Beyond the current UI bug in the skills interface, it is not intuitive to understand how agents interact with skills.
For example, when clicking on an agent, it shows:
Skills: -
“-” is not clear information. Does this mean there is an issue, or simply that no skills are loaded? If it means none are loaded, it would be clearer to explicitly display “None”.
Beyond this small UI detail, it is unclear how agents actually use their skills.
I have an agent with several attached skills, yet I have never seen them being used in task logs, even when the context seemed highly relevant.
Additionally, the current basic skill selector becomes tedious when attaching multiple skills to an agent. A multi-select list UI would be much more practical.
Once the skill display bug in the dedicated skills interface is fixed, it would also be useful to verify whether the list of enabled skills actually matches the skills shown inside each agent.
There is also a conceptual issue regarding skill behavior (assuming they are actually being used despite the lack of clear logs): default agents.
It should be clarified whether enabling a skill in the dedicated interface:
- Allows all agents, including default agents, to use those skills
or
- Only makes those skills visible/selectable inside each agent configuration.
Personal opinion on the expected behavior
Skills are conceptually designed to be loaded by the LLM only when the request context is relevant to the skill’s name or description.
This means that regardless of the number of skills — even if an agent has 300 — the agent should not read all of them, only the relevant ones.
Because of that, defining whether a skill is enabled or disabled per agent does not seem particularly meaningful.
Instead, there should visually be:
- A list of all discovered skills
- A list of all enabled skills
Agents should naturally be able to consult enabled skills automatically, similar to how it works in Claude Code.
However, users should also have the ability to force a skill to be consulted.
This is why I believe the agent interface should clarify the Skills section with something like:
By default, all enabled skills can be consulted automatically. If you want to force the usage of specific skills, select them in the list below.
In the case of a forced skill, the agent would be required to read it before executing its task.
The primary role of a skill is to train the agent for a specific use case or provide up-to-date documentation before it starts coding or planning work.
Beyond the current UI bug in the skills interface, it is not intuitive to understand how agents interact with skills.
For example, when clicking on an agent, it shows:
Skills: -
“-” is not clear information. Does this mean there is an issue, or simply that no skills are loaded? If it means none are loaded, it would be clearer to explicitly display “None”.
Beyond this small UI detail, it is unclear how agents actually use their skills.
I have an agent with several attached skills, yet I have never seen them being used in task logs, even when the context seemed highly relevant.
Additionally, the current basic skill selector becomes tedious when attaching multiple skills to an agent. A multi-select list UI would be much more practical.
Once the skill display bug in the dedicated skills interface is fixed, it would also be useful to verify whether the list of enabled skills actually matches the skills shown inside each agent.
There is also a conceptual issue regarding skill behavior (assuming they are actually being used despite the lack of clear logs): default agents.
It should be clarified whether enabling a skill in the dedicated interface:
or
Personal opinion on the expected behavior
Skills are conceptually designed to be loaded by the LLM only when the request context is relevant to the skill’s name or description.
This means that regardless of the number of skills — even if an agent has 300 — the agent should not read all of them, only the relevant ones.
Because of that, defining whether a skill is enabled or disabled per agent does not seem particularly meaningful.
Instead, there should visually be:
Agents should naturally be able to consult enabled skills automatically, similar to how it works in Claude Code.
However, users should also have the ability to force a skill to be consulted.
This is why I believe the agent interface should clarify the Skills section with something like:
By default, all enabled skills can be consulted automatically. If you want to force the usage of specific skills, select them in the list below.
In the case of a forced skill, the agent would be required to read it before executing its task.
The primary role of a skill is to train the agent for a specific use case or provide up-to-date documentation before it starts coding or planning work.