pi-cache-optimizer 2.2.0 → 2.3.0

package/README.md

@@ -40,6 +40,19 @@ This release keeps the original DeepSeek behavior and adds read-only stats adapt
 
 Generic OpenAI-compatible proxies are **not** treated as OpenAI-family just because they use an OpenAI-shaped API or provider id. If the active model id/name is ambiguous, the extension hides the footer stats instead of guessing.
 
+## Platform support
+
+This extension is pure Node.js — no shell exec, no native bindings, no platform-specific paths — so it runs on every OS Pi itself supports:
+
+| OS | Notes |
+|---|---|
+| Linux | Native. |
+| macOS | Native. |
+| Windows | Works through the bash shell Pi requires on Windows (Git Bash, Cygwin, MSYS2, or WSL). See Pi's [Windows setup](https://github.com/earendil-works/pi-coding-agent/blob/main/docs/windows.md). |
+| Termux / Android | Works inside Pi's Termux setup. |
+
+State files under `~/.pi/agent/` are resolved via Node's `os.homedir()`, so on Windows the path automatically expands to `C:\Users\<you>\.pi\agent\...`. All shell snippets in this README are bash, matching the shell Pi runs in on every supported platform; no PowerShell or `cmd.exe` translation is needed when commands are executed inside (or for) Pi.
+
 ## Quickstart
 
 1. (Optional but recommended) Read the official Pi + DeepSeek onboarding guide: [`pi_mono.zh-CN.md`](https://github.com/deepseek-ai/awesome-deepseek-agent/blob/main/docs/pi_mono.zh-CN.md). It covers Pi installation and core configuration.

package/README.zh-CN.md

@@ -43,6 +43,19 @@
 
 Generic OpenAI-compatible 代理**不会**仅因为使用 OpenAI 形状 API 或 provider id 就被当作 OpenAI-family。如果当前 model id/name 语义不明确，扩展会隐藏底部统计，而不是猜测。
 
+## 平台支持
+
+本扩展是纯 Node.js 实现 —— 不调用 shell、没有原生绑定、不写死平台相关路径 —— 因此与 Pi 自身保持一致，支持以下系统：
+
+| 操作系统 | 说明 |
+|---|---|
+| Linux | 原生支持。 |
+| macOS | 原生支持。 |
+| Windows | 通过 Pi 在 Windows 下要求的 bash shell 运行（Git Bash、Cygwin、MSYS2 或 WSL）。详见 Pi 的 [Windows setup](https://github.com/earendil-works/pi-coding-agent/blob/main/docs/windows.md)。 |
+| Termux / Android | 在 Pi 的 Termux 环境中可用。 |
+
+状态文件 `~/.pi/agent/` 通过 Node 的 `os.homedir()` 解析，所以在 Windows 上会自动展开为 `C:\Users\<你>\.pi\agent\...`。本文档中所有 shell 命令均使用 bash 语法，与 Pi 在每个受支持平台下运行的 shell 一致；只要在 Pi 内（或为 Pi 而执行）运行，就**不需要**改写为 PowerShell 或 `cmd.exe` 形式。
+
 ## 快速开始
 
 1. （可选但推荐）先读一遍官方 Pi + DeepSeek 接入指南：[`pi_mono.zh-CN.md`](https://github.com/deepseek-ai/awesome-deepseek-agent/blob/main/docs/pi_mono.zh-CN.md)。它讲了 Pi 安装与基础配置。

package/index.ts

@@ -1433,6 +1433,40 @@ export default function (pi: ExtensionAPI) {
   });
 
   pi.on("before_agent_start", async (event, _ctx) => {
+    // ────────────────────────────────────────────────────────────────
+    // OpenAI Responses API bypass (codex-responses + responses)
+    //
+    // OpenAI's Responses API endpoints — both the Codex backend
+    // (openai-codex-responses, chatgpt.com) and the public
+    // Responses API (openai-responses, api.openai.com / Copilot) —
+    // have two properties that make client-side prompt reordering
+    // unnecessary and potentially harmful:
+    //
+    //  1. Server-managed caching: both APIs send `prompt_cache_key`
+    //     (= Pi session id) in every request body, so the server
+    //     already maintains a stable cache without prefix ordering.
+    //     Client-side reordering adds no cache benefit.
+    //
+    //  2. Stricter content-safety filtering: the Codex backend in
+    //     particular has a product-level safety filter that flags
+    //     reordered prompts (tool snippets / guidelines lifted above
+    //     the assistant role) as potential prompt-injection, returning
+    //     `content_filter` and blocking tool calls (notably
+    //     `subagent`). The public Responses API shares the same
+    //     filter framework and could behave similarly.
+    //
+    // We therefore skip ALL prompt modifications (churn strip, skill
+    // compression, reorder) for these APIs. Third-party providers
+    // that use openai-completions are unaffected.
+    // ────────────────────────────────────────────────────────────────
+    const model = _ctx.model;
+    if (model) {
+      const api = lower(model.api);
+      if (api === "openai-codex-responses" || api === "openai-responses") {
+        return {};
+      }
+    }
+
     // Step 1: strip per-turn churn from <session-overview>.
     // Removing RECENT COMMITS, Working directory status, and
     // Journal line count makes more of the session-overview stable

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-cache-optimizer",
-  "version": "2.2.0",
+  "version": "2.3.0",
   "description": "Pi extension that improves provider-side KV/prompt cache hit rates (DeepSeek, OpenAI, Claude, Gemini) by reordering the system prompt, requesting long retention, and showing footer cache stats. Renamed from pi-deepseek-cache-optimizer.",
   "keywords": [
     "pi-package",