* fix(codex-responses): gracefully recover from invalid_encrypted_content (salvage #10144)
When an OpenAI-compatible Responses API surface accepts an initial
request but later rejects the replayed `codex_reasoning_items`
encrypted blob with HTTP 400 `invalid_encrypted_content`, the
session previously got stuck retrying the same poisoned payload.
Recovery: classify the error as a dedicated FailoverReason, and on the
first hit disable encrypted reasoning replay for the rest of the
session, strip cached items from message history, and retry once.
Changes:
* error_classifier: add FailoverReason.invalid_encrypted_content
branch in _classify_400 (before context_overflow so the messages
that mention 'encrypted content … could not be verified' don't trip
context heuristics), in _classify_by_error_code, and extend
_extract_error_code to peek inside wrapped JSON in error.message and
ignore the bare '400' as a code.
* agent_init: initialize `_codex_reasoning_replay_enabled = True` on
every agent.
* run_agent: add AIAgent._disable_codex_reasoning_replay() helper
that flips the flag and pops cached items.
* codex_responses_adapter: thread a `replay_encrypted_reasoning`
kwarg through _chat_messages_to_responses_input so that when the
flag is False we don't replay codex_reasoning_items.
* transports/codex.py: read `replay_encrypted_reasoning` from params,
thread it into the adapter, and gate the
`include=['reasoning.encrypted_content']` request hint on it.
* chat_completion_helpers: pass the agent's replay flag through to
the transport.
* conversation_loop: in the retry loop, add an
invalid_encrypted_content recovery branch that fires once per
session, only when api_mode == codex_responses, only when replay is
still enabled, and only when at least one assistant message in
history actually carries cached reasoning items (otherwise the 400
has nothing to do with our cache and the normal retry path handles
it).
Tests:
* test_error_classifier: new wrapped-JSON _extract_error_code case;
new TestClassifyApiError cases proving the 400 is retryable with
no fallback, that the broad message match doesn't catch a generic
'parsed' message, and that the error code match is
case-insensitive.
* test_run_agent_codex_responses: end-to-end test of the recovery
branch firing once and disabling replay, plus a sibling test that
proves the branch does *not* fire (and the flag stays True) when
history has no cached reasoning items.
Salvages PR #10144 onto the post-refactor module layout
(error_classifier / codex_responses_adapter / transports/codex /
conversation_loop / agent_init) since the original diff was written
against the pre-refactor monolithic run_agent.py.
* chore(release): map victorGPT in AUTHOR_MAP for #10144 salvage
---------
Co-authored-by: victorGPT <wuxuebin1993@gmail.com>
* fix(minimax-oauth): refresh short-lived access tokens per request
MiniMax OAuth issues ~15-minute access tokens. The Anthropic SDK caches
api_key as a static string at client construction, so a session that
resolves credentials once at startup keeps sending the same bearer until
MiniMax returns 401 mid-session.
Swap the static string for a callable token provider, reusing the existing
Entra-ID bearer-hook infrastructure in build_anthropic_client. The callable
re-reads auth.json on each invocation and calls _refresh_minimax_oauth_state,
which is a no-op when the token still has more than 60s of life left and
refreshes proactively otherwise. Refreshes persist to auth.json so other
processes (gateway, cron) see them immediately.
The wire-up lives at the agent-init / model-switch boundary rather than in
resolve_runtime_provider, so aux client paths that hand the api_key string
to OpenAI(api_key=...) are unaffected.
* docs: add infographic for minimax-oauth token refresh
The memory-provider gate added in the prior commit closes one of two
blind-injection sites in agent_init.py. The context engine block (lines
~1445) follows the identical pattern: agent.context_compressor.get_tool_schemas()
(lcm_grep, lcm_describe, lcm_expand) was appended to agent.tools unconditionally,
ignoring enabled_toolsets.
Same bug class, same local-model latency penalty, same one-line gate — using
'context_engine' as the toolset name (matches the existing plugin-system
convention in plugins.py, plugins_cmd.py, etc.).
Also adds Lempkey to scripts/release.py AUTHOR_MAP for the prior commit's
authorship.
MemoryManager.get_all_tool_schemas() output was appended to AIAgent.tools
unconditionally — bypassing the enabled_toolsets / platform_toolsets filter.
Setting `platform_toolsets: telegram: []` had no effect: fact_store and other
memory provider tools still leaked into the tool surface on every session.
Impact on local models (per @thundercat49's benchmarks on Qwen3-30B-A3B Q4_K_M /
RTX 3090): tool-formatted prompts process at 134 tok/s vs 1,230 tok/s for plain
text. With 8 memory tool schemas injected, a simple 'hello' on Telegram took
~42s instead of ~1.7s. Small models also entered tool-call loops when memory
tools were the only tools present.
Gate condition (matches the natural meaning of enabled_toolsets):
None → no filter, inject (backward compat)
contains 'memory' → user opted in, inject
otherwise (including []) → skip injection
Co-authored-by: Teknium <127238744+teknium1@users.noreply.github.com>
Allow custom OpenAI-compatible providers declared under `custom_providers:`
to set provider-specific `extra_body` fields and have Hermes merge them into
chat-completions requests when the matching custom endpoint is active.
This is a manual per-provider override rather than a model-name heuristic.
OpenAI-compatible Gemma thinking support is real, but the on-wire payload
shape is backend-specific: some servers want top-level `enable_thinking`,
while vLLM Gemma and NIM-style endpoints expect `chat_template_kwargs`.
A per-provider override is safer than picking one assumed payload.
Example config:
```yaml
custom_providers:
- name: gemma-local
base_url: http://localhost:8080/v1
model: google/gemma-4-31b-it
extra_body:
enable_thinking: true
reasoning_effort: high
```
For vLLM Gemma or NIM-style endpoints, use the nested shape those servers
expect:
```yaml
extra_body:
chat_template_kwargs:
enable_thinking: true
```
Changes:
- `hermes_cli/config.py`: preserve `extra_body` in normalized
`custom_providers:` entries and allow it in the validated field set.
- `hermes_cli/runtime_provider.py`: propagate custom-provider `extra_body`
as `request_overrides.extra_body` for named custom runtime resolution,
including credential-pool paths.
- `agent/agent_init.py`: at agent init, locate the matching custom-provider
entry by `base_url` (+ optional model) and merge its `extra_body` into
`AIAgent.request_overrides`, with caller-provided overrides winning on
conflicting top-level keys.
- `plugins/model-providers/custom/__init__.py`: keep existing CustomProfile
behavior (Ollama `num_ctx`, `think=False` when reasoning disabled);
user-configured `extra_body` flows through `request_overrides`.
- `website/docs/integrations/providers.md`: document the explicit
`extra_body` override and the vLLM/Gemma `chat_template_kwargs` variant.
- Tests cover config normalization, runtime propagation, model matching,
trailing-slash equivalence, fallback when no `model` field is set, and
caller-override merging precedence.
Verified end-to-end against `CustomProfile` via `ChatCompletionsTransport`:
configured `extra_body` reaches `kwargs.extra_body` on the wire request,
and coexists with profile-generated entries (Ollama `num_ctx`, `think=False`)
without clobber.
Salvaged from #29022 onto current `main`. Cosmetic typing edit in
`plugins/model-providers/custom/__init__.py` and a stale-base docs revert
in `providers.md` were dropped during cherry-pick.
Closes#29022
PR #29182 deleted the per-session JSON snapshot writer outright because
state.db is canonical and the snapshots had no in-tree consumer. Some
users have external tooling that reads `~/.hermes/sessions/session_{sid}.json`
directly, so reintroduce the writer behind a config flag that defaults
to off.
- Add `sessions.write_json_snapshots` (default False) to DEFAULT_CONFIG
- Restore `AIAgent._save_session_log` + `_clean_session_content` as
gated methods. When the flag is off the call is a fast no-op; when
on, the writer behaves as before (atomic write, truncation guard
preserved, REASONING_SCRATCHPAD → think tag normalization)
- Re-derive the target path from `agent.session_id` on each call so
`/branch` and `/compress` re-points happen automatically — no need
to restore the explicit re-point bookkeeping at call sites
- Wire the single call site in `_persist_session` (the cleanup-on-exit
hook). Did NOT restore the 7 intra-turn calls the original PR deleted
— those were redundant writes within the same turn that doubled disk
I/O without adding any persistence guarantee `_persist_session` does
not already provide
- Read the flag once at agent init via `load_config()`, cache as
`agent._session_json_enabled`
- Update `TestNoSessionJsonSnapshot` → `TestSessionJsonSnapshotOptIn`
to pin behavior: default off (no file), opt-in true (file written),
no-op method on default agents, logs_dir retained unconditionally
- Update CONTRIBUTING.md and the bundled `hermes-agent` skill to
document the flag and its default
`AIAgent.__init__` was eagerly calling
`_check_compression_model_feasibility()` which probes the auxiliary
provider chain and runs `get_model_context_length()` (potentially
network-bound) to decide whether the configured auxiliary model can
fit a full compression-threshold window. That cost ~440ms cold on
every agent construction.
Most `chat -q` invocations finish in 1-5 seconds and never accumulate
enough context to trip the compression threshold, so the feasibility
check is pure overhead. The result is also only consumed when
compression actually fires (the function adjusts the live threshold
downward if the aux model can't fit; absent that mutation, the gate
in `conversation_loop.py:442` would never fire anyway).
Defer to first `compress_context()` call via
`agent._compression_feasibility_checked` sentinel. Runs at most once
per agent lifetime, just before the first compression pass. The
warning storage (`_compression_warning`) and gateway replay
machinery is unchanged — it still emits to status_callback on the
first turn that actually needs compression.
E2E timing (chat -q 'hi', 3 runs each):
BEFORE AFTER delta
median wall 2.03s 1.86s -8% (-169ms)
min wall 1.92s 1.63s -15% (-293ms)
Real cold-start observation (synthetic 31-turn agent loop): identical
behavior since feasibility check fires once on first compression and
caches. No semantic difference for sessions that DO compress.
UX trade-off: users with broken auxiliary-provider config no longer
see the warning at session start. They see it when compression first
fires — which is exactly when it matters. For users with working
config (the vast majority), the warning never fires anyway, so the
deferral is invisible.
Tests:
- tests/run_agent/test_compression_feasibility.py — 16/16 pass
(the one test that asserted call-at-init was updated to drive the
lazy check explicitly via agent._check_compression_model_feasibility())
- Live tmux session: 2-turn conversation + tool call completes clean,
zero errors in agent.log
Salvages #24402 by @RyanRana. The KANBAN_GUIDANCE block (~835 tokens)
is session-static — the dispatcher decides at spawn time whether the
process is a kanban worker via the kanban_show tool's check_fn (gated
on HERMES_KANBAN_TASK env var). Re-checking 'kanban_show' in
valid_tool_names and re-loading the reference on every system-prompt
rebuild (init + each context compression) is wasted work.
Caches the resolved string on agent._kanban_worker_guidance once in
agent_init and consumes it in system_prompt.build_system_prompt(),
with a getattr fallback for code paths that bypass agent_init.
PR #28102 made the summary-failure abort path the unconditional default,
changing established behavior. Gate it behind config.yaml flag
`compression.abort_on_summary_failure` (default False = historical
fallback-placeholder behavior).
- hermes_cli/config.py: new `compression.abort_on_summary_failure` key,
default False, documented inline.
- agent/agent_init.py: read the flag from compression config and pass to
ContextCompressor.
- agent/context_compressor.py: `__init__` accepts `abort_on_summary_failure`
(default False). `compress()` failure branch gates the abort on the
flag; when False, falls through to the restored legacy fallback path
(static "summary unavailable" placeholder + drop middle window).
- tests: restore original fallback expectations as default; add new
TestAbortOnSummaryFailure class for the opt-in mode.
Gateway/CLI plumbing (force=True on /compress, hygiene/handler abort
detection, locale `gateway.compress.aborted` key) from PR #28102 stays
intact — those paths only fire when `_last_compress_aborted` is True,
which now only happens when the flag is enabled.
Original commit 8d756a421 by austrian_guy targeted __init__ in
pre-refactor run_agent.py. The body now lives in
agent/agent_init.init_agent — re-applied there.
Co-authored-by: austrian_guy <33156212+ether-btc@users.noreply.github.com>
Original commit 13c3d4b4e by kchantharuan touched __init__ and
_apply_client_headers_for_base_url in pre-refactor run_agent.py. Re-applied to:
- __init__: agent/agent_init.py (3 hunks — NVIDIA branch + _custom_headers
fallback in routed-client and fallback-client paths)
- _apply_client_headers_for_base_url: still in run_agent.py (1 hunk)
build_nvidia_nim_headers was already present in agent/auxiliary_client.py
from the prior merge — no additional port needed.
Co-authored-by: kchantharuan <kchantharuan@nvidia.com>
Original commit b62c99797 by Jaaneek targeted six locations in
pre-refactor run_agent.py. Re-applied to the extracted post-PR locations:
- api_mode dispatch → agent/agent_init.py
- is_xai_responses build_api_kwargs → agent/chat_completion_helpers.py
- codex_auth_retry block + 401 hint → agent/conversation_loop.py
- _try_refresh_codex_client_credentials body → run_agent.py (kept)
The non-run_agent.py portions of the commit (auxiliary_client, codex
transport, hermes_cli/auth, tools/xai_http, tests, docs) merged cleanly
from main via the prior merge commit.
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
The largest method left on AIAgent (60+ parameters, the entire startup
sequence — credential resolution, provider auto-detection, context
engine bootstrap, memory store hydration, plugin lifecycle hooks)
moves into agent/agent_init.py.
AIAgent.__init__ is now a thin wrapper that calls
agent.agent_init.init_agent(self, ...) with the original full
parameter list preserved.
Module-level run_agent names referenced in the body (_openrouter_prewarm_done,
_qwen_portal_headers, _routermint_headers, _hermes_home, OpenAI,
get_tool_definitions, check_toolset_requirements) are resolved through
_ra() so test patches on those names keep working. agent_init's logger
warnings are routed via _ra().logger so tests patching run_agent.logger
capture them (TestStringKSuffixContextLengthWarns,
TestCustomProvidersInvalidContextLengthWarns).
Live E2E reconfirmed on three model paths (openai/gpt-5.4,
anthropic/claude-sonnet-4.6, moonshotai/kimi-k2-thinking).
tests/run_agent/ + tests/agent/: 4313 passed (same pre-existing
test_auxiliary_client failure).
run_agent.py: 5944 -> 4564 lines (-1380).
Total reduction since baseline: 16083 -> 4564 (-11519, 72%).
