jif-oai

This commit introduces **enhanced maintenance for the logs database** within the `codex-rs/state` module, representing a **new capability** and significant **maintenance improvement**. It configures and utilizes **incremental auto-vacuum** for the logs database, ensuring more frequent and efficient clean-up operations. The new `run_logs_startup_maintenance` function, called during initialization, now handles log retention, WAL checkpointing, and incremental vacuuming. This refactoring also updates the `LOGS_DB_VERSION` and streamlines database connection options, leading to **improved performance and reduced disk usage** for log storage.

This commit **refactors the multi-agent tool argument handling** by **removing the `interrupt` parameter** from the `send_message` tool's public API and internal specification. The core change involved splitting `MessageToolArgs` into `SendMessageArgs` and `AssignTaskArgs`, with `SendMessageArgs` explicitly dropping support for `interrupt`. Consequently, the `send_message` and `assign_task` tool handlers were updated to parse and pass arguments explicitly, and the `create_send_message_tool` function in `codex-rs/tools/src/agent_tool.rs` was modified to reflect this API change. This ensures that `send_message` operations are no longer configurable for interruption, with updated tests verifying the new behavior and specification.

This commit **enforces a stricter API contract for the `spawn_agent` v2 tool**, requiring the use of the `items` field and completely removing the deprecated `message` field. Specifically, the **`SpawnAgentArgs` structure** in `codex-rs/core/src/tools/handlers/multi_agents_v2/spawn.rs` is updated to make `items` mandatory and remove `message`. This **feature enhancement** streamlines agent spawning inputs, impacting the `codex-rs/tools` definition and necessitating comprehensive test updates to validate the new schema and reject legacy usage.

This commit provides a **critical bug fix** within the `codex-rs/core` module, specifically addressing the **thread rollout truncation** logic. It **corrects the computation of fork boundaries** by updating the `fork_turn_positions_in_rollout` function to identify rollback boundaries based on *all instruction turns*, rather than exclusively user turns. This ensures that the system correctly handles scenarios involving **rolled-back assistant instruction turns**, preventing incorrect truncations or state management issues in conversational threads. A new test case has been added to validate the accurate handling of these rolled-back assistant turns, improving the overall robustness of the `truncate_rollout_to_last_n_fork_turns` functionality.

This commit introduces a **V2 fork pattern** for the **multi-agent system**, providing more granular control over the historical context passed to newly spawned agents. It adds a new `fork_turns` property to the `multi_agents_v2` tool, allowing users to specify `none`, `all`, or a positive integer for the number of recent turns to fork. This **new capability** involves significant **refactoring** in the `agent/control` and `thread_rollout_truncation` modules to implement the new `SpawnAgentForkMode` and associated history truncation logic. The change enhances the flexibility and efficiency of multi-agent interactions by precisely controlling the scope of shared context, with extensive **test coverage** for the new modes and truncation logic.

This commit performs a **minor text refinement** within the **`codex-rs/core`** module's **`tasks`** subsystem. It updates the constant string `TURN_ABORTED_INTERRUPTED_GUIDANCE` in `codex-rs/core/src/tasks/mod.rs`. The change involves removing a phrase related to verifying the current state, thereby streamlining the **guidance message** provided when a turn is aborted. This is a **maintenance** task to improve clarity without altering any functional behavior.

This commit **refactors** the **`wait_agent` tool** within the `codex-rs/tools` module to significantly simplify its parameter structure. It removes the `targets` parameter from `wait_agent_tool_parameters_v2`, making `timeout_ms` the sole and optional parameter for the tool. This **cleanup** effort streamlines the `wait_agent`'s API, making its configuration and usage more straightforward. Corresponding **specification tests** in `codex-rs/core` and `codex-rs/tools` have been updated to align with these simplified parameters, ensuring the tool's continued correctness and consistency.

This commit **fixes a bug** in the **session turn management** within the `codex-rs` core library by **refactoring** how new turns are initiated. It updates the `codex_thread` module to use `maybe_start_turn_for_pending_work` after submission responses and clarifies turn initiation in `inter_agent_communication` within the `codex` module. The core changes in the **`tasks`** module reorder logic in `start_turn` and rename turn initiation functions, ensuring that pending work reliably triggers subsequent turns. This improves the correctness and flow of the **agent's turn-taking mechanism**, preventing potential stalls or incorrect state transitions.

This commit delivers a **bug fix** for an issue causing incorrect termination of the "end of turn watcher" within the **`MultiAgentV2` system**. It **conditionally disables** the problematic completion watcher for `MultiAgentV2` threads in the **agent control module** (`codex-rs/core/src/agent/control.rs`) during `spawn_agent` and `resume_agent` calls to prevent incorrect behavior. To ensure proper agent lifecycle management, new functionality was added to the **core codex module** (`codex-rs/core/src/codex.rs`) to **notify parent agents of terminal turn events** from their `MultiAgentV2` child agents. This crucial change prevents premature agent termination and is thoroughly validated by **new tests** covering various completion scenarios.

This commit provides a **minor fix** and **refactoring** to the **multi-agent v2 waiting mechanism** within `codex-rs/core`. Specifically, it reorders the declaration of `mailbox_seq_rx` within the `handle` function in `codex-rs/core/src/tools/handlers/multi_agents_v2/wait.rs`. This ensures that the mailbox sequence receiver is properly initialized before a `session.send_event` call, preventing potential race conditions or incorrect state. The change improves the **robustness and correctness** of event handling during **multi-agent interactions**.

This commit introduces a **new `Mailbox` system** within the `codex-rs/core` crate, enabling robust **inter-agent communication** by allowing messages to be sent, received, and tracked. The existing **`wait_agent` tool** has been **re-implemented** to utilize this mailbox, simplifying its interface by removing the need for explicit `target` agents. This **new capability** integrates deeply with the `Session` management and task handling logic, providing a more structured approach to agent interactions. The change affects the **`agent` module**, `Session` struct, and core task management, improving the overall communication infrastructure and streamlining the `wait` mechanism.

This commit introduces a **new `Op` enum** to standardize agent input and communication, specifically enabling **inter-agent communication during agent spawning**. It involves a significant **refactoring** of the core **agent control logic** in `codex-rs/core/src/agent/control.rs` to utilize this new enum for `spawn_agent` and `send_input` operations. Existing multi-agent tool handlers and memory components are updated to adapt to the `Op` enum, and the `render_input_preview` functionality is centralized for consistency. This **feature** enhances the robustness and flexibility of **multi-agent interactions** by providing a unified protocol for initial agent inputs and communication.

This commit **refactors** the rendering of the review exit success message within the **`codex-rs/core` review task subsystem**. It transitions the templating mechanism to leverage the `codex_utils_template` library, standardizing how such messages are generated. Specifically, the `exit_review_mode` and `render_review_exit_success` functions in `review.rs` are updated to use this new library, and the `exit_success.xml` template's placeholder syntax is adjusted from `{results}` to `{{results}}` to align with `codex_utils_template`'s requirements. This **refactoring** improves consistency and maintainability by adopting a common templating solution, without altering the user-facing output.

This commit **refactors** the **tool description generation** within the `codex-rs/core` module to enhance maintainability and correctness. Specifically, the `create_tool_search_tool` and `create_tool_suggest_tool` functions now utilize the `codex_utils_template` library for templating, replacing previous direct string replacement logic. A corresponding **test enhancement** was added to `spec_tests.rs` to assert that tool suggest descriptions are correctly rendered without exposing raw template placeholders. This internal improvement ensures robust and accurate tool descriptions for the affected search and suggest tools.

This commit **refactors** the internal handling of **collaboration mode presets** within the `codex-rs/core` module. It transitions the generation of default mode instructions to leverage the **`codex-utils-template` library** for more robust and maintainable templating, affecting `collaboration_mode_presets.rs`. Concurrently, duplicated preset logic and constants are eliminated from the **`tui_app_server`**, which now imports these definitions directly from the core module, improving code consolidation. This effort **centralizes the management of collaboration mode instructions**, significantly enhancing code consistency and maintainability across the application, with corresponding test updates in `collaboration_mode_presets_tests.rs`.

This commit **refactors** the generation of **sandbox mode prompts** within the `codex-rs/protocol` crate to utilize the `codex_utils_template` library. The `sandbox_text` generation logic now employs `LazyLock` for efficient template parsing and rendering, improving the system's performance and maintainability. Existing prompt templates for `danger_full_access`, `read_only`, and `workspace_write` modes have been updated to adopt the new `{{network_access}}` template syntax. This **enhancement** standardizes how prompts are constructed, making them more flexible and easier to manage for various sandbox permission levels. A new test case for `DangerFullAccess` mode ensures the correct functioning of the updated templating system.

This commit **refactors** the **review prompt generation** within the `codex-rs/core` module. It migrates the implementation of functions like `review_prompt` and `render_review_prompt` to leverage the `codex-utils-template` library. This change standardizes the templating mechanism for review prompts, improving maintainability and consistency. Additionally, new tests are introduced to ensure the reliability of the refactored prompt generation logic.

This commit **refactors** the **login error page rendering** within the **`codex-rs/login` service** to utilize the `codex_utils_template` library. The `error.html` template was updated to use `{{...}}` placeholders, and the `render_login_error_page` function in `server.rs` now employs the new utility for parsing and rendering. This **refactoring** effort standardizes template syntax and rendering logic, promoting **consistency and maintainability** across the project's templating infrastructure. Additionally, new test cases were introduced to ensure the robustness of the updated rendering process.

This commit introduces a **new feature** that makes the `task_name` parameter **mandatory** when spawning agents within the **multi-agent V2 system**. The **`codex-rs/core` multi-agent V2 spawn handler** (`multi_agents_v2/spawn.rs`) and its associated **tool specification** (`spec.rs`) are updated to enforce this requirement, modifying functions like `handle` and `create_spawn_agent_tool_v2`. This ensures that all newly created V2 agents are explicitly named, enhancing clarity and traceability for spawned processes. **Tests** in `multi_agents_tests.rs` and `spec_tests.rs` have been added or updated to validate this new mandatory parameter.

This commit performs **general code cleanup** by removing unnecessary or redundant elements from the codebase. This **maintenance** task aims to streamline the project, potentially reducing its footprint and improving overall clarity. While specific modules are not detailed, the change contributes to overall project hygiene. The scope of this work is internal, with no direct impact on user-facing features or functionality.