Conversation
code-crusher
commented
Jan 29, 2026
code-crusher
commented
- Implement skill tool functionality and update dependencies
- Update CHANGELOG.md for v5.3.0
- Update assistant message handling and tool integration
- Fix type error in useSkillTool - use ask instead of say
- fix skill use tool
- clean up skill tool
ContextSummary By MatterAI
🔄 What ChangedIntroduced a Workspace Skills system allowing the assistant to load specialized instructions from 🔍 Impact of the ChangeEnables users to extend assistant capabilities via local markdown files, providing niche context (e.g., specific security rules or framework patterns). Improves performance and resource usage by persisting indexing state across sessions. 📁 Total Files ChangedClick to Expand
🧪 Test Added/RecommendedAdded
🔒 Security Vulnerabilities
ImplementationThe feature uses Screenshots
How to Test
Get in TouchN/A |
![matter-ai-bot[bot]](https://avatars.githubusercontent.com/in/474549?s=60&v=4){kind=small}
There was a problem hiding this comment.
🧪 PR Review is completed: The Skills system implementation is robust, with clean integration into the toolchain. The Code Indexing state persistence changes are well-handled. I've identified a prompt optimization to reduce token usage and correct typos.
Skipped files
CHANGELOG.md: Skipped file pattern
| const skillList = skills | ||
| .map((skill) => { | ||
| return ` - ${skill.metadata.name}: ${skill.metadata.description}` | ||
| }) | ||
| .join("\n") | ||
|
|
||
| return `You are provided Skills below, these skills are to be used by you as per your descretion. The purpose of these skills is to provide you additional niche context for you tasks. You might get skills for React, Security or even third-party tools. Use the tool use_skill to get the skill context: | ||
| ${skillList}` |
There was a problem hiding this comment.
🟠 Logic/Performance
Issue: Listing all skills in the system prompt is redundant as they are already listed in the use_skill tool description (which is required for the model to know valid arguments). This consumes unnecessary tokens. Additionally, there are typos in the text ("descretion", "for you tasks").
Fix: Remove the detailed list from the system prompt and direct the model to refer to the use_skill tool definition.
Impact: Reduces token usage and improves prompt clarity.
| const skillList = skills | |
| .map((skill) => { | |
| return ` - ${skill.metadata.name}: ${skill.metadata.description}` | |
| }) | |
| .join("\n") | |
| return `You are provided Skills below, these skills are to be used by you as per your descretion. The purpose of these skills is to provide you additional niche context for you tasks. You might get skills for React, Security or even third-party tools. Use the tool use_skill to get the skill context: | |
| ${skillList}` | |
| // Skills are already listed in the use_skill tool description | |
| return `You have access to a library of skills that provide specialized instructions for tasks. Use the 'use_skill' tool to access them. Refer to the 'use_skill' tool definition for the list of available skills.` |
|
/matter review-max |
|
I couldn't clone the repository. |
1 participant
