Python: [Feature]: Run argument skills support

[supraano]

Description

Scope

Currently, it is possible to specify tools that are only available for a specific agent run loop. It is desired that for skills this is also supported.

Why is this useful?

• For mode switches: Agentic applications could have different operating modes with more or less skills available
• Permissions and user management: Agentic applications should expose skills only to specific users

Code Sample

@tool(...)
def get_weather(...):
    ...

agent = Agent(...)
agent.run(messages="Whats the weather like in NY?", tools=[get_wheather]) # this is already supported
agent.run(messages="Whats the wheather like in NY?", skills=[...]) # this is the added behavior

Language/SDK

Python

[semenshi]

Currently, AiContextProvider is configuration-time only - providers are set via BaseAgent.__init__(context_providers=...) and run() does not accept a context_providers parameter.

There are two approaches to achieve per-run skill scoping:

1. Use FilteringSkillsSource with caching disabled
Wrap your skills source with a FilteringSkillsSource whose predicate evaluates runtime context (user, mode, etc.). Make sure to set disable_caching=True on the SkillsProvider so the predicate is re-evaluated on each run rather than serving stale results.

2. Create a context-aware delegating SkillsSource
Implement a custom SkillsSource that delegates skill discovery to other sources and applies context-based filtering. This gives you full control over which underlying sources are consulted and how skills are filtered depending on the current user or operating mode.