b60439152b
- harness/es.py: append-only event store + проекции (образец — кооп-портал) - harness/levels.py: уровни 1с/1м/1ч/1д, контекст ∝ масштабу - harness/scheduler.py: уровневые потоки, идемпотентные тики, absorbed-стабы - harness/isolated_runner.py + _llm_worker.py: LLM-тик в сабпроцессе (глобальный asyncio-рантайм Strands не переживает 2+ create_harness) - cli.py: init/task/tasks/status/events + requeue (написан самим минутным агентом) - 9router: отдельный ключ sk-simbiot, модель auto (glm-5.3 первым)
69 lines
2.5 KiB
Python
69 lines
2.5 KiB
Python
"""Strategy presets — named building blocks that resolve to concrete strategy configurations.
|
|
|
|
Preset definitions (thresholds, methods, conditions) are internal defaults and may
|
|
change between releases as we tune based on real-world usage data. Treat the preset
|
|
name as the stable contract, not its expansion. Use raw ``Offload.*`` strategies when
|
|
you need a specific, pinned configuration.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from .strategies.offload import Offload
|
|
from .types import ContextStrategy
|
|
|
|
StrategyPresetName = str
|
|
"""Known preset name strings accepted in the strategies array.
|
|
|
|
Preset names are the stable contract. The strategies they resolve to may change
|
|
between releases.
|
|
|
|
- ``'proactive_summarization'`` — batch summarize oldest messages at 70% utilization
|
|
- ``'large_tool_offloading'`` — truncate tool results over 2500 tokens to a 1000-token preview
|
|
- ``'overflow_protection'`` — truncate oldest messages when the context window is full
|
|
- ``'stale_tool_cleanup'`` — drop tool results older than 5 messages
|
|
"""
|
|
|
|
STRATEGY_PRESET_NAMES: tuple[str, ...] = (
|
|
"proactive_summarization",
|
|
"large_tool_offloading",
|
|
"overflow_protection",
|
|
"stale_tool_cleanup",
|
|
)
|
|
|
|
|
|
def _resolve_preset(name: str) -> list[ContextStrategy]:
|
|
"""Resolve a preset name to its default strategy array.
|
|
|
|
Args:
|
|
name: The preset name.
|
|
|
|
Returns:
|
|
The strategy array for the given preset.
|
|
"""
|
|
if name == "proactive_summarization":
|
|
return [Offload.summarize("*").when(utilization=0.7, preserve_recent=0.7)]
|
|
if name == "large_tool_offloading":
|
|
return [Offload.truncate("tool_results", {"preview_tokens": 1000}).when(threshold=2500)]
|
|
if name == "overflow_protection":
|
|
return [Offload.truncate("*").when(utilization=1.0, preserve_recent=4)]
|
|
if name == "stale_tool_cleanup":
|
|
return [Offload.drop("tool_results").when(preserve_recent=5)]
|
|
raise ValueError(f'Unknown strategy preset: "{name}". Valid presets: {", ".join(STRATEGY_PRESET_NAMES)}')
|
|
|
|
|
|
def _resolve_strategies(entries: list[ContextStrategy | str]) -> list[ContextStrategy]:
|
|
"""Resolve a mixed list of strategies and preset names into a flat strategy list.
|
|
|
|
Args:
|
|
entries: List of raw strategies and/or preset name strings.
|
|
|
|
Returns:
|
|
Flattened list of concrete strategies.
|
|
"""
|
|
strategies: list[ContextStrategy] = []
|
|
for entry in entries:
|
|
if isinstance(entry, str):
|
|
strategies.extend(_resolve_preset(entry))
|
|
else:
|
|
strategies.append(entry)
|
|
return strategies
|