Skip to content

ISSUES.md

Latest commit

 

History

History
30 lines (23 loc) · 2.03 KB

ISSUES.md

File metadata and controls

30 lines (23 loc) · 2.03 KB
Raw

Issues

Review Scope

  • Updated OpenCode autosave implementation in .opencode/plugins/mempalace-autosave.js, mempalace-autosave-sync.py, mempalace-autosave-mine.sh, .gitignore, and README.md
  • Focused on the previously identified security and performance issues in autosave persistence and sync behavior

Overall Assessment

  • safe to merge

Priority Summary

  1. No unresolved material security or performance issues were found in the current autosave implementation.

By Type

Missing or Weak Tests

  • Existing coverage now includes a Node plugin regression test for sanitized transcript persistence and no-op idle flushes, plus a Python unittest for sync idempotency and stale chunk deletion.
  • Add a permission-focused test for the autosave directory and transcript files where the platform supports chmod semantics.
  • Add a long-session benchmark or regression test to quantify idle flush cost on large transcripts.

Positive Notes

  • Raw tool output is no longer persisted; assistant fallback content is reduced to bounded structural markers such as tool status and patch file names.
  • Autosave directories and transcript files are created with restrictive modes (0700/0600) on platforms that honor them.
  • The plugin now maintains an incremental in-memory session cache from OpenCode events and only hydrates from the server when needed after startup.
  • The sync script skips unchanged chunks and deletes stale chunk IDs, substantially reducing ChromaDB churn versus blind full upserts.
  • .mempalace-autosave/ is ignored by git to reduce accidental publication of local transcript spools.

Unverified Areas

  • End-to-end plugin behavior against a live OpenCode runtime was not exercised here; validation was limited to code inspection plus syntax checks.
  • Very large or very long-lived sessions may still merit profiling because transcript rendering and chunk comparison remain proportional to changed session size at flush time.
  • Filesystem permission hardening is best-effort and may not be enforced uniformly on all platforms or mounts.