Mint v1.4.2
Mint Release Notes
v1.4.2 - Web Search & Specialized Personas
Mint 1.4.2 introduces powerful information retrieval capabilities with integrated Web Search, smarter file discovery, and a suite of specialized agent personas with consistent "Mint" personality traits.
Highlights
- Integrated Web Search: The agent can now search the internet to provide up-to-date information.
- Specialized Personas: Six distinct agent personas (Default, Coder, Researcher, Creative, Manager, Reviewer) with a consistent "Mint" personality.
- Smart Path Discovery: New
find_pathtool for finding files and folders by name or partial path. - Enhanced Thai Support: Refined personality and polite particles ("ค่ะ", "นะคะ") when communicating in Thai.
- Deep Context Integration: Better chat history management for seamless multi-turn conversations.
New Features
-
Web Search Integration
- Added
web_searchtool tocode_agent. - Enables the agent to fetch real-time data from the web when it lacks local knowledge.
- Added
-
Persona-Driven Experts
- Introduced the
Agent Orchestratorwith six specialized roles:- Mint Default: Friendly and versatile assistant.
- Mint Coder: Technical expert focusing on best practices.
- Mint Researcher: Analytical role focused on data and facts.
- Mint Creative: Expressive partner for brainstorming.
- Mint Manager: Efficiency-focused role for task planning.
- Mint Reviewer: Senior critic for code and content quality.
- Introduced the
-
Improved File Operations
- New
find_pathutility to locate files and directories within the workspace. - Smarter path resolution for all file system actions.
- New
-
Refined Agent Personality
- Standardized "Mint" personality across all modules.
- Added specific Thai language support to ensure polite and consistent responses in Thai.
Improvements & Fixes
-
CLI Stability
- Improved CLI input handling and streaming performance.
- Better reporting of tool execution in the terminal UI.
-
Router Improvements
- Enhanced intent detection in
chat_routerto better distinguish between general chat and complex coding tasks. - Added more Thai keyword support for better localization.
- Enhanced intent detection in
-
Tool Execution
- Hardened tool execution paths and improved error reporting when tools fail.
- Better handling of large tool outputs with truncation logic.
Documentation & Maintenance
- Major overhaul of
README.mdto reflect new capabilities and architecture. - Updated
BUILD_AND_RELEASE.mdwith refined release procedures. - Cleaned up dependency list in
package.json.
Testing
- Added tests for
findPathandwebSearchhelpers. - Verified persona switching and personality consistency.
- Current test status:
- 85 tests passed
v1.4.1 - CLI Agent Hardening & Stability
Mint 1.4.1 focuses on making the CLI and coding agent behave more like a real agent, while tightening several unsafe execution paths and improving reliability across provider selection, workspace handling, and action execution.
Highlights
- CLI one-shot chat now executes returned actions, not just text responses
- Code Mode now uses only supported coding providers instead of silently falling through
- Agent loop is more resilient when a model returns malformed JSON
- Workspace detection is safer and no longer matches sibling paths by prefix
- Multiple shell execution paths were hardened to reduce command injection risk
CLI & Agent Improvements
-
One-shot CLI action execution
mint chat <message>and equivalent one-shot paths now execute structured actions the same way as interactive chat.- This fixes cases where Mint would say it was opening or searching for something but would not actually perform the action.
-
Better Code Mode provider routing
- Code Mode now selects from providers that are actually supported by the coding workflow:
geminianthropicopenailocal_openai
- Unsupported coding providers such as
ollamaorhuggingfaceno longer appear to be active while silently falling back elsewhere.
- Code Mode now selects from providers that are actually supported by the coding workflow:
-
More robust agent loop
- Code Mode now attempts JSON repair when a provider returns malformed structured output.
- Instead of failing immediately on the first invalid JSON response, Mint asks the model to reformat the reply into valid JSON and retries.
-
Accurate step reporting
- Code Mode now reports the actual number of executed steps instead of always showing the maximum configured step count.
Security & Safety Fixes
-
Hardened URL and file opening paths
- Replaced shell-string execution with argument-based process execution in browser and file action handlers.
-
Hardened app launching
open_appno longer builds launcher commands through shell interpolation for common launch paths.
-
Hardened Docker plugin
- Docker actions now use argument-based execution instead of shell-string concatenation.
-
Safer workflow process matching
- Custom workflow process-name matching now escapes regex characters before building a matcher.
Workspace & Config Fixes
-
Workspace path matching fixed
- Workspace detection no longer treats sibling folders with the same prefix as the same project.
- Example:
/projectno longer incorrectly matches/project-two
-
Workspace test isolation
- Workspace storage can now be redirected for tests through an override path, making tests more reliable in restricted environments.
-
Proactive cooldown config consistency
- Proactive suggestion cooldown now reads from the same config source as the rest of the app.
-
Provider fallback logic improved
- Non-Gemini providers are no longer blocked by a missing Gemini API key before routing even begins.
Documentation
- Rewrote the README to better reflect Mint’s current state as a desktop assistant plus CLI coding agent
- Clarified current agent capabilities, supported workflows, and provider behavior
Testing
-
Expanded regression coverage for:
- Docker plugin execution
- Provider routing helpers
- Code agent helper behavior
- Workspace path boundary handling
-
Current test status:
- 77 tests passed
Notes
Mint 1.4.1 is not a major feature release. It is a stabilization release aimed at making existing agentic workflows safer, more honest about what provider is actually being used, and more reliable in day-to-day CLI usage.