mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-24 16:54:43 +00:00
fix(context-engine): honor quiet compaction status
This commit is contained in:
Stephen Schoettler
Teknium
parent
2ca38e5df4
commit
4035d70bbe
4 changed files with 227 additions and 3 deletions
|
|
@ -53,6 +53,39 @@ def sanitize_memory_context(memory_context: str) -> str:
|
|||
)
|
||||
|
||||
|
||||
def automatic_compaction_status_message(
|
||||
engine: Any,
|
||||
*,
|
||||
phase: str,
|
||||
default_message: str,
|
||||
**context: Any,
|
||||
) -> str | None:
|
||||
"""Resolve host-visible status for an automatic compaction event.
|
||||
|
||||
Engines can suppress routine automatic status with
|
||||
``emit_automatic_compaction_status = False`` or customize it by defining
|
||||
``get_automatic_compaction_status_message(...)``. Empty strings and
|
||||
``None`` mean "do not emit a lifecycle status".
|
||||
"""
|
||||
if not getattr(engine, "emit_automatic_compaction_status", True):
|
||||
return None
|
||||
|
||||
formatter = getattr(engine, "get_automatic_compaction_status_message", None)
|
||||
if callable(formatter):
|
||||
message = formatter(
|
||||
phase=phase,
|
||||
default_message=default_message,
|
||||
**context,
|
||||
)
|
||||
else:
|
||||
message = default_message
|
||||
|
||||
if message is None:
|
||||
return None
|
||||
message = str(message).strip()
|
||||
return message or None
|
||||
|
||||
|
||||
class ContextEngine(ABC):
|
||||
"""Base class all context engines must implement."""
|
||||
|
||||
|
|
@ -89,6 +122,12 @@ class ContextEngine(ABC):
|
|||
protect_first_n: int = 3
|
||||
protect_last_n: int = 6
|
||||
|
||||
# User-visible lifecycle status for automatic host-triggered compaction.
|
||||
# Alternative engines that treat compaction as routine background
|
||||
# maintenance can set this false to keep successful automatic passes silent;
|
||||
# warnings, errors, and explicit manual commands should still surface.
|
||||
emit_automatic_compaction_status: bool = True
|
||||
|
||||
# -- Core interface ----------------------------------------------------
|
||||
|
||||
@abstractmethod
|
||||
|
|
@ -156,6 +195,27 @@ class ContextEngine(ABC):
|
|||
"""
|
||||
return False
|
||||
|
||||
def get_automatic_compaction_status_message(
|
||||
self,
|
||||
*,
|
||||
phase: str,
|
||||
default_message: str,
|
||||
**context: Any,
|
||||
) -> str | None:
|
||||
"""Return user-visible status for automatic host-triggered compaction.
|
||||
|
||||
Return ``None`` to suppress successful automatic lifecycle status for
|
||||
this compaction event. ``phase`` identifies the host call site (for
|
||||
example ``"preflight"`` or ``"compress"``). ``context`` contains
|
||||
best-effort fields such as ``approx_tokens`` and ``threshold_tokens``.
|
||||
|
||||
This hook does not control warning/error messages or explicit manual
|
||||
commands such as ``/compress``.
|
||||
"""
|
||||
if not self.emit_automatic_compaction_status:
|
||||
return None
|
||||
return default_message
|
||||
|
||||
# -- Optional: manual /compress preflight ------------------------------
|
||||
|
||||
def has_content_to_compress(self, messages: List[Dict[str, Any]]) -> bool:
|
||||
|
|
|
|||
|
|
@ -42,7 +42,10 @@ from datetime import datetime
|
|||
from pathlib import Path
|
||||
from typing import Any, Optional, Tuple
|
||||
|
||||
from agent.context_engine import sanitize_memory_context
|
||||
from agent.context_engine import (
|
||||
automatic_compaction_status_message,
|
||||
sanitize_memory_context,
|
||||
)
|
||||
from agent.model_metadata import estimate_request_tokens_rough
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
|
@ -1143,7 +1146,20 @@ def compress_context(
|
|||
f"{approx_tokens:,}" if approx_tokens else "unknown", agent.model,
|
||||
focus_topic,
|
||||
)
|
||||
agent._emit_status(COMPACTION_STATUS)
|
||||
_compaction_status = COMPACTION_STATUS
|
||||
if not force:
|
||||
_compaction_status = automatic_compaction_status_message(
|
||||
agent.context_compressor,
|
||||
phase="compress",
|
||||
default_message=_compaction_status,
|
||||
approx_tokens=approx_tokens,
|
||||
message_count=_pre_msg_count,
|
||||
model=agent.model,
|
||||
focus_topic=focus_topic,
|
||||
)
|
||||
_compaction_status_emitted = bool(_compaction_status)
|
||||
if _compaction_status:
|
||||
agent._emit_status(_compaction_status)
|
||||
_compaction_done_emitted = False
|
||||
|
||||
def _complete_compaction_lifecycle() -> None:
|
||||
|
|
@ -1151,7 +1167,11 @@ def compress_context(
|
|||
if _compaction_done_emitted:
|
||||
return
|
||||
_compaction_done_emitted = True
|
||||
_emit_compaction_done(agent)
|
||||
# A suppressed start (quiet context engine) opened no visible
|
||||
# compaction phase — emit no terminal edge either. Failure warnings
|
||||
# go through agent._emit_warning and are never suppressed here.
|
||||
if _compaction_status_emitted:
|
||||
_emit_compaction_done(agent)
|
||||
|
||||
# ── Compression lock ────────────────────────────────────────────────
|
||||
# Atomic, state.db-backed lock per session_id. Without this, two
|
||||
|
|
|
|||
|
|
@ -36,6 +36,7 @@ from agent.conversation_compression import (
|
|||
PRE_API_COMPRESSION_STATUS_TEMPLATE,
|
||||
conversation_history_after_compression,
|
||||
)
|
||||
from agent.context_engine import automatic_compaction_status_message
|
||||
from agent.display import KawaiiSpinner
|
||||
from agent.error_classifier import FailoverReason, classify_api_error
|
||||
from agent.iteration_budget import IterationBudget
|
||||
|
|
|
|||
|
|
@ -774,6 +774,57 @@ class TestPreflightCompression:
|
|||
assert new_system_prompt == "rebuilt without memory"
|
||||
build_prompt.assert_called_once_with("system prompt")
|
||||
|
||||
def test_compress_context_suppresses_automatic_status_when_engine_opts_out(self, agent):
|
||||
"""Plugin engines can make successful automatic compaction silent."""
|
||||
events = []
|
||||
agent.status_callback = lambda ev, msg: events.append((ev, msg))
|
||||
agent.context_compressor.emit_automatic_compaction_status = False
|
||||
|
||||
def _fake_compress(messages, current_tokens=None, focus_topic=None):
|
||||
events.append(("compress", "started"))
|
||||
return [{"role": "user", "content": f"{SUMMARY_PREFIX}\nPrevious conversation"}]
|
||||
|
||||
with (
|
||||
patch.object(agent.context_compressor, "compress", side_effect=_fake_compress),
|
||||
patch.object(agent, "_build_system_prompt", return_value="new system prompt"),
|
||||
patch("run_agent.estimate_request_tokens_rough", return_value=42),
|
||||
):
|
||||
compressed, new_system_prompt = agent._compress_context(
|
||||
[{"role": "user", "content": "hello"}],
|
||||
"system prompt",
|
||||
approx_tokens=1234,
|
||||
)
|
||||
|
||||
assert compressed == [{"role": "user", "content": f"{SUMMARY_PREFIX}\nPrevious conversation"}]
|
||||
assert new_system_prompt == "new system prompt"
|
||||
assert events == [("compress", "started")]
|
||||
|
||||
def test_compress_context_force_keeps_manual_status_when_engine_opts_out(self, agent):
|
||||
"""Manual /compress remains visible even for quiet automatic engines."""
|
||||
events = []
|
||||
agent.status_callback = lambda ev, msg: events.append((ev, msg))
|
||||
agent.context_compressor.emit_automatic_compaction_status = False
|
||||
|
||||
def _fake_compress(messages, current_tokens=None, focus_topic=None, force=False):
|
||||
events.append(("compress", "started"))
|
||||
return [{"role": "user", "content": f"{SUMMARY_PREFIX}\nPrevious conversation"}]
|
||||
|
||||
with (
|
||||
patch.object(agent.context_compressor, "compress", side_effect=_fake_compress),
|
||||
patch.object(agent, "_build_system_prompt", return_value="new system prompt"),
|
||||
patch("run_agent.estimate_request_tokens_rough", return_value=42),
|
||||
):
|
||||
agent._compress_context(
|
||||
[{"role": "user", "content": "hello"}],
|
||||
"system prompt",
|
||||
approx_tokens=1234,
|
||||
force=True,
|
||||
)
|
||||
|
||||
assert events[0][0] == "lifecycle"
|
||||
assert "Compacting context" in events[0][1]
|
||||
assert events[1] == ("compress", "started")
|
||||
|
||||
def test_preflight_compresses_oversized_history(self, agent):
|
||||
"""When loaded history exceeds the model's context threshold, compress before API call."""
|
||||
agent.compression_enabled = True
|
||||
|
|
@ -826,6 +877,98 @@ class TestPreflightCompression:
|
|||
for ev, msg in status_messages
|
||||
)
|
||||
|
||||
def test_preflight_suppresses_status_when_context_engine_opts_out(self, agent):
|
||||
"""LCM-style engines can keep routine automatic preflight maintenance silent."""
|
||||
agent.compression_enabled = True
|
||||
agent.context_compressor.context_length = 200_000
|
||||
agent.context_compressor.threshold_tokens = 100_000
|
||||
agent.context_compressor.emit_automatic_compaction_status = False
|
||||
|
||||
big_history = []
|
||||
for i in range(20):
|
||||
big_history.append({"role": "user", "content": f"Message {i} padded"})
|
||||
big_history.append({"role": "assistant", "content": f"Response {i} padded"})
|
||||
|
||||
ok_resp = _mock_response(content="After quiet preflight", finish_reason="stop")
|
||||
agent.client.chat.completions.create.side_effect = [ok_resp]
|
||||
status_messages = []
|
||||
agent.status_callback = lambda ev, msg: status_messages.append((ev, msg))
|
||||
|
||||
_rough_calls = {"n": 0}
|
||||
|
||||
def _rough_estimate(*_args, **_kwargs):
|
||||
_rough_calls["n"] += 1
|
||||
return 114_000 if _rough_calls["n"] == 1 else 40_000
|
||||
|
||||
with (
|
||||
patch("agent.conversation_loop.estimate_request_tokens_rough", side_effect=_rough_estimate),
|
||||
patch.object(agent, "_compress_context") as mock_compress,
|
||||
patch.object(agent, "_persist_session"),
|
||||
patch.object(agent, "_save_trajectory"),
|
||||
patch.object(agent, "_cleanup_task_resources"),
|
||||
):
|
||||
mock_compress.return_value = (
|
||||
[{"role": "user", "content": f"{SUMMARY_PREFIX}\nPrevious conversation"}],
|
||||
"new system prompt",
|
||||
)
|
||||
result = agent.run_conversation("hello", conversation_history=big_history)
|
||||
|
||||
mock_compress.assert_called_once()
|
||||
assert result["completed"] is True
|
||||
assert not any(
|
||||
ev == "lifecycle" and "Preflight compression" in msg
|
||||
for ev, msg in status_messages
|
||||
)
|
||||
|
||||
def test_preflight_uses_context_engine_custom_status_message(self, agent):
|
||||
"""Plugin engines can replace generic built-in-compressor wording."""
|
||||
agent.compression_enabled = True
|
||||
agent.context_compressor.context_length = 200_000
|
||||
agent.context_compressor.threshold_tokens = 100_000
|
||||
|
||||
def _custom_status(**kwargs):
|
||||
assert kwargs["phase"] == "preflight"
|
||||
assert kwargs["approx_tokens"] == 114_000
|
||||
assert kwargs["threshold_tokens"] == 100_000
|
||||
return "🔧 LCM context maintenance: preparing compacted context."
|
||||
|
||||
agent.context_compressor.get_automatic_compaction_status_message = _custom_status
|
||||
|
||||
big_history = []
|
||||
for i in range(20):
|
||||
big_history.append({"role": "user", "content": f"Message {i} padded"})
|
||||
big_history.append({"role": "assistant", "content": f"Response {i} padded"})
|
||||
|
||||
ok_resp = _mock_response(content="After custom preflight", finish_reason="stop")
|
||||
agent.client.chat.completions.create.side_effect = [ok_resp]
|
||||
status_messages = []
|
||||
agent.status_callback = lambda ev, msg: status_messages.append((ev, msg))
|
||||
|
||||
_rough_calls = {"n": 0}
|
||||
|
||||
def _rough_estimate(*_args, **_kwargs):
|
||||
_rough_calls["n"] += 1
|
||||
return 114_000 if _rough_calls["n"] == 1 else 40_000
|
||||
|
||||
with (
|
||||
patch("agent.conversation_loop.estimate_request_tokens_rough", side_effect=_rough_estimate),
|
||||
patch.object(agent, "_compress_context") as mock_compress,
|
||||
patch.object(agent, "_persist_session"),
|
||||
patch.object(agent, "_save_trajectory"),
|
||||
patch.object(agent, "_cleanup_task_resources"),
|
||||
):
|
||||
mock_compress.return_value = (
|
||||
[{"role": "user", "content": f"{SUMMARY_PREFIX}\nPrevious conversation"}],
|
||||
"new system prompt",
|
||||
)
|
||||
result = agent.run_conversation("hello", conversation_history=big_history)
|
||||
|
||||
mock_compress.assert_called_once()
|
||||
assert result["completed"] is True
|
||||
lifecycle_messages = [msg for ev, msg in status_messages if ev == "lifecycle"]
|
||||
assert "🔧 LCM context maintenance: preparing compacted context." in lifecycle_messages
|
||||
assert not any("Preflight compression" in msg for msg in lifecycle_messages)
|
||||
|
||||
def test_preflight_defers_when_recent_real_usage_fit(self, agent):
|
||||
"""A noisy rough estimate should not re-compact a recently fitting request."""
|
||||
agent.compression_enabled = True
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue