VT Code

What's Changed

0.82.5 - 2026-02-25

Bug Fixes

• Update versioning format to remove 'v' prefix in tags and URLs (081b6f73) (@vinhnx)
• Resolve critical scrolling issue and remove unused slash command handlers (bcb81434) (@vinhnx)
• Correct exec_code policy and update TODO for markdown rendering issue (9b87f88b) (@vinhnx)
• Prevent footer panic when hint is absent, refactor path argument to &Path, and optimize sidebar string truncation. (59525d4b) (@vinhnx)
• Update default model in configuration to glm-5:cloud (1700a7e4) (@vinhnx)
• Correct changelog generation to use the previous release tag instead of a fixed version. (b0437d27) (@vinhnx)
• Update chat input placeholders for clarity and improved user guidance (284a2390) (@vinhnx)
• Update chat input placeholders for clarity and improved user guidance (2e178d32) (@vinhnx)
• Disable scroll indicator in status bar (c5d4373d) (@vinhnx)
• Improve output messages for run completion and error handling (833564dd) (@vinhnguyen2308)

Documentation

• Update documentation and improve clarity on execution plans, architectural invariants, and quality scores (27c61ef7) (@vinhnx)
• Update contributing guidelines to reference CONTRIBUTING.md and AGENTS.md. (b4a1ef7a) (@vinhnx)
• Update documentation for TECH_DEBT_TRACKER and QUALITY_SCORE; add tests for subagent loading and file operations (fb7c0944) (@vinhnx)
• Add a guide for adding new models to AGENTS.md. (8648b584) (@vinhnx)
• Update TODO.md with additional PTY truncate display information and test references (cd3a3850) (@vinhnx)
• Update TODO.md with new tasks and references (049b6cae) (@vinhnx)
• Update TODO.md with examples and improve TUI display for truncated outputs (269395b1) (@vinhnx)

Features

• Add MiniMax M2.5 model support across various providers and update related constants (968963f1) (@vinhnx)
• Add Qwen3 Coder Next model support and update related constants (5a4303e0) (@vinhnx)
• Add skill bundle import/export functionality with zip support (dea5b5b7) (@vinhnx)
• Implement plan mode toggle and strip proposed plan blocks in rendering (79f0327d) (@vinhnx)
• Implement in-process teammate runner and enhance team protocol messaging (4aff9318) (@vinhnx)
• Add /share-log command to export session log as JSON for debugging (64305820) (@vinhnx)
• Implement agent steering mechanism to control runloop execution, including stop, pause, resume, and input injection capabilities. (ac806e0e) (@vinhnx)
• Enhance reasoning display by introducing structured ReasoningSegment with stages and improved rendering in the TUI. (68d07c91) (@vinhnx)
• Use configurable constants for agent session limits and expose the default max context tokens function. (21d1183f) (@vinhnx)
• Introduce agent legibility guidelines and refine steering message variants for clarity and structured output. (52b13dd1) (@vinhnx)
• Add Kimi K2.5 model support across OpenRouter, Ollama, and HuggingFace. (fddc4887) (@vinhnx)
• Add sanitizer module for secret redaction and integrate into output handling (4263808d) (@vinhnx)
• Implement credential storage using OS keyring and file fallback (1e94c71a) (@vinhnx)
• Add timeout handling for turn metadata collection (1b1f91d4) (@vinhnx)
• Implement mouse scroll handling for improved navigation (24a2d640) (@vinhnx)
• Add Qwen3.5-397B-A17B model with hybrid architecture and update configuration (26a9a7ee) (@vinhnx)
• Implement secure storage for custom API keys using OS keyring (3da5a60a) (@vinhnx)
• Add CI workflows for building Linux and Windows binaries; optimize release process (090bebb4) (@vinhnx)
• Add full CI mode to release script for all platforms (326a2c8c) (@vinhnx)
• Refactor build process to use conditional cross compilation for Linux and Windows (d15bb558) (@vinhnx)
• Implement mouse scroll support for TUI session and history picker, and update default agent configuration to Ollama. (db99f4db) (@vinhnx)
• Render GFM tables inside markdown code blocks as tables and prevent word-wrapping for table lines in the TUI. (c90f06e3) (@vinhnx)
• Implement mouse text selection in the TUI and add a new vtcode.toml configuration file. (83567152) (@vinhnx)
• Add Claude Sonnet 4.6 model support and integrate it across model definitions, parsing, catalog, and documentation. (d460c56d) (@vinhnx)
• Implement Gemini 3.1 Pro Preview models with updated token limits and system prompt handling. (dc0742c0) (@vinhnx)
• Implement Gemini prompt caching with TTL using a new CacheControl part and add support for Gemini 3.1 Pro preview models. (8b5b42a1) (@vinhnx)
• Add prompt_cache_key to OpenAI requests for improved cache locality and simplify Responses API usage logic. (45c9002e) (@vinhnx)
• Add top-level cache control to Anthropic requests, with TTL determined by breakpoint consumption. (91c0c9e4) (@vinhnx)
• Standardize MiniMax-M2.5 model identifier, promote it as the default, and update configuration defaults. (ff6dcef6) (@vinhnx)
• Introduce CI cost optimization strategies, add a new --ci-only release mode, and document release workflow details. (dd2f3168) (@vinhnx)
• Add prompt cache key to LLM requests and enhance unified_file tool execution diagnostics. (4073aed6) (@vinhnx)
• Refactor Ollama non-streaming request handling and add a fallback to non-streaming for initial stream failures. (30683331) (@vinhnx)
• Improve spooled tool output handling by verifying file existence and add a mechanism to suppress agent follow-up prompt detection for auto-generated prompts. (083ae71a) (@vinhnx)
• Enhance error handling and recovery mechanisms across various components (7902206c) (@vinhnx)
• Implement tool reentrancy guard to prevent recursive execution and improve panic reporting with better-panic. (44351bf5) (@vinhnx)
• Implement chunked reading for spooled tool outputs with improved agent messaging and update default LLM provider configuration. (b5b0c230) (@vinhnx)
• Add chunked file read spool progress tracking and refine token usage calculation for context management. (ef8f162d) (@vinhnx)
• Generate consolidated checksums.txt for releases and centralize script utilities into common.sh. (4bdb0808) (@vinhnx)
• Implement TaskTracker tool and enhance agent guards and documentation based on NL2Repo-Bench insights. (45a9a159) (@vinhnx)
• Integrate AI agent best practices into system prompts and loop detection for improved planning, root cause analysis, and uncertainty recognition. (91e5e295) (@vinhnx)
• Enhance documentation on grounding, uncertainty, and regression verification; improve loop detection guidance (064ea630) (@vinhnx)
• Enhance AskUserChoice with freeform input, custom labels, placeholders, and default selections. (53e0e111) (@vinhnx)
• Implement freeform text input for wizard modals, guided by system prompt and toggled by the Tab key. (9b54cdd6) (@vinhnx)
• Refine plan mode transitions by adding more aliases, enabling contextual exit confirmations, and providing user guidance. (02bf1476) (@vinhnx)
• Set custom terminal title for VT Code TUI (9459f366) (@vinhnx)
• Migrate changelog generation to git-cliff and update related documentation (34d07804) (@vinhnx)
• Add support for additional tree-sitter languages and make them optional (b1c01414) (@vinhnguyen2308)
• Make tree-sitter language parsers optional and update dependencies to disable default features (5e268705) (@vinhnguyen2308)

Other

• Add architectural invariants, core beliefs, execution plans, quality score, and tech debt tracker documentation (80cf6d68) (@vinhnx)
• Add support for inline skill bundles and network policies in skills (ddb9d280) (@vinhnx)
• Adjust provider configurations and logging. (a649d023) (@vinhnx)
• Refactor exec_policy and command validation; consolidate HTTP client utilities; enhance error handling; improve file operations; update middleware to async; clean up validation cache; and adjust rate limiting implementation. (e6245af6) (@vinhnx)
• Refactor caching logic for improved performance; enhance error context handling; update tests for accurate cache statistics (5a10f0d7) (@vinhnx)
• Refactor apply_patch handler for improved output handling; streamline sandbox policy mapping (67c0e05b) (@vinhnx)
• Refactor agent runner to use AgentSessionState for session management and update related components (083fcac6) (@vinhnx)
• Enhance configuration and logging for agent and hooks; add mock MCP server for integration tests (59a48db6) (@vinhnx)
• Add tests for model picker and prompt refinement functionality (0305390e) (@vinhnx)
• Implement unified session loop for agent execution and remove plugin marketplace commands (17d9597f) (@vinhnx)
• Refactor file operations to use utility functions for directory creation and file reading/writing (7f800b3d) (@vinhnx)
• Implement streaming response rendering with reasoning support and helper functions (b3ba347b) (@vinhnx)
• Refactor LLM request handling, improve reasoning processing, and enhance tool command parsing (896f6c69) (@vinhnx)
• Refactor agent runner and tool registry for improved error handling and normalization; enhance loop detection and add tests for new functionality. (101a2b07) (@vinhnx)
• Enhance agent runner settings and improve loop detection; update CLI commands for reasoning effort and verbosity; refine patch handling in file operations; adjust markdown diff summary; modify configuration for LLM provider. (ff221d1d) (@vinhnx)
• Update LLM provider configuration to use Ollama and remove unused typos.toml file (ca83bdcd) (@vinhnx)
• Remove custom prompts feature and related code (dfc3ec5d) (@vinhnx)
• Remove custom prompts feature and related code; update error handling for agent actions; refine UI shortcuts and command handling (cfcc9765) (@vinhnx)
• Add additional_agent_dirs configuration option to vtcode.toml (0f6d1747) (@vinhnx)
• Update default model in vtcode.toml to minimax-m2:cloud; format code for consistency (e024a48d) (@vinhnx)
• Add scripts for documentation link validation and markdown location checks (7290fc5b) (@vinhnx)
• Enhance model behavior configuration for LLM providers (0fa12334) (@vinhnx)
• Update TODO.md (3391d6b9) (@vinhnx)
• Add spool chunk read tracking and enforce limits per turn (c09ef6d3) (@vinhnx)
• Increase max tool calls per turn to 48 and implement budget warning system in harness (7b6cade0) (@vinhnx)
• Rename UpdatePlanCommand to TaskTrackerCommand and refactor related files and documentation (b4520273) (@vinhnx)
• Add plan-mode task tracker with CRUD functionality and integrate with existing tools (56c7e3b8) (@vinhnx)
• Add navigation loop guidance and improve plan mode handling (5d162ae6) (@vinhnx)
• Refactor user input option generation and enhance markdown rendering in modals (a72a63c0) (@vinhnx)
• Add reference to git-cliff for changelog generation (8dd703f1) (@vinhnx)
• Refactor tool call handling and configuration limits (fcc783dd) (@vinhnx)
• Refactor registry and facade methods for improved error handling and readability; update configuration defaults and themes. (5ea7b7e1) (@vinhnx)
• Refactor error handling in various modules to improve resilience and logging; enhance cache management and concurrency controls. (035a14bb) (@vinhnx)
• Enhance TTY integration using crossterm's IsTty for consistent terminal handling (2a4e71d6) (@vinhnx)
• Implement unified exec terminal panel and enhance PTY session handling (f8505ba9) (@vinhnx)
• Refactor run command output handling and improve command summary rendering (59e8fab8) (@vinhnx)
• Implement plan artifact write handling and adjust LLM timeout logic (5edb5d66) (@vinhnx)
• Enhance timeout messages to include "continue" nudge for stalled turns in plan mode (03cd0a65) (@vinhnx)
• Enhance plan mode interview options with recommended choices for goals, constraints, and verification (3263981c) (@vinhnx)
• Refactor tool interaction: replace ask_user_question with request_user_input, enhance error classification, and improve function declaration caching. (2af17e60) (@vinhnx)
• Remove unified symbol extractor and related patterns; update language summarization to use file extensions; enhance README for clarity and feature descriptions; clean up unused imports in tools module. (aca4dbef) (@vinhnguyen2308)

Refactors

• Remove unused ReasoningSegment import from turn_processing.rs (9f4672d7) (@vinhnx)
• Reimplement LLM streaming and event handling using AgentSessionController and its event sink mechanism. (95bcd08f) (@vinhnx)
• Extract large event handler modules into smaller files for improved navigation (9eda39e4) (@vinhnx)
• Streamline file operations and enhance workspace path handling (f7ebb78d) (@vinhnx)
• Replace manual file operations with shared utility functions for consistency (54c447af) (@vinhnx)
• Replace manual file operations with shared utility functions for consistency (7fcbe5f2) (@vinhnx)
• Replace manual file operations with shared utility functions for consistency (15c45f9a) (@vinhnx)
• Consolidate duplicated logic across workspace crates into shared utility functions (a9df41fd) (@vinhnx)
• Streamline MCP tool management and indexing in ToolRegistry (a5c3677b) (@vinhnx)
• Remove Rust cache step from release workflow (e36c5f55) (@vinhnx)
• Clean up code formatting and improve readability across multiple files (b1ae3ad9) (@vinhnx)
• Remove unused imports and enhance configuration for credential storage (b79f2bd2) (@vinhnx)
• Make TUI signal cleanup and dotfile permission backup UNIX-specific. (14a4f2d2) (@vinhnx)
• Improve previous SemVer tag identification by searching commit history in release scripts (0fcdce3d) (@vinhnx)
• Remove unified tool resolver module to streamline codebase (1384fe03) (@vinhnx)
• Remove unused TUI components and improve event handling for focus changes (99fa2620) (@vinhnx)
• Standardize continuation prefix handling in wrap_block_lines functions (0596e899) (@vinhnx)

Full Changelog: https://github.com/vinhnx/vtcode/compare/0.82.5