vibe.runtime_models.session¶

Session state validation models.

Provides Pydantic models for validating: - Assistant messages (SystemMessage, UserMessage, AssistantMessage, ToolResult) - Tool calls and tool results - Interview session metadata - Session serialization/deserialization

These models are used at session load/save boundaries and when deserializing messages from persisted .vibestate files.

Note: These are validation-only models. The actual dataclass instances from vibe.assistant.strructures are used at runtime. These Pydantic models validate dict data before constructing dataclasses.

ToolCallModel ¶

Validation model for LLM tool invocations.

Validates the structure of tool calls made by the LLM, ensuring required fields are present and well-formed.

validate_non_empty ¶

validate_non_empty(v: str) -> str

Ensure id and name are non-empty strings.

decode_thought_signature ¶

decode_thought_signature(v: bytes | dict[str, Any] | None) -> bytes | None

Decode base64-encoded thought_signature from JSON storage.

SystemMessageModel ¶

Validation model for system messages.

System messages contain instructions and context for the LLM.

UserMessageModel ¶

Validation model for user messages.

User messages contain user input, either as plain text or structured form data.

AssistantMessageModel ¶

Validation model for assistant (LLM) messages.

Assistant messages contain LLM-generated text and optional tool calls.

ToolResultModel ¶

Validation model for tool execution results.

Tool results contain the output from executing a tool call.

validate_call_id ¶

validate_call_id(v: str) -> str

Ensure call_id is non-empty.

MessageHistoryModel ¶

Validation model for a list of messages.

Ensures message history is well-formed and maintains required ordering.

validate_tool_result_references ¶

validate_tool_result_references() -> MessageHistoryModel

Ensure tool results reference valid tool calls.

Each ToolResult must have a corresponding ToolCall earlier in history.

DraftBlockModel ¶

Validation model for assistant draft blocks.

Draft blocks are sections of generated content that may reference questions and contain LLM-generated text.

AssistantSessionModel ¶

Validation model for per-assistant session state.

Each assistant maintains its own session data including message history, draft blocks, and turn tracking.

InterviewSessionModel ¶

Validation model for interview session state.

Validates the core interview session structure stored in Flask session. Uses FlexibleModel to allow for extensions and backward compatibility.

validate_non_empty ¶

validate_non_empty(v: str) -> str

Ensure IDs are non-empty strings.