mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-05-29 06:31:32 +00:00
fix(anthropic): reactive recovery for OAuth 1M-context beta rejection (#17752)
Keep context-1m-2025-08-07 in OAuth requests by default so 1M-capable subscriptions retain full context. When Anthropic rejects a request with 400 'long context beta is not yet available for this subscription', disable the beta for the rest of the session, rebuild the client, and retry once. Addresses #17680 (thanks @JayGwod for the clean reproduction) without forcing every OAuth user off the 1M context window. Changes: - agent/error_classifier.py: new FailoverReason.oauth_long_context_beta_forbidden; pattern matches 400 + 'long context beta' + 'not yet available'. Narrow enough that the existing 429 tier-gate pattern keeps its own reason. - agent/anthropic_adapter.py: _common_betas_for_base_url, build_anthropic_client, build_anthropic_kwargs gain drop_context_1m_beta kwarg. Default=False (1M stays). OAuth OAUTH_ONLY_BETAS unchanged. - agent/transports/anthropic.py: build_kwargs forwards the flag. - run_agent.py: self._oauth_1m_beta_disabled flag, retry-once guard, recovery branch next to the image-shrink path. _rebuild_anthropic_client honors the flag. The main build_kwargs call site threads it through for fast-mode extra_headers. - hermes_cli/doctor.py, hermes_cli/models.py: sibling OAuth /v1/models probes get the same reactive retry — previously they'd falsely report the Anthropic API as unreachable for affected subscriptions. Tests: 2190 tests/agent/ + 94 adjacent integration tests pass. New unit tests cover the classifier pattern (including the collision guard against the 429 tier-gate) and the drop_context_1m_beta adapter behavior (default keeps 1M, flag strips only 1M while preserving every other beta).
This commit is contained in:
parent
4d363499db
commit
828d3a320b
8 changed files with 264 additions and 23 deletions
|
|
@ -54,6 +54,7 @@ class FailoverReason(enum.Enum):
|
|||
# Provider-specific
|
||||
thinking_signature = "thinking_signature" # Anthropic thinking block sig invalid
|
||||
long_context_tier = "long_context_tier" # Anthropic "extra usage" tier gate
|
||||
oauth_long_context_beta_forbidden = "oauth_long_context_beta_forbidden" # Anthropic OAuth subscription rejects 1M context beta — disable beta and retry
|
||||
|
||||
# Catch-all
|
||||
unknown = "unknown" # Unclassifiable — retry with backoff
|
||||
|
|
@ -450,6 +451,25 @@ def classify_api_error(
|
|||
should_compress=True,
|
||||
)
|
||||
|
||||
# Anthropic OAuth subscription rejects the 1M-context beta header.
|
||||
# Observed error body: "The long context beta is not yet available for
|
||||
# this subscription." Returned as HTTP 400 from native Anthropic when
|
||||
# the subscription doesn't include 1M context, even though the request
|
||||
# carries ``anthropic-beta: context-1m-2025-08-07``. The recovery path
|
||||
# in run_agent.py rebuilds the Anthropic client with the beta stripped
|
||||
# and retries once. Pattern is narrow enough that it won't collide with
|
||||
# the 429 tier-gate pattern above (different status, different phrase).
|
||||
if (
|
||||
status_code == 400
|
||||
and "long context beta" in error_msg
|
||||
and "not yet available" in error_msg
|
||||
):
|
||||
return _result(
|
||||
FailoverReason.oauth_long_context_beta_forbidden,
|
||||
retryable=True,
|
||||
should_compress=False,
|
||||
)
|
||||
|
||||
# ── 2. HTTP status code classification ──────────────────────────
|
||||
|
||||
if status_code is not None:
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue