Repository navigation
Restore a single memory change from its journal - #822
Conversation
Apply the inverse of a matching add, remove or replace while refusing drifted entries.
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. |
![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: b4240aa6db
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".
…topic cap A restore wrote a line from the change log back verbatim: a value saved as a secret since would return in the clear, a fact remembered again under another topic or new metadata would duplicate, and a removed topic could be rebuilt past the 40-topic cap. The line going back now passes the checks a new entry does, and the scope-wide normalized dedup and topic cap match addEntry.
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 937a435c16
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".
A replace whose before-text was remembered again under another topic or new metadata still restored it beside the new copy. The restore now runs the same scope-wide normalized-text dedup as a remove, skipping only the matched after entry itself — an after that squashes to the before text is no duplicate.
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9790a1daee
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".
A forged remove row with removedTopic and a file like ../../team/MEMORY.md rebuilt its topic at a joined path, writing a forged entry over another scope\u0027s file while inside() still passed — the target stays inside the memory root. Every file a restore names is now checked before anything is built or published: only MEMORY.md or a plain topic file name is a change that can be restored.
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 46c051c699
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".
A filtered split renumbered the rows behind every blank line, so a restore could undo the wrong change; the index now counts physical lines and a blank one is refused. A before-line going back is also held to a single non-empty bullet, so a forged heading or multi-line value cannot corrupt the file or hide entries from recall. And flatFile only checks the name is a flat basename now, so hand-named topic files like Travel.md restore again.
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 64be9492ed
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".
… characters A change-log row's before line is the whole entry line, metadata included. The secret check ran only on the entry text, so a value that became a saved secret after the row was written could still be put back from the line's metadata, to the file and to the restore's own row. The whole line now goes through secretEntry on both the remove and the replace path. A file name with a control character passed the flat-name check, and the write it led to failed after the pending record was accepted, leaving every later memory change stuck on a pending record that cannot finish. Flat names now refuse control characters. The bullet prefix was stripped with ^[-*] while the entry-line check and parseMemory accept an indented bullet, so a before line with leading spaces kept its bullet into the text check and the duplicate check missed a fact already saved; the restore then wrote a duplicate. The same extraction now strips leading whitespace everywhere a row's line becomes text, including the reported text.
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e0c425f00b
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".
… a secret name On a file system that folds case, the file a removed topic's row names can still exist under another case; the case-sensitive lookup missed it and a fresh file written under the row's name replaced it, losing what was saved since. When the named path reports existing, the loaded variant whose name folds to it is used instead, and the change log records the name of the file actually written so a restore of that row finds the line again. A case-sensitive file system keeps the names apart and rebuilds as before. A removed topic is rebuilt with its name written into the file's heading and the Index link. If the name became a saved secret since the row was written, the restore would put it back in the clear even when the entry itself held nothing sensitive. The rebuild now checks the topic name with the same secretEntry check an entry gets and refuses.
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c5a6ecb98f
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: c5a6ecb98f
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".
… secrets The temporary name writeText makes beside a file carried the file's whole path, so a legal topic file name near the system's per-name limit made a temporary name over it. The write then failed after the pending record had accepted the change, and every later memory change stopped on a pending record that cannot finish. The temporary name is now a short name in the same folder — a piece of the file's name, the pid, and a uuid — so it stays within the limit however long the file's name is, cannot collide with a parallel write, and still moves over the file in place. Undoing an add or a replace copies the line the row wrote into the restore's own row as its before, journaled in the clear along with the pending record. When the value on that line became a saved secret since the row was written, the restore put it back into the change log. The whole after-line now faces the same secretEntry check the line going back already did, on both paths.
|
@codex review |
|
Codex Review: Didn't find any major issues. More of your lovely PRs please. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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". |
# Conflicts: # ds-bot/src/host/memory.js
The header line kept the summary's update time on a page that is not the summary; the history page keeps the space empty like the dialog makes it.
Issue
Fixes #464
What changed / 改了什么
memory-restoreendpoint, selecting a journal row by its 1-based line number (plusbotIdfor a private Bot's journal).restored: <original id>, enforcecheckSize(), and document the design indocs/plans/2026-11-01-memory-restore.md.Changelog
A single memory change can now be safely undone from its journal without hand-editing Markdown files.
How I checked it / 怎么验证的
npm test(inds-bot/) — 1,436 passed, 2 skipped; includes restore, redo, drift, topic cleanup, and range errorsnode scripts/build-client.mjs --checknpm run lint— existing warnings onlysetsidor DSH checkout; journal writes and recovery are unit-testedChecklist / 检查
src/client/changed →npm run buildran andclient.jsis committed. (No client source changed.)Co-authored-by.