Allow for granular choices in existing context to conserve tokens

[cfdude]

Power users understand that every query sends all pieces of context to the API provider burning tokens and raising cost when some of that context is unnecessary for the query being asked. Power users want the ability to turn off/on parts of the context that gets sent to the API provider. Can include:

• Instructions for all modes
• Mode Role
• Mode custom instructions
• .clinerules
• Additional context that is currently not visible normally

If not an off/on type of process, then it would be good to have a generated context that is ready to send on the next request that can be edited or even shortened and potentially saved like a scratch reference that can be easily switched to instead of manually editing the file each time.

Ideally, it would also be good to have some control over rules that define what parts of the context get sent and in what situations. If we're in "Ask" mode, maybe we don't need to send so much context and we would use a 'lite' context queue with reduced output to send.

[veeragoni]

+1 adding ability to add more 'modes' probably solves this.
Essentially a "ask" or "chat" mode that typically acts like a ChatGPT in VSCode with optional MCP tools (search and others). This way, user never to have to leave IDE.
And giving ability to add more mode's might also give option for custom prompt templates (ex: summarize YouTube videos and take that as context into code etc)

[DoiiarX]

I have been hoping to add token management similar to prompt poet for a long time.