Make scoped memory writes, retrieval and workspace processing reliable - #201
Conversation
…ng, gist format, schema indexes - Add DeterministicContextPacker with inter-candidate clause redundancy pruning, score-elbow gating, and ContextPackResult NamedTuple API - Vectorize personalized_pagerank with NumPy (np.add.at scatter, vectorized dangling mass, L1 convergence) for graph performance on large stores - Add early-return guard when incidence_memory_ids is empty in recall pipeline - Reduce list_memory_ids candidate limit 12000 to 500 for lower latency - Add 4 schema indexes: entities workspace_created, edge dst_visibility, mem_links a_valid and b_valid for graph/link traversal performance - Add format='gist' to engraphis_recall_context and smart_recall_context for 60-80% token savings with one-line memory summaries - Add diagnostics pruning: strip verbose default-valued fields when diagnostics=False for cleaner MCP responses - Export ContextPackResult, DeterministicContextPacker, pack_context from engraphis.core public API - Add 29 tests for context packer and 2 tests for MCP gist/diagnostics
… format param in skills - Defer idx_mem_links_b_valid creation in Store._apply_schema alongside idx_mem_links_temporal to preserve legacy v5 migration compatibility - Document format parameter for engraphis_recall_context in portable skill reference and MCP_TOOLS.md - Refresh .claude-plugin/skill-assets.sha256 digest
… and add multi-mode slider test tool - Direct cross-encoder reranker config to ~/.engraphis/config.env rather than CWD .env - Soften security disclosure phrasing in CHANGELOG.md - Add multi-mode slider regression harness tools/galaxy_mode_test.js
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: 696aa07c84
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".
…s, and MCP contract normalization
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b48850a9cf
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: 7148b72fe3
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".
1 participant
Description
Repairs cases where distinct evidence disappeared during packing, independent engines accepted duplicate writes, stale indexes missed canonical memories, and dashboard search or temporal filtering disagreed with the engine.
The findings register, compatibility decisions, remaining backlog, migration and rollback instructions, and reproducible evidence are in
docs/RELIABILITY_PROGRAM.mdanddocs/evidence/reliability/. The review includes the original four unmerged feature commits and existing local edits. Historical branches and the stash were reconciled separately; no unique recovery patch was missing.Type
Verification
Four bounded internal reviewers returned their findings; the parent reviewed and integrated the changes.
pr-review-followup.json,operator-policy-review.jsonandbrowser-followup-review.jsonpin the reviewed source bytes and exact validation boundaries.Final GitHub validation at
e577ea33bf18e25df56c9212530d68ba0890602c: all 23 checks passed. The complete CI run passed 4,825 tests with 16 skips on Python 3.12, 3,877 tests with 117 skips on the Python 3.9 floor, and all 112 browser scenarios, with 85.07% coverage. The supported Python 3.10–3.14 jobs, security, packaging, Docker and integration checks are green. Automated review completed on this head without additional findings; the six addressed earlier threads remain open for reviewer confirmation.Local, remote and fetched PR heads agree, and the locally computed merge tree matches GitHub's merge tree. Earlier failed and successful checkpoints remain preserved in the evidence directory rather than being relabeled as final-head results.
Compatibility and rollout
Schema 17 requires the documented backup and recovery path. Its repair-order index is added idempotently when an existing version-17 database reopens. External-index erasure must own its transaction, so callers must finish a manually opened transaction before erasing; this prevents a later rollback from discarding repair debt after provider deletion.
Roll out the compatible cloud policy migration and services before this client. Existing readable uploads remain paused until explicit workspace approval.
The twelve storage/index benchmark cells remain historical measurements at their recorded source hashes. Store and schema changed after measurement; those timings do not describe this final head or prove the full 100,000-memory coding-agent target. Production restore, independent task-quality evidence and paid model evaluations remain separate gates.
This PR does not authorize merge, deployment, credential rotation or paid evaluation. Review threads are retained for reviewer confirmation.