[FEATURE] Strategy for improving and optimizing context usage in a specific but frequent situation.

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

One of the largest inhibitors of LLM tools like Claude Code are context size constraints and context poisoning vulnerabilities. A common cause of context poisoning and pollution, is when Claude Code proposes a suggesting that is mostly perfect but has minor mistakes. Often these are not necessarily products of poor prior context that need to be corrected with user rejection feedback; but either the arbitrary decision that is slightly off from the users desire, or the minor momentary mistake condemned by the prior context coinciding with a pattern matching a tapestry of poor input data it was trained on. In these common scenarios, instead of the context maintaining the erroneous suggestion, adding the users rejection feedback, along with the new appropriate suggestion; there should be an "reject with context revision" (I'll leave it up to Anthropic to compose a more UX amenable name) option, where instead the revised version replaces the original suggestion in the context, and the rejection exchange is dropped - for all intended purposes this new suggestion should be identical to if the LLM suggested it in the first place, and the user can accept or reject (or context revise reject) this new one like any other, the context will be sufficient to elegantly continue it's work appropriately, and the AI can go on to live happily ever after. Anyway, thank you for listening to my TED talk - is this a stupid idea or what?

Proposed Solution

Refer to the above.

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

_No response_

Additional Context

_No response_