Files
hermes-agent/tests/tools/test_bot_failure_reasons.py
kshitijk4poor a802212171 fix(bot-relay): classify the named-secret spelling of a scope refusal too
UnscopedSecretError reads "could not read this profile's OPENROUTER_API_KEY" when get_secret
passes the name, which the salvaged rule missed (fell to unknown). Key on the lede both
spellings share. The dedicated test duplicated the parametrized table (and retry_action of
a non-retryable code); the named spelling is a table row built from the real exception.
2026-10-01 16:18:46 +05:30

103 lines
4.7 KiB
Python

"""Tests: typed failure-reason codes (tools/bot_failure_reasons.py, #93091).
Pins the closed reason vocabulary, the ordered classifier (incl. the
auth-beats-quota precedence seen in real Anthropic 401 bodies), the three
real-world fixtures from live bot runs, and the auto-retryable set.
"""
import pytest
from agent.secret_scope import UnscopedSecretError
from tools import bot_failure_reasons as fr
# Real error text captured from live bot turns.
FIXTURE_ANTHROPIC_401 = (
"Error code: 401 - {'type': 'error', 'error': {'type': 'authentication_error', "
"'message': 'Your API key is invalid, blocked or out of funds...'}}"
)
FIXTURE_NO_PROVIDER = (
"agent init failed: No LLM provider configured. Run `hermes model` to select "
"a provider, or run `hermes setup` for first-time configuration."
)
FIXTURE_NO_TOKEN = "agent init failed: No access token found for Nous Portal login."
# Target-scope spawn refusals, verbatim from a relay ledger (the named-secret spelling is built live).
FIXTURE_TARGET_SCOPE = (
"Hermes could not read this profile's API key (an internal profile-scoping bug on the "
"multiplexed gateway, not your configuration). Run `hermes gateway restart`; if it keeps "
"happening, report it with `hermes debug share`."
)
FIXTURE_TARGET_SCOPE_DETAILED = (
"served_profile_child_env(inherit_credentials=True) called with no target home and no profile "
"secret scope bound while multiplexing is on; the child would inherit the launch profile's "
"credentials. Bind the profile scope (or pass target_home) at the spawn site."
)
@pytest.mark.parametrize(
("text", "code"),
[
("Error code: 403 - forbidden", fr.PROVIDER_AUTH_OR_ACCESS),
("Invalid API key provided", fr.PROVIDER_AUTH_OR_ACCESS),
("Error code: 402 - payment required", fr.PROVIDER_QUOTA_LIMIT),
("insufficient balance, top up your account", fr.PROVIDER_QUOTA_LIMIT),
("You exceeded your current quota", fr.PROVIDER_QUOTA_LIMIT),
("Error code: 429 - Too Many Requests", fr.PROVIDER_RATE_LIMIT),
("Rate limit reached for gpt-4o", fr.PROVIDER_RATE_LIMIT),
("Error code: 500 - internal server error", fr.PROVIDER_SERVER_ERROR),
("Error code: 529 - overloaded_error: Overloaded", fr.PROVIDER_SERVER_ERROR),
("This model's maximum context length is 128000 tokens", fr.CONTEXT_OVERFLOW),
("context_overflow: prompt too large", fr.CONTEXT_OVERFLOW),
("missing config: no provider block in config.yaml", fr.MISSING_CONFIG),
("model 'gpt-9' not found", fr.MODEL_UNAVAILABLE),
("The model `foo-bar` does not exist", fr.MODEL_UNAVAILABLE),
("model_not_found", fr.MODEL_UNAVAILABLE),
("status: 401 unauthorized", fr.PROVIDER_AUTH_OR_ACCESS),
("upstream server error", fr.PROVIDER_SERVER_ERROR),
(FIXTURE_TARGET_SCOPE, fr.TARGET_SCOPE_UNRESOLVED),
(FIXTURE_TARGET_SCOPE_DETAILED, fr.TARGET_SCOPE_UNRESOLVED),
(str(UnscopedSecretError("OPENROUTER_API_KEY")), fr.TARGET_SCOPE_UNRESOLVED),
# bare numbers WITHOUT a status-code context must not classify —
# they feed AUTO_RETRYABLE and a misfire could auto-retry a
# permanent local failure (review finding on #93101).
("gate check failed: error at line 502 of module", fr.UNKNOWN),
("took 429 ms to fail", fr.UNKNOWN),
("something inexplicable happened", fr.UNKNOWN),
("", fr.UNKNOWN),
(None, fr.UNKNOWN),
],
)
def test_classify_agent_error_rules(text, code):
assert fr.classify_agent_error(text) == code
def test_fixture_anthropic_401_auth_beats_quota():
# The live 401 body ALSO says "out of funds" — auth wins by precedence.
assert "out of funds" in FIXTURE_ANTHROPIC_401
assert fr.classify_agent_error(FIXTURE_ANTHROPIC_401) == fr.PROVIDER_AUTH_OR_ACCESS
def test_precedence_authentication_error_type_alone_beats_quota_words():
text = "authentication_error: account is out of funds"
assert fr.classify_agent_error(text) == fr.PROVIDER_AUTH_OR_ACCESS
# but plain quota text without any auth marker classifies as quota
assert fr.classify_agent_error("account is out of funds") == fr.PROVIDER_QUOTA_LIMIT
def test_fixture_no_provider_configured_is_missing_config():
assert fr.classify_agent_error(FIXTURE_NO_PROVIDER) == fr.MISSING_CONFIG
def test_fixture_no_access_token_is_missing_config():
assert fr.classify_agent_error(FIXTURE_NO_TOKEN) == fr.MISSING_CONFIG
def test_auto_retryable_set_and_predicate():
for code in fr.AUTO_RETRYABLE:
assert fr.is_auto_retryable(code)
for code in fr.ALL_REASONS - fr.AUTO_RETRYABLE:
assert not fr.is_auto_retryable(code)
assert not fr.is_auto_retryable("")
assert not fr.is_auto_retryable("nonsense")