GitHub user wenjin272 added a comment to the discussion: Agent Skills 
Investigation

> Skills are typically triggered by prompts

Yes. To use skills, the framework must provide a system prompt to the chat 
model, like:
```
## Available Skills

<usage>
Skills provide specialized capabilities and domain knowledge. Use them when 
they match your current task.

How to use skills:
- Load skill: load_skill_through_path(skillId="<skill-id>", path="SKILL.md")
- The skill will be activated and its documentation loaded with detailed 
instructions
- Additional resources (scripts, assets, references) can be loaded using the 
same tool with different paths

Path Information:
When you load a skill, the response will include:
- Exact paths to all skill resources
- Code examples for accessing skill files
- Usage instructions specific to that skill

Template fields explanation:
- <name>: The skill's display name
- <description>: When and how to use this skill
</usage>

<available_skills>
...
```
and after the `available_skills` will be the name and description of each skill.

I think the main difference is, for conversation agents, user could dynamic add 
or remove skills, but for event-driven agents, user must register all the 
skills may be used when define the agents and can't add new skills during the 
runtime.

But, in the future, I think we could also achieve dynamic adding or removing 
agent skills by retrieving them from an MCP server. Because FASTMCP now 
supports skills provider: https://gofastmcp.com/servers/providers/skills. And 
we plan to support dynamic loading tools and prompts from mcp server: 
https://github.com/apache/flink-agents/discussions/543.

GitHub link: 
https://github.com/apache/flink-agents/discussions/536#discussioncomment-15982135

----
This is an automatically sent email for [email protected].
To unsubscribe, please send an email to: [email protected]

Reply via email to