mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-06-09 08:21:50 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 12

fix(compress): abort instead of dropping messages when summary LLM fails (#28102)

Browse source 
When auxiliary compression's summary generation returns None (aux model
errored, returned non-JSON, timed out, etc.) the compressor previously
still dropped every middle message between compress_start..compress_end
and replaced them with a static 'Summary generation was unavailable'
placeholder. The session kept going but the user silently lost N turns
of context for nothing.

New behavior: on summary failure, compress() aborts entirely — returns
the input messages unchanged and sets _last_compress_aborted=True. The
existing _summary_failure_cooldown_until gate (30-60s) keeps the aux
model from being burned on every turn. Auto-compress callers detect
the no-op (len(after) == len(before)) and stop looping. The chat is
'frozen' at its current size until the next /compress or /new.

Manual /compress (CLI + gateway) now passes force=True which clears
the cooldown so users can retry immediately after an auto-abort. If
the manual retry also fails, the user gets a visible warning telling
them nothing was dropped and how to retry.

- agent/context_compressor.py: compress() gains force= kwarg; failure
  branch sets _last_compress_aborted and returns messages unchanged
  instead of inserting placeholder.
- run_agent.py: _compress_context() detects abort, surfaces warning,
  skips session-rotation entirely, returns messages unchanged.
- cli.py + gateway/run.py: manual /compress paths pass force=True.
- gateway/run.py: hygiene + /compress handlers detect _last_compress_aborted
  and emit the new 'Compression aborted' warning (gateway.compress.aborted)
  instead of the old 'N historical messages were removed' message.
- locales/*.yaml: new gateway.compress.aborted key in all 16 locales.
- tests: updated to assert the abort contract (messages preserved,
  compression_count not incremented, abort flag set, no placeholder
  leaked). New test_force_true_bypasses_failure_cooldown covers the
  manual-retry path.
This commit is contained in:
Teknium 2026-05-18 10:19:40 -07:00 • committed by GitHub
parent 65e0c49b77
commit 1634397ddb
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
24 changed files with 249 additions and 103 deletions
Download patch file Download diff file Expand all files Collapse all files

113
tests/agent/test_context_compressor.py
View file

@ -64,21 +64,31 @@ class TestCompress:
result = compressor.compress(msgs)
assert result == msgs
def test_truncation_fallback_no_client(self, compressor):
# compressor has client=None, so should use truncation fallback
def test_no_client_aborts_compression_with_messages_preserved(self, compressor):
"""compressor has no provider configured, so _generate_summary returns
None → compression aborts entirely. Messages must be returned
unchanged (no placeholder, no drop) and _last_compress_aborted set."""
msgs = [{"role": "system", "content": "System prompt"}] + self._make_messages(10)
result = compressor.compress(msgs)
assert len(result) < len(msgs)
# Should keep system message and last N
assert result[0]["role"] == "system"
assert compressor.compression_count == 1
# Abort path: messages preserved byte-for-byte
assert result == msgs
assert compressor._last_compress_aborted is True
# Compression count NOT incremented on abort — nothing was compressed.
assert compressor.compression_count == 0
def test_compression_increments_count(self, compressor):
msgs = self._make_messages(10)
compressor.compress(msgs)
assert compressor.compression_count == 1
compressor.compress(msgs)
assert compressor.compression_count == 2
mock_resp = MagicMock()
mock_resp.choices = [MagicMock()]
mock_resp.choices[0].message.content = "summary text"
with patch("agent.context_compressor.call_llm", return_value=mock_resp):
compressor.compress(msgs)
assert compressor.compression_count == 1
# Reset cooldown isn't needed (no prior failure) but reset
# iterative-summary state so the next call follows the same
# path as the first.
compressor.compress(msgs)
assert compressor.compression_count == 2
def test_protects_first_and_last(self, compressor):
msgs = self._make_messages(10)
@ -128,7 +138,11 @@ class TestGenerateSummaryNoneContent:
{"role": "user" if i % 2 == 0 else "assistant", "content": f"msg {i}"}
for i in range(10)
]
result = c.compress(msgs)
mock_resp = MagicMock()
mock_resp.choices = [MagicMock()]
mock_resp.choices[0].message.content = "summary text"
with patch("agent.context_compressor.call_llm", return_value=mock_resp):
result = c.compress(msgs)
assert len(result) < len(msgs)
@ -716,11 +730,14 @@ class TestAuxModelFallbackSurfacedToCallers:
class TestSummaryFailureTrackingForGatewayWarning:
"""When summary generation fails, the compressor must record dropped count
+ fallback flag so gateway hygiene & /compress can surface a visible
warning instead of silently dropping context."""
"""When summary generation fails, the compressor must ABORT compression
entirely (return the original messages unchanged) and set the abort flag
so gateway hygiene & /compress can surface a visible warning. Previous
behavior of inserting a static "summary unavailable" placeholder while
silently dropping the middle window has been removed — losing N turns
of context is worse than freezing the chat until the user retries."""
def test_compress_records_fallback_and_dropped_count_on_summary_failure(self):
def test_compress_aborts_and_preserves_messages_on_summary_failure(self):
with patch("agent.context_compressor.get_model_context_length", return_value=100000):
c = ContextCompressor(model="test", quiet_mode=True, protect_first_n=2, protect_last_n=2)
@ -740,16 +757,23 @@ class TestSummaryFailureTrackingForGatewayWarning:
with patch("agent.context_compressor.call_llm", side_effect=Exception("404 model not found")):
result = c.compress(msgs)
assert c._last_summary_fallback_used is True
assert c._last_summary_dropped_count > 0
# Abort flag set, error recorded
assert c._last_compress_aborted is True
assert c._last_summary_error is not None
# Result must still be well-formed (fallback summary present).
assert any(
# No fallback inserted, no messages dropped
assert c._last_summary_fallback_used is False
assert c._last_summary_dropped_count == 0
# Original messages preserved byte-for-byte — the agent loop's
# "did compression help?" check (len(after) < len(before)) sees a
# no-op and stops looping.
assert result == msgs
# No "Summary generation was unavailable" placeholder leaked in.
assert not any(
isinstance(m.get("content"), str) and "Summary generation was unavailable" in m["content"]
for m in result
)
def test_compress_clears_fallback_flag_on_subsequent_success(self):
def test_compress_clears_abort_flag_on_subsequent_success(self):
mock_response = MagicMock()
mock_response.choices = [MagicMock()]
mock_response.choices[0].message.content = "summary text"
@ -768,18 +792,57 @@ class TestSummaryFailureTrackingForGatewayWarning:
{"role": "user", "content": "msg 7"},
]
# First call fails, second succeeds — flag must reset on second compress.
# First call fails, second succeeds — abort flag must reset on second compress.
with patch("agent.context_compressor.call_llm", side_effect=Exception("boom")):
c.compress(msgs)
assert c._last_summary_fallback_used is True
assert c._last_compress_aborted is True
# Reset cooldown to allow retry on second compress
c._summary_failure_cooldown_until = 0.0
with patch("agent.context_compressor.call_llm", return_value=mock_response):
c.compress(msgs)
assert c._last_compress_aborted is False
assert c._last_summary_fallback_used is False
assert c._last_summary_dropped_count == 0
def test_force_true_bypasses_failure_cooldown(self):
"""Manual /compress passes force=True so it can retry immediately
after an auto-compress abort instead of waiting out the 30-60s
cooldown."""
mock_response = MagicMock()
mock_response.choices = [MagicMock()]
mock_response.choices[0].message.content = "summary text"
with patch("agent.context_compressor.get_model_context_length", return_value=100000):
c = ContextCompressor(model="test", quiet_mode=True, protect_first_n=2, protect_last_n=2)
msgs = [
{"role": "system", "content": "sys"},
{"role": "user", "content": "msg 1"},
{"role": "assistant", "content": "msg 2"},
{"role": "user", "content": "msg 3"},
{"role": "assistant", "content": "msg 4"},
{"role": "user", "content": "msg 5"},
{"role": "assistant", "content": "msg 6"},
{"role": "user", "content": "msg 7"},
]
# Pre-populate an active cooldown (as if a prior auto-compress aborted).
import time as _time
c._summary_failure_cooldown_until = _time.monotonic() + 999.0
# Without force, _generate_summary would short-circuit on cooldown
# and return None → abort. With force=True the cooldown is cleared
# and the call goes through.
with patch("agent.context_compressor.call_llm", return_value=mock_response):
result = c.compress(msgs, force=True)
assert c._last_compress_aborted is False
# Cooldown was cleared and a real summary attempt was made.
assert c._summary_failure_cooldown_until == 0.0
# Result is actually compressed (shorter than input).
assert len(result) < len(msgs)
class TestSummaryPrefixNormalization:
def test_legacy_prefix_is_replaced(self):
@ -1338,7 +1401,11 @@ class TestSummaryTargetRatio:
+ [{"role": "user" if i % 2 == 0 else "assistant", "content": f"msg {i}"}
for i in range(8)]
)
result = c.compress(msgs)
mock_resp = MagicMock()
mock_resp.choices = [MagicMock()]
mock_resp.choices[0].message.content = "summary text"
with patch("agent.context_compressor.call_llm", return_value=mock_resp):
result = c.compress(msgs)
# System prompt (msg[0]) survives as head
assert result[0]["role"] == "system"
assert result[0]["content"].startswith("System prompt")