feat: add support for skills

[twishabansal]

Does not yet contain support for multiple sub-toolsets for cloudsql postgres

[averikitsch]

Have you tested with the settings in the .env and having the extension load in the env vars? Does this allow the agents to use these variables without additional input from the user?

[twishabansal]

Haoyu has edited his implementation to update the scripts to get the env vars from the global extension path. It is automatically set when the user runs gemini extensions install.

[averikitsch]

For the first iteration, we will have to manually add the descriptions since Grouping (with the descriptions) most likely wont be complete.

[twishabansal]

Do you mean grouping will be complete but without the descriptions? For now, I will add the descriptions manually.

[helloeve]

Thank you Twisha for authoring this PR! After looking at the generated config, I realized there should be some improvements done to the skill generation logic we previously introduced.

• As Averi pointed out, the env variable placeholder for defining the sources are not propagated as the current generation logic will rely on the parsed version of the environment variable rather than leaving it as a placeholder.
• The generated skill.md is too large and I think we may want to move the tool-specific parameter definition into its own file under references/ to take advantages from skill's dynamic loading capability.