed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Per Teknium: the caps should bound a single agent loop, not accumulate
over the whole session. Rename SessionCapConfig -> LoopCapConfig and the
config section session_caps -> loop_caps; move the counters into
reset_for_turn (invoked per turn via turn_context) so each turn starts
with a fresh budget; retune defaults 200 -> 50 (a single turn issuing 50
web searches / spawning 50 subagents is already pathological). Block
codes session_*_cap -> loop_*_cap and messages updated to drop the
/new-resets-the-budget guidance (irrelevant now that it resets per turn).
Tests flipped: the old persists-across-turn-resets assertion becomes
resets-each-turn.
Add per-session lifetime caps on web_search calls and subagent spawns
(defaults 200/200, matching Claude Code v2.1.212). Unlike the existing
per-turn tool-loop guardrails, these count over the whole session and
reset only when a fresh agent is built (/new, /clear). Hitting a cap
blocks the offending call and halts the turn cleanly.
- agent/tool_guardrails.py: SessionCapConfig + session counters on the
controller (in __init__, not reset_for_turn, so they persist across
turns). before_call() enforces caps first, independent of
hard_stop_enabled. delegate_task batches count each task.
- hermes_cli/config.py: tool_loop_guardrails.session_caps defaults.
- docs + tests (unit + E2E validated against a real AIAgent).
One relay adapter fronts N platforms on one WS, but the capability surface
(MAX_MESSAGE_LENGTH / message_len_fn) was a scalar from whichever descriptor
resolved the handshake — and the transport's read loop OVERWROTE it on every
descriptor frame (last-writer-wins). A Discord chat on a gateway whose
applied descriptor was Telegram's inherited the 4,096-char cap and over-sent
into Discord's 2,000-char API 400 (observed live: 2,543/2,641-char replies
silently lost while inbound kept working).
- ws_transport: accumulate one descriptor per platform in
_descriptors_by_platform (exposed via descriptor_for_platform); the FIRST
descriptor of a connection generation stays the session default instead of
last-writer-wins; the map resets on re-dial.
- BasePlatformAdapter: new max_message_length_for_chat /
message_len_fn_for_chat hooks defaulting to the scalar surface (native
single-platform adapters unchanged).
- RelayAdapter: overrides resolve the chat's platform from _platform_by_chat
(the same map per-frame egress uses) and look up that platform's negotiated
descriptor; falls back to the scalar for unknown chats/transports.
- stream_consumer (streaming budget, _raw_message_limit, fallback-continuation
chunking) + run.py tool-progress limit now resolve per-chat.
Tests: tests/gateway/relay/test_relay_per_platform_caps.py (7) — verified
fail-without/pass-with (all 7 fail with the fix stashed). Relay + stream
consumer suites green (213 + 223).
