Repository navigation
fix(memory): use agent recall feedback during project curation - #172
Open
ScriptedAlchemy wants to merge 1 commit into
Open
ScriptedAlchemy wants to merge 1 commit into
ScriptedAlchemy wants to merge 1 commit into
Conversation
|
ScriptedAlchemy
enabled auto-merge
November 20, 2026 04:38
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Agents could upvote or downvote a recalled fact, but the curator never saw those votes. A valid fact recalled for the wrong task also had no accurate rating.
Agents can now mark a recall as irrelevant without lowering trust. The recall includes the fact ID and revision required by the feedback tool. During each project’s scheduled review, the curator receives the fact’s trust and vote counts since its last edit. Ratings help it identify claims to check. Retirement still requires evidence and respects the existing protections for owner-maintained memories.
Verification included 102 changed-file tests and nine additional MCP and provider tests. All 21 real-model curation samples passed, including retaining a durable rule with low trust and repeated negative or off-task feedback.
In Kimi K3.1 tool-use replays, both helpful votes and both irrelevant votes were recorded. The model submitted four of five expected downvotes. In the remaining replay it identified the stale fact in its final reply and skipped the tool call.
Implemented with GPT-6 in Codex.