v0.32.2

[0.32.2] - 2026-07-12

Changed

• CLI Agent Prefix: Renamed cli: prefix to agent: across all providers, session, config, and discovery modules.
• CLI Agent Registry: Removed gemini-cli, claude, aider from supported CLI agents. Added pi, copilot (GitHub Copilot CLI), and devin.
• Gemini-Specific Logic Removed: Stripped GEMINI.md prompt file injection and Gemini session management from configure_cli_agent().
• Dead Code Cleanup: Removed Gemini API fetcher, Claude/Aider fallback model lists, and associated unused imports from model_discovery.py.
• Status Bar Model Display: Status bar now shows the underlying CLI agent model (e.g., opencode / o3-pro) instead of just the agent name.

Fixed

• CLI Argument Order: Fixed argument construction for CLI agents — subcommand (exec, run) now correctly precedes the --model flag. Was producing codex --model o4-mini exec --json instead of codex exec --json --model o4-mini.
• CLI Agent Prompt Header: Prompt now includes ENGINE and MODEL lines so the external agent knows which model it's running as.

v0.32.1

[0.32.1] - 2026-07-12

Fixed

• Empty Response Handling: When thinking-only models (e.g., gemma4) produce only `` content with no trailing text, the thinking is now emitted as the text response so the renderer has content to display. Guarded chunk["choices"][0] access to prevent `IndexError` on empty chunks.
• Error Event Handling: Orchestrator error events are now handled gracefully in stream_response() instead of silently dropped.

v0.32.0

v0.32.0 — 2026-07-12

Security (CRITICAL)

• Sandbox Validation: Added command validation against dangerous shell patterns (command substitution, /etc redirects, mkfs, dd) while preserving normal shell features.
• Subprocess Safety: Routed 12+ direct subprocess.run calls through core/subprocess_safety.py with command whitelisting in git_logic, system_tools, model_discovery, paths, and ui_utils.
• CLI Provider: Replaced 6 bare excepts with specific exception types (ValueError, JSONDecodeError) for proper error tracking.
• SECURITY.md: Customized with real project versions (0.31.x, 0.30.x) and vulnerability reporting guidelines.

Code Quality (HIGH)

• repl_tool.py: Eliminated self.__init__() anti-pattern → clean _restart() method with extracted _start_process().
• Type Annotations: Typed CLIProvider.config as SettingsProvider instead of Any; removed dead ui_adapter parameter from ChatAgent.
• Dead Code Removal: Removed dead load_api_key from GemStyleRenderer, inline class-body imports (logging/os/Any), and no-op if/pass block in compression.py.

Features

• Rate Limit Headers: Added parse_rate_limit_headers() utility to parse Retry-After, X-RateLimit-Remaining, and X-RateLimit-Reset from API responses.

v0.31.0

v0.31.0 — 2026-07-12

Fixed

• i18n Auto-Detection: Restored the commented-out _detect_language() call so mentask auto-detects the system locale instead of always defaulting to English.
• Exception Handling: Replaced 37 bare except Exception: blocks with proper exc_info=True logging across core modules (history, memory, metrics, rag, ollama, audit, trust).
• Sandbox Logging: Added debug logging for command execution in LocalSandbox and DockerSandbox.
• Logger Unification: Migrated 30+ files from generic getLogger("mentask") to module-specific loggers (e.g., mentask.core.i18n, mentask.agent.chat).

Changed

• ChatAgent Decomposition: Extracted tool registry factory into agent/tool_factory.py, reducing ChatAgent's responsibilities.
• Config Typing: Added SettingsProvider Protocol in config_manager.py; typed BaseProvider.config instead of Any.
• Command Registry: Replaced monolithic if/elif chain in commands.py with O(1) dict lookup via _build_command_registry().
• Legacy Tools Cleanup: Relocated FILE_SESSIONS to core/constraints.py, decoupling orchestrator from legacy tools/ directory.
• Retry Strategy: Enhanced to detect transient API errors (HTTP 429/503, network errors) with exponential backoff and jitter.

v0.30.0

v0.30.0 — 2026-06-20

Added

• RAG Persistent Cache: SQLite-backed disk cache for indexed workspaces. Bypasses re-computation on subsequent queries unless files are modified.
• Git-Cliff Configuration: Setup automatic generation of CHANGELOG.md via cliff.toml and a custom Makefile recipe make changelog.
• Dynamic CLI Changelog: The --list changelog option now dynamically parses and formats the latest release notes from CHANGELOG.md using Rich Markdown.
• Developer Guidelines: Created a comprehensive CONTRIBUTING.md outlining project setup, code quality checks, pytest conventions, and Conventional Commits.

Changed

• MCP Resource Management: Migrated MCP stdio and client sessions to use AsyncExitStack for guaranteed lifecycle cleanup on failures or shutdown.
• Strict Docstring Linting: Integrated Google Style docstring linting (D rules) with Ruff and formatted all existing modules to comply.

v0.29.0

v0.29.0 — 2026-06-19

Added

• Multi-Agent CLI Bridge: Seamless integration for external CLI binaries (gemini-cli, codex, opencode) with hierarchical model discovery and dynamic autocompletion.
• Git Mastery Integration: Introduced commit_changes tool for structured Conventional Commits, including automated staging and diff previsualization.
• New Core Tools: Added GitCommitTool and git_logic.py to ensure high-quality repository history.
• Dynamic API Key Detection: Configured ConfigManager to automatically resolve and switch the default model to available alternatives depending on which API keys are configured.

Changed

• Slash Commands Modularization: Refactored the command handling logic to split it from the dispatcher class into a structured package src/mentask/agent/core/command_handlers/ with category-scoped modules.
• Google GenAI Decoupling: Deferred imports of google.genai in tools_registry.py and changed signatures to Any to prevent startup errors on environments without the SDK installed.

Fixed

• CLI Bridge Event Leak: Resolved a visual bug where raw JSONL events leaked to the terminal due to stdout buffering.
• CLI Streaming Parsers: Extended JSONL streaming to support Codex, OpenCode, and Gemini schemas.
• Actual Token Metrics: Extracted precise input_tokens and output_tokens directly from CLI JSONL streams.
• Context Snapping Crashes: Fixed a KeyError on status-only chunks and a reference bug that mutated history.
• Test Suite Resilience: Resolved critical integration failures related to Ollama PATH issues.
• Provider Stability: Enabled streaming of tool_call, info, and error events for all external providers.

v0.27.9

[0.27.9] - 2026-05-12

Fixed

• Worktree Tools: Fixed an IndentationError in src/mentask/tools/worktree_tools.py that prevented the CLI from starting.
• Improved Tooling: Added comprehensive unit tests for worktree tools to ensure stability.

v0.27.8

[0.27.8] - 2026-05-12

Added

• Thought Process Toggle: Introduced the /thinking [true|false] command to show or hide the agent's internal reasoning loop.
• Unified Status Line: Merged the session status bar and turn divider into a single, cohesive line that matches the user's prompt theme.

Fixed

• CLI Bridge Robustness: Refactored CLIProvider to use stdin for prompt delivery, bypassing OS command-line length limits on Windows.
• Improved Diagnostics: Implemented parallel reading of stdout and stderr for CLI providers, enabling the capture of system warnings and errors.
• UI Layout: Removed aggressive grid-based padding in agent responses for a more natural conversational flow.
• Error Filtering: Automatically suppress repetitive system warnings (e.g., Windows 10 detected, Ripgrep missing) in CLI output.

v0.27.7

[0.27.7] - 2026-05-12

Fixed

• GemStyleRenderer: Resolved a critical NameError: name 'self' is not defined during initialization by fixing the default parameter signature in the __init__ method.

v0.27.5

[0.27.5] - 2026-05-12

Added

• Selective Memory via Side-Query: Implemented a lightweight "sidechain" selection mechanism to identify and inject only the most relevant memory files into the context.
• Memory Metadata Caching: Introduced .metadata_cache.json in the memories directory to support thousands of learning files with zero performance lag.
• Intelligent Context Pre-selection: Refactored ContextManager and ChatAgent to automatically pre-select relevant context based on the user query before every turn.