@rahularya01/pi-cursor 1.4.29 → 1.4.30
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +9 -0
- package/dist/index.js +32 -32
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,14 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## [1.4.30] - 2026-09-02
|
|
4
|
+
|
|
5
|
+
### Fixed
|
|
6
|
+
|
|
7
|
+
- **Cursor 3.18.9 additive `agent.v1` envelope fields no longer report as wire-drift.** `InteractionUpdate.message_started_at_ms`, `ExecServerMessage.accept_hook_additional_contexts`, `ConversationStateStructure` start timestamp/timezone, and `AgentServerMessage.ttft_breakdown` are decoded so `/cursor.doctor` and Connect `internal` errors stop attaching this inventory. Fixes [#16](https://github.com/Rahularya01/pi-cursor/issues/16).
|
|
8
|
+
- **Tool-schema slimming no longer drops user-defined properties named `description`, `title`, or `default`.** Annotation keys are stripped only on schema nodes; named maps (`properties`, `$defs`, …) keep their keys. Prevents invalid schemas (dangling `required`) that Gemini rejected as `resource_exhausted`.
|
|
9
|
+
- **GetUsableModels `cursor-*` rows no longer keep a guessed 200K window when Cursor already published `contextTokenLimit`.** Parameterized discovery lists `grok-4.6` at 256K, but Pi also registers `cursor-grok-4.6` from GetUsableModels with `inferCursorContextWindow` (default 200K). Different ids meant both survived catalog merge, so compaction used the 200K alias. The parameterized window is now copied onto the `cursor-` prefixed twin. `inferCursorContextWindow` also treats Grok 4.5/4.6 as 256K when metadata is missing.
|
|
10
|
+
- **Cursor GPT-5.6 Luna/Sol/Terra no longer advertise a 1M window that OpenAI rejects at 500k.** Live `GetUsableModels` labels the 272k default "GPT-5.6 Luna 1M High", so Pi inferred `contextWindow: 1000000` and skipped auto-compaction until ~983k. OpenAI (via Cursor) then 400'd: `This model's maximum prompt length is 500000 but the request contains 503167 tokens.` Display-name "1M" is ignored unless the id has `-1m`; 1m variants are capped at 500k so threshold compaction fires first. Claude 1M rows are unchanged.
|
|
11
|
+
|
|
3
12
|
## [1.4.29] - 2026-08-30
|
|
4
13
|
|
|
5
14
|
### Changed
|