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)
137 lines
5 KiB
Python
137 lines
5 KiB
Python
"""Intervention action types.
|
|
|
|
Each action represents a typed decision that a handler returns after evaluating
|
|
an event. The framework uses these to compose decisions across multiple handlers.
|
|
"""
|
|
|
|
from collections.abc import Callable
|
|
from dataclasses import dataclass, field
|
|
from typing import Any
|
|
|
|
from ..hooks.events import (
|
|
AfterModelCallEvent,
|
|
AfterToolCallEvent,
|
|
BeforeInvocationEvent,
|
|
BeforeModelCallEvent,
|
|
BeforeToolCallEvent,
|
|
)
|
|
|
|
LifecycleEvent = (
|
|
BeforeInvocationEvent | BeforeToolCallEvent | AfterToolCallEvent | BeforeModelCallEvent | AfterModelCallEvent
|
|
)
|
|
|
|
_APPROVED_RESPONSES = {"y", "yes"}
|
|
|
|
|
|
def default_evaluate(response: Any) -> bool:
|
|
"""Default evaluate function for the confirm action.
|
|
|
|
Accepts: True, 'y'/'yes' (case-insensitive, whitespace-trimmed).
|
|
|
|
Args:
|
|
response: The human's response value to evaluate.
|
|
|
|
Returns:
|
|
True if the response is considered an approval, False otherwise.
|
|
"""
|
|
if response is True:
|
|
return True
|
|
if isinstance(response, str):
|
|
return response.lower().strip() in _APPROVED_RESPONSES
|
|
return False
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class Proceed:
|
|
"""Allow the operation to continue unchanged.
|
|
|
|
Args:
|
|
reason: Optional metadata for debugging/logging. Not shown to the model.
|
|
"""
|
|
|
|
type: str = field(default="proceed", init=False)
|
|
reason: str | None = None
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class Deny:
|
|
"""Block the operation. The reason is shown to the model as the cancellation message."""
|
|
|
|
type: str = field(default="deny", init=False)
|
|
reason: str = ""
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class Guide:
|
|
"""Provide feedback to steer behavior.
|
|
|
|
On beforeToolCall/beforeInvocation, sets cancel so the model sees the feedback.
|
|
On beforeModelCall, injects feedback as a user message.
|
|
On afterModelCall, the response is discarded and the model retries with feedback.
|
|
|
|
.. warning::
|
|
On ``after_model_call``, Guide triggers a model retry. Handlers **must** ensure
|
|
convergence (e.g., by tracking retry count and escalating to Deny after repeated
|
|
failures). The framework imposes no retry cap on guide-triggered retries.
|
|
|
|
.. note::
|
|
On ``before_model_call`` and ``after_model_call``, guidance messages are injected
|
|
directly into ``agent.messages`` and bypass session management. Session managers
|
|
will not track these injected messages.
|
|
"""
|
|
|
|
type: str = field(default="guide", init=False)
|
|
feedback: str = ""
|
|
reason: str | None = None
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class Confirm:
|
|
"""Request human approval before proceeding. Only supported on beforeToolCall.
|
|
|
|
Two modes depending on whether response is provided:
|
|
- With response: passed as a preemptive value to the interrupt system, agent never pauses.
|
|
- Without response: breaks out of the agent loop to pause for external resume.
|
|
"""
|
|
|
|
type: str = field(default="confirm", init=False)
|
|
prompt: str = ""
|
|
reason: str | None = None
|
|
response: Any = None
|
|
evaluate: Callable[[Any], bool] | None = field(default=default_evaluate)
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class Transform:
|
|
"""Modify event content in-place.
|
|
|
|
The apply function mutates the event before execution proceeds.
|
|
Later handlers in the pipeline see the transformed content.
|
|
"""
|
|
|
|
type: str = field(default="transform", init=False)
|
|
apply: Callable[[LifecycleEvent], None] = field(default=lambda e: None)
|
|
reason: str | None = None
|
|
|
|
|
|
InterventionAction = Proceed | Deny | Guide | Confirm | Transform
|
|
"""Union of all intervention actions a handler can return.
|
|
|
|
Action-to-event compatibility matrix::
|
|
|
|
| Action | before_invocation | before_tool_call | before_model_call | after_tool_call | after_model_call |
|
|
|-----------|-------------------|------------------|-------------------|-----------------|------------------|
|
|
| Proceed | — | — | — | — | — |
|
|
| Deny | cancel | cancel | cancel | — | — |
|
|
| Guide | cancel+ | cancel+ | inject | — | inject + retry |
|
|
| Confirm | — | confirm | — | — | — |
|
|
| Transform | apply | apply | apply | apply | apply |
|
|
|
|
— = no-op (warns at runtime)
|
|
cancel = sets event.cancel/cancel_tool, short-circuits (remaining handlers skipped)
|
|
cancel+ = sets cancel with accumulated feedback from all guiding handlers
|
|
confirm = uses preemptive response or interrupt, checks with evaluate, sets cancel if denied
|
|
inject = appends accumulated feedback as a user message so the model sees it on this call
|
|
inject + retry = appends accumulated feedback and retries so the model sees guidance
|
|
apply = calls action.apply(event) for in-place mutation, later handlers see the change
|
|
"""
|