feat: models.py - OpenEnv 5-step structure
Browse files
models.py
ADDED
|
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
models.py — Step 1: Define Types
|
| 3 |
+
|
| 4 |
+
Action, Observation dataclasses for the Cross-Session Continuity environment.
|
| 5 |
+
These extend openenv.core types so the framework can serialize/deserialize them.
|
| 6 |
+
"""
|
| 7 |
+
|
| 8 |
+
from openenv.core.env_server.types import Action, Observation
|
| 9 |
+
from pydantic import Field
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
class ContinuityAction(Action):
|
| 13 |
+
"""
|
| 14 |
+
Action for the Cross-Session Continuity environment.
|
| 15 |
+
|
| 16 |
+
The agent specifies which tool to call and its arguments.
|
| 17 |
+
"""
|
| 18 |
+
tool: str = Field(..., description="Tool name: read_file | write_file | run_tests | write_handoff | parse_handoff | submit")
|
| 19 |
+
path: str = Field(default="", description="File path (for read_file / write_file)")
|
| 20 |
+
content: str = Field(default="", description="File content (for write_file) or handoff note (for write_handoff)")
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
class ContinuityObservation(Observation):
|
| 24 |
+
"""
|
| 25 |
+
Observation returned after each action.
|
| 26 |
+
|
| 27 |
+
Provides the agent with rich feedback about the current episode state,
|
| 28 |
+
session, test results, and any errors or warnings.
|
| 29 |
+
"""
|
| 30 |
+
output: str = Field(
|
| 31 |
+
default="",
|
| 32 |
+
description="Primary text output of the tool call",
|
| 33 |
+
)
|
| 34 |
+
session: int = Field(
|
| 35 |
+
default=1,
|
| 36 |
+
description="Current session number (1 or 2)",
|
| 37 |
+
)
|
| 38 |
+
passed: int = Field(
|
| 39 |
+
default=0,
|
| 40 |
+
description="Number of tests passed (run_tests only)",
|
| 41 |
+
)
|
| 42 |
+
total: int = Field(
|
| 43 |
+
default=0,
|
| 44 |
+
description="Total number of tests (run_tests only)",
|
| 45 |
+
)
|
| 46 |
+
auxiliary_reward: float = Field(
|
| 47 |
+
default=0.0,
|
| 48 |
+
description="Shaped reward for this step (training signal only)",
|
| 49 |
+
)
|
| 50 |
+
error: str = Field(
|
| 51 |
+
default="",
|
| 52 |
+
description="Error message if action was invalid or rejected",
|
| 53 |
+
)
|
| 54 |
+
warning: str = Field(
|
| 55 |
+
default="",
|
| 56 |
+
description="Warning message (e.g. approaching step limit)",
|
| 57 |
+
)
|
| 58 |
+
message: str = Field(
|
| 59 |
+
default="",
|
| 60 |
+
description="Informational message (e.g. session transition)",
|
| 61 |
+
)
|
| 62 |
+
retries_left: int = Field(
|
| 63 |
+
default=3,
|
| 64 |
+
description="Remaining retry budget for invalid actions",
|
| 65 |
+
)
|
| 66 |
+
done: bool = Field(
|
| 67 |
+
default=False,
|
| 68 |
+
description="Whether the episode has ended",
|
| 69 |
+
)
|
| 70 |
+
reward: float = Field(
|
| 71 |
+
default=0.0,
|
| 72 |
+
description="Final reward (only set when done=True)",
|
| 73 |
+
)
|