Fix memory integrity and introduce project-centered dashboard workflows - #203
Fix memory integrity and introduce project-centered dashboard workflows#203Coding-Dev-Tools wants to merge 13 commits into
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. |
![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: 82357fc02c
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".
| return { | ||
| "id": candidate.id, | ||
| "approved_from": old.id, | ||
| "op": "noop", | ||
| } |
There was a problem hiding this comment.
When a pending memory already has a live approved successor—for example, after a lost response—the retry path returns only the successor ID and op. approve_for_prompt() then reports the reviewer supplied by the retry rather than the reviewer stored in the successor's approval metadata. A retry made under a different reviewer therefore falsely attributes the existing approval while its audit record still names the original reviewer; return the stored reviewer from this branch.
Useful? React with 👍 / 👎.
| if memory_id in (meta.get("supersedes") or []) or meta.get("corrects") == memory_id: | ||
| if (memory_id in (meta.get("supersedes") or []) | ||
| or meta.get("corrects") == memory_id or meta.get("approved_from") == memory_id | ||
| or memory_id in (meta.get("promoted_from") or [])): |
There was a problem hiding this comment.
When caller-controlled metadata contains a non-list promoted_from value, such as 1, this membership test raises TypeError. Because the preceding SQL selects any same-workspace metadata containing the target ID, an otherwise unrelated record like {"note":"<target-id>","promoted_from":1} makes both inspection and the new history endpoint fail for that target. Type-check promoted_from as is already done in predecessors() before testing membership.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 957f52aae0
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".
| if not vector_writes_ready: | ||
| break |
There was a problem hiding this comment.
During a bulk/startup repair, if vector writes are unavailable and the earliest queue entry is an upsert, this break exits the entire repair loop before later deletion entries are inspected. After an external deletion failure, an older upsert can therefore leave a securely erased or quarantined vector searchable until embedding compatibility is restored, even when the provider could now process deletes; skip blocked upserts while continuing cleanup work.
AGENTS.md reference: AGENTS.md:L172-L173
Useful? React with 👍 / 👎.
| def validate_promotion() -> Optional[dict]: | ||
| replay = command.validate() | ||
| if replay is not None: | ||
| return replay |
There was a problem hiding this comment.
An identical retry of a successfully committed session-memory promotion cannot reach this receipt replay after the source session has closed, because the earlier session-status check raises first. This is reproducible by promoting a session memory, ending the session, and repeating the promotion: it raises ValueError despite the committed receipt, so perform replay before applying transient session-state preconditions.
Useful? React with 👍 / 👎.
| def validate_merge() -> Optional[dict]: | ||
| replay = command.validate() | ||
| if replay is not None: | ||
| return replay |
There was a problem hiding this comment.
A retry of a successful session-scoped merge after that session ends fails at the earlier active-session check and never reaches this durable receipt replay. Thus a caller retrying after a lost response receives an error even though the merge committed; check for the existing command result before requiring the session to remain active.
Useful? React with 👍 / 👎.
| if root is None: | ||
| raise MemoryConflict("memory was erased while opening history") | ||
| members = [record for record in self._chain_for(root, wid) | ||
| if (rid is None or record.repo_id == rid) |
There was a problem hiding this comment.
When a repository memory is promoted to workspace scope, its successor intentionally has repo_id=None, so this exact-repository filter removes it whenever history is requested with the source repository. The endpoint consequently reports an incomplete lineage and incorrect total_count, while inspect() returns both records for the same request; authorize the root against repo, then retain visible same-workspace lineage members or apply ancestor-scope semantics.
AGENTS.md reference: AGENTS.md:L174-L178
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: cc3642d4b7
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".
| if _transactional_validator is not None: | ||
| replay = _transactional_validator() | ||
| if replay is not None: | ||
| return replay |
There was a problem hiding this comment.
When a revision/correction has committed but its response is lost, the durable receipt is not checked until this transactional validator runs; remember_with_resolution() has already called self.embedder.embed(...). If the configured embedder is temporarily unavailable on the retry, the identical request raises that backend error instead of returning its committed result, defeating the new recoverable/idempotent revision contract. Check the command receipt before entering the write preparation path, while retaining this in-transaction validation for races.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c5dc4137ce
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".
| root.setAttribute('aria-busy', 'true'); | ||
| status.textContent = 'Loading saved versions…'; | ||
| try { | ||
| const params = new URLSearchParams({ workspace, limit: '50' }); |
There was a problem hiding this comment.
When a project is selected, this request still sends only workspace; MemoryService.memory_history() consequently receives repo=None and follows lineage pointers across every repository in that workspace. For example, a sibling-project record whose metadata points at the inspected memory will appear in the selected project's history, even though the service already supports filtering by repository. Pass the active project into this component and add it to these parameters.
AGENTS.md reference: AGENTS.md:L177-L178
Useful? React with 👍 / 👎.
1 participant
Description
Concurrent governed writes could create multiple successors and delayed sync publication could restore erased vectors. Packing could also discard repository-specific facts, and synthesis could remove exceptions while still reporting a grounded answer. This change makes those transitions authoritative and gives project-memory tasks a consistent dashboard journey.
The original checkout's active graph/layout work stays separate. The implementation and remaining dependency order are recorded in docs/REWORK_EXECUTION.md.
Type
Verification
31da32c06a5ade989608504472e00dfca6f13f94: CodeQL and ci. This includes the NumPy-only Python 3.9 floor, full Python 3.10–3.14 stacks, encryption matrix, coverage, browser, packaging, Docker, type and hygiene gates. The Python 3.12 full suite passed 5,067 tests with 18 optional/platform skips.Migration and remaining release gates
Schema 18 uses the verified pre-migration backup path. Never downgrade a migrated live database in place: restore into a separate fenced location and reconcile later erasures/revocations before serving it. Production restore/cutover, both physical capacity targets, the independent 400-task corpus, human usability and the consented pilot remain open. Paid evaluations require the separate concrete proposal's approval. No merge, release or deployment is included.
Related private processing/recovery work: Coding-Dev-Tools/engraphis-cloud#67. Website release contracts: Coding-Dev-Tools/engraphis.com#12.