Spaces:
Sleeping
Sleeping
File size: 9,881 Bytes
116524e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 | """ClaudeCode — Claude Code CLI runner with ACE learning."""
from __future__ import annotations
from collections.abc import Iterable, Sequence
from pathlib import Path
from typing import Any, Optional
from pydantic_ai.settings import ModelSettings
from pipeline import Pipeline
from pipeline.protocol import SampleResult, StepProtocol
from ..core.context import ACEStepContext, SkillbookView
from ..core.skillbook import Skillbook
from ..integrations import wrap_skillbook_context
from ..integrations.claude_code import ClaudeCodeExecuteStep, ClaudeCodeToTrace
from ..protocols import (
DeduplicationConfig,
DeduplicationManagerLike,
ReflectorLike,
SkillManagerLike,
)
from ..steps import learning_tail
from .base import ACERunner
class ClaudeCode(ACERunner):
"""Claude Code CLI with ACE learning pipeline.
INJECT skillbook → EXECUTE Claude Code → LEARN (Reflect → Tag → Update → Apply).
Two construction paths:
1. ``ClaudeCode.from_roles(reflector, skill_manager, working_dir=..., ...)``
— pre-built roles.
2. ``ClaudeCode.from_model(working_dir=..., ace_model="gpt-4o-mini", ...)``
— builds ACE roles from a model string.
Example::
runner = ClaudeCode.from_model(working_dir="./my_project")
results = runner.run([
"Add unit tests for utils.py",
"Refactor the auth module",
])
runner.save("code_expert.json")
"""
@classmethod
def build_steps(
cls,
*,
reflector: ReflectorLike,
skill_manager: SkillManagerLike,
skillbook: Skillbook | None = None,
skillbook_path: Optional[str] = None,
working_dir: Optional[str] = None,
timeout: int = 600,
model: Optional[str] = None,
allowed_tools: Optional[list[str]] = None,
dedup_config: Optional[DeduplicationConfig] = None,
dedup_manager: DeduplicationManagerLike | None = None,
dedup_interval: int = 10,
checkpoint_dir: str | Path | None = None,
checkpoint_interval: int = 10,
) -> list[StepProtocol]:
"""Return the steps that ``from_roles()`` would compose.
Use this to inspect, modify, or extend the pipeline before
constructing it yourself::
steps = ClaudeCode.build_steps(reflector=r, skill_manager=sm, ...)
steps.insert(2, MyCustomStep())
pipe = Pipeline(steps)
runner = ACERunner(pipeline=pipe, skillbook=skillbook)
Args:
reflector: Reflector role for analysing execution traces.
skill_manager: SkillManager role for update operations.
skillbook: Starting skillbook. Creates an empty one if ``None``.
skillbook_path: Path to load skillbook from.
working_dir: Directory where Claude Code executes.
timeout: Execution timeout in seconds.
model: Optional Claude model override.
allowed_tools: Optional list of allowed tools.
dedup_config: Deduplication configuration.
dedup_manager: Optional pre-built deduplication manager.
dedup_interval: Samples between deduplication runs.
checkpoint_dir: Directory for checkpoint files.
checkpoint_interval: Samples between checkpoint saves.
"""
# Resolve skillbook
if skillbook_path:
skillbook = Skillbook.load_from_file(skillbook_path)
elif skillbook is None:
skillbook = Skillbook()
# Resolve dedup manager
dm = dedup_manager
if dm is None and dedup_config is not None:
from ..deduplication import DeduplicationManager
dm = DeduplicationManager(dedup_config)
steps: list[StepProtocol[ACEStepContext]] = [
ClaudeCodeExecuteStep(
working_dir=working_dir,
timeout=timeout,
model=model,
allowed_tools=allowed_tools,
),
ClaudeCodeToTrace(),
*learning_tail(
reflector,
skill_manager,
skillbook,
dedup_manager=dm,
dedup_interval=dedup_interval,
checkpoint_dir=checkpoint_dir,
checkpoint_interval=checkpoint_interval,
),
]
return steps
@classmethod
def from_roles(
cls,
*,
reflector: ReflectorLike,
skill_manager: SkillManagerLike,
skillbook: Skillbook | None = None,
skillbook_path: Optional[str] = None,
working_dir: Optional[str] = None,
timeout: int = 600,
model: Optional[str] = None,
allowed_tools: Optional[list[str]] = None,
dedup_config: Optional[DeduplicationConfig] = None,
dedup_manager: DeduplicationManagerLike | None = None,
dedup_interval: int = 10,
checkpoint_dir: str | Path | None = None,
checkpoint_interval: int = 10,
) -> ClaudeCode:
"""Construct from pre-built role instances.
Args:
reflector: Reflector role for analysing execution traces.
skill_manager: SkillManager role for update operations.
skillbook: Starting skillbook. Creates an empty one if ``None``.
skillbook_path: Path to load skillbook from.
working_dir: Directory where Claude Code executes.
timeout: Execution timeout in seconds.
model: Optional Claude model override.
allowed_tools: Optional list of allowed tools.
dedup_config: Deduplication configuration.
dedup_manager: Optional pre-built deduplication manager.
dedup_interval: Samples between deduplication runs.
checkpoint_dir: Directory for checkpoint files.
checkpoint_interval: Samples between checkpoint saves.
"""
# Resolve skillbook (must match build_steps resolution)
if skillbook_path:
skillbook = Skillbook.load_from_file(skillbook_path)
elif skillbook is None:
skillbook = Skillbook()
steps = cls.build_steps(
reflector=reflector,
skill_manager=skill_manager,
skillbook=skillbook,
working_dir=working_dir,
timeout=timeout,
model=model,
allowed_tools=allowed_tools,
dedup_config=dedup_config,
dedup_manager=dedup_manager,
dedup_interval=dedup_interval,
checkpoint_dir=checkpoint_dir,
checkpoint_interval=checkpoint_interval,
)
return cls(pipeline=Pipeline(steps), skillbook=skillbook)
@classmethod
def from_model(
cls,
*,
working_dir: Optional[str] = None,
ace_model: str = "gpt-4o-mini",
ace_max_tokens: int = 2048,
ace_temperature: float = 0.0,
**kwargs: Any,
) -> ClaudeCode:
"""Build ACE roles from a model string.
Args:
working_dir: Directory where Claude Code executes.
ace_model: Model identifier for ACE roles.
ace_max_tokens: Max tokens for ACE LLM responses.
ace_temperature: Sampling temperature for ACE roles.
**kwargs: Forwarded to :meth:`from_roles`.
"""
from ..implementations import Reflector, SkillManager
model_settings = ModelSettings(
temperature=ace_temperature,
max_tokens=ace_max_tokens,
)
return cls.from_roles(
reflector=Reflector(ace_model, model_settings=model_settings),
skill_manager=SkillManager(ace_model, model_settings=model_settings),
working_dir=working_dir,
**kwargs,
)
def run(
self,
tasks: Sequence[str] | Iterable[str] | str,
epochs: int = 1,
*,
wait: bool = True,
) -> list[SampleResult]:
"""Run coding tasks with learning.
Args:
tasks: Single task string or list of task strings.
Must be a ``Sequence`` for ``epochs > 1``.
epochs: Number of passes over all tasks.
wait: If ``True``, block until background learning completes.
"""
if isinstance(tasks, str):
tasks = [tasks]
return self._run(tasks, epochs=epochs, wait=wait)
def _build_context( # type: ignore[override]
self,
task: str,
*,
epoch: int,
total_epochs: int,
index: int,
total: int | None,
global_sample_index: int,
**_: Any,
) -> ACEStepContext:
"""Place a raw task string on ``ctx.sample``."""
return ACEStepContext(
sample=task,
skillbook=SkillbookView(self.skillbook),
epoch=epoch,
total_epochs=total_epochs,
step_index=index,
total_steps=total,
global_sample_index=global_sample_index,
)
# ------------------------------------------------------------------
# Convenience lifecycle methods
# ------------------------------------------------------------------
def get_strategies(self) -> str:
"""Return formatted skillbook strategies for display."""
if not self.skillbook.skills():
return ""
return wrap_skillbook_context(self.skillbook)
# Backward-compat aliases
save_skillbook = ACERunner.save
load_skillbook = ACERunner.load
wait_for_learning = ACERunner.wait_for_background
|