Skip to content

Open Pstack 1.11.0 incorporates Cursor 0.15.9

Choose a tag to compare

View all tags
@arjitj2 arjitj2 released this 04 Oct 01:04
v1.11.0
1ad7ccc
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Incorporates Cursor Pstack 0.15.9 at e43c7ee26e0038c6c1fa8380dd34ce86ff94cb2a, including the pending 0.15.6–0.15.8 updates.

Changes

  • Add manual /correct to find repeated mistakes in repository history and prevent them through structural enforcement.
  • Screen designs for split ownership, duplicate APIs, importable internals, and hand-maintained duplicate lists.
  • Try performance improvements in order, beginning with removing unused work. Hillclimb keeps its separate target and iteration floor.
  • Add benchmark and numerical-evidence guidance, fresh workers for new tasks, hourly audit guidance, PR headings, and schema-first boundary validation.

The shared Claude Code/Codex tree, saved model choices, provider dispatch, and recovery policies remain intact. Adaptations and exclusions identify intentional differences from Cursor.

Validation

PR #111 records passing CI, 766 Bun tests, strict typechecks, static/plugin/documentation/maintenance checks, and independent exact-head review. The 216-file package was exercised in real Codex CLI 0.159.2 with GPT-6.1 Sol and Claude Code 2.1.285 with Opus 5.5. Live fixtures covered skill entry, benchmark and numerical-evidence rejection, schema validation, three fresh native workers, workflow decisions, historical mistake correction, architecture screening, and performance stop semantics. Reintroducing historical mistakes failed the correction checks; restoration passed.

The published package tree matches the tested candidate. Copilot was quota-limited and provided no review. The read-only design/performance fixtures establish guidance, not a measured performance gain or a complete architecture panel. Scheduler execution, remote PR operations, and provider recovery were not changed or revalidated by these fixtures.

Codex CLI 0.158.0 rejected GPT-6.1 Sol before inference; the app-bundled 0.159.2 accepted the same model and login. No model downgrade was required.

Update and rollback

Follow the update instructions. main remains the installation source. This release does not automatically change existing installations or model assignments. Previous tested checkpoint: v1.10.3.

Assets 2
Loading