Repository navigation
Sonny: memory reinforces what the person confirms - #431
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013jkdGGVwTdQTtcHeKUqRa6
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d16ee49c69
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d4c106f477
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0117027dc0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 805d1eb4c5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 133ca5db0c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 06942bcc1b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Facts carry confirmed and pushed_back counts, shown in the index. The nightly said gives the person's reactions with the reply each answers, counted only while still on, and ask answers with what the ask recommended. The cleanup saves what worked, never agreement. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013jkdGGVwTdQTtcHeKUqRa6
ceec538 to
70284fe
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 70284fe872
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013jkdGGVwTdQTtcHeKUqRa6
Requested by George · project thread
This is the second half of the memory-first audit (write-up, first half #410). It adds a feedback loop: the person's praise and pushback reach memory and carry weight with later agents. George chose to ship it in 0.7.0. The coach thread's eval of #403 runs at its own pinned commit, so this PR doesn't touch it.
What changes
confirmedandpushed_backin its front matter, written only when not zero. The index line shows them, for example- split-reviews (feedback, verified 2026-10-30, confirmed 3, pushed back 1): ..., so a later agent sees a preference's weight without opening it. This follows ExpeL's insight counts and ACE's helpful/harmful counters.savetakesconfirmed/pushed_back; the script takes--confirmed N/--pushed-back N.memory_invalid), the same as a badexternal.saidsees feedback. Until now it dropped reactions entirely. It now includes:reacted_to: the first line of the reply it answers. A reaction to the bot's last reply reached the bot as a bare emoji, sosaidpairs it with the reply before it in history. Feedback stays tied to what it was about, as ChatGPT Pulse keeps feedback with its card.ask, withrecommended: the label of the option the ask recommended, taken from theaskcall.confirmedby the times seen. Pushback, or a pick against the recommendation, raisespushed_back. It removes a fact pushed back on more than it was confirmed, as ExpeL drops an insight at zero. Picking against the recommendation on one kind of question again and again is saved as a preference.Checks
memory::tests::a_facts_counts_show_in_the_index_and_outlast_a_save_without_themcovers file and index rendering, keeping counts across a save without them, setting zero, script flags (once each, whole numbers,saveonly) and an invalid count.the_cleanup_reads_the_persons_reactions_and_picks_with_what_each_answerscovers:cargo test --workspace,cargo fmt --checkandcargo clippy --workspace --all-targets -D warningspass locally.Not evaluated yet
The coach thread's two-day eval measures #403 alone. This loop needs its own run: on day 1 the scripted person praises one approach and pushes back on another, and on day 2 a fresh lead should repeat the first and avoid the second, without its feedback turning softer.
Generated by Claude Code