pi-cache-optimizer 2.8.1 → 2.8.2

package/README.md

@@ -55,7 +55,7 @@ Run `/reload` in Pi after install/update/remove so extension hooks refresh.
 
 On Pi 0.79.7 and newer, `pi update` updates Pi itself only. To update installed Pi packages such as this extension, run `pi update --extensions` (packages only) or `pi update --all` (Pi + packages).
 
-This extension is validated against Pi 0.83.0 and remains designed for Pi 0.82+. It uses the existing extension hooks, `getAgentDir()`, and prompt options shared by those versions; it does not depend on Pi 0.83-only APIs such as `ctx.scopedModels` or the bundled TypeBox 1.3 aliases.
+This extension is validated against Pi 0.84.1 and remains designed for Pi 0.82+. It uses the existing extension hooks, `getAgentDir()`, and prompt options shared by those versions; it does not depend on Pi 0.83+ APIs such as `ctx.scopedModels` or the bundled TypeBox 1.3 aliases.
 
 ## Commands
 
@@ -109,7 +109,7 @@ The explicit setting is stored in `pi-cache-optimizer-config.json` under Pi's ag
 
 Third-party `openai-completions` proxies (LiteLLM / OneAPI / NewAPI / OpenRouter-like channels) often route one session across multiple upstream backends. That splits provider-side prompt caches.
 
-Pi 0.81+ also has a built-in `llama.cpp` provider using an OpenAI-shaped transport. Pi 0.82+ core generates a session `prompt_cache_key` for it when cache retention is enabled, so this extension preserves that key and may add the same conservative fallback when missing. The built-in provider's explicit compat fingerprint is excluded from generic proxy routing/session-affinity advice, but a custom or overridden provider that merely reuses the id `llama.cpp` is treated like any other OpenAI-compatible channel. `prompt_cache_retention` remains subject to the normal safety rule: keep it only for official OpenAI or an explicit effective `supportsLongCacheRetention: true` opt-in in `models.json`; otherwise strip it before sending. Effective values follow Pi's precedence: `modelOverrides[modelId].compat` first, then the matching `models[].compat`, then provider-level `compat`. An explicit `false` at a higher layer overrides `true` below it.
+Pi 0.84.1 also fixes built-in Fireworks compatibility for models that reject `prompt_cache_retention`; keep the extension's diagnostics focused on the effective model compat exposed by Pi rather than adding a provider-name special case. Pi 0.81+ also has a built-in `llama.cpp` provider using an OpenAI-shaped transport. Pi 0.82+ core generates a session `prompt_cache_key` for it when cache retention is enabled, so this extension preserves that key and may add the same conservative fallback when missing. The built-in provider's explicit compat fingerprint is excluded from generic proxy routing/session-affinity advice, but a custom or overridden provider that merely reuses the id `llama.cpp` is treated like any other OpenAI-compatible channel. `prompt_cache_retention` remains subject to the normal safety rule: keep it only for official OpenAI or an explicit effective `supportsLongCacheRetention: true` opt-in in `models.json`; otherwise strip it before sending. Effective values follow Pi's precedence: `modelOverrides[modelId].compat` first, then the matching `models[].compat`, then provider-level `compat`. An explicit `false` at a higher layer overrides `true` below it.
 
 For real proxies, start with session affinity:
 
@@ -150,7 +150,7 @@ Some proxies rewrite or insert hidden 5-minute breakpoints after Pi's request ho
 
 ## Adaptive thinking models
 
-Claude models from opus-4.6 / sonnet-4.6 (including Opus 5 and Sonnet 5) / fable-5 onwards require `forceAdaptiveThinking: true` in compat. Kimi Coding K3 (`k3`) and `kimi-for-coding` also use adaptive thinking and need `allowEmptySignature: true` so replayed empty-signature thinking blocks remain valid. Without the required compat, Pi may send a legacy thinking payload or replay thinking incorrectly. Pi 0.83.0's native Opus 5 catalog is covered by the same adaptive-thinking detection; custom `anthropic-messages` channels still need the compat flag when Pi does not provide it.
+Claude models from opus-4.6 / sonnet-4.6 (including Opus 5 and Sonnet 5) / fable-5 onwards require `forceAdaptiveThinking: true` in compat. Kimi Coding K3 (`k3`) and `kimi-for-coding` also use adaptive thinking and need `allowEmptySignature: true` so replayed empty-signature thinking blocks remain valid. Without the required compat, Pi may send a legacy thinking payload or replay thinking incorrectly. Pi 0.83+ native Opus 5 catalogs are covered by the same adaptive-thinking detection; custom `anthropic-messages` channels still need the compat flag when Pi does not provide it.
 
 Pi's built-in catalog already sets this flag for official models. Custom channels in `models.json` that override these models must include the flag:
 
@@ -283,10 +283,10 @@ Pi 0.79+ also includes a built-in footer `CH` marker for the latest prompt cache
 Example footer:
 
 ```text
-· OpenAI cache 3/10 · 0.002M/0.005M tok (40%) ⚠️ compat
+· OpenAI cache 3/10·0.002M/0.005M 40.0% ⚠️ compat
 ```
 
-The leading `· ` is owned by this extension and separates its status from statuses published by other extensions in the same footer. It is present for normal, disabled, router-restored, and warning-suffixed statuses. The stats format remains `<label> <hit requests>/<total requests> · <cached input tokens>/<total input tokens> tok (<token hit rate>)`; `/cache-optimizer stats` output is unchanged. Some adapters may also append `· write <tokens> tok`, and runtime diagnostics may append `⚠️ compat` or `⚠️ integrity`.
+The leading `· ` is owned by this extension and separates its status from statuses published by other extensions in the same footer. It is present for normal, disabled, router-restored, and warning-suffixed statuses. The compact footer format is `<label> <hit requests>/<total requests>·<cached input tokens>/<total input tokens> <token hit rate>`; token hit rate keeps one decimal place and the footer omits the redundant `tok` suffix. `/cache-optimizer stats` output is unchanged. Some adapters may also append `·write <tokens>`, and runtime diagnostics may append `⚠️ compat` or `⚠️ integrity`.
 
 Supported footer labels include: DS, Claude, OpenAI, Gemini, Kimi, Qwen, GLM, MiniMax, Mimo, Hunyuan, Mistral, Grok, Llama, Nemotron, Cohere, Yi, Doubao, ERNIE, Baichuan, StepFun, Spark, InternLM, Gemma, Phi, Jamba, Solar, Sonar, Nova, Reka, Falcon, DBRX, MPT, StableLM, Aquila, EXAONE, HyperCLOVA, Luminous, Hermes, Granite, Arctic, Pangu, SenseNova, Zhinao, MiniCPM, XVERSE, Orion, OpenChat, Vicuna, Wizard, Zephyr, Dolphin, OpenOrca, Starling, BLOOM, RWKV, and Aya.
 

package/README.zh-CN.md

@@ -55,7 +55,7 @@ pi remove npm:pi-deepseek-cache-optimizer && pi install npm:pi-cache-optimizer
 
 Pi 0.79.7 及之后，`pi update` 默认只更新 Pi 本体。若要更新已安装的 Pi package（包括本扩展），请运行 `pi update --extensions`（只更新 packages）或 `pi update --all`（Pi 与 packages 一起更新）。
 
-本扩展已使用 Pi 0.83.0 验证，并继续面向 Pi 0.82+ 设计。它使用这些版本共有的 extension hooks、`getAgentDir()` 和 prompt options，不依赖 Pi 0.83 专有 API（例如 `ctx.scopedModels` 或 bundled TypeBox 1.3 aliases）。
+本扩展已使用 Pi 0.84.1 验证，并继续面向 Pi 0.82+ 设计。它使用这些版本共有的 extension hooks、`getAgentDir()` 和 prompt options，不依赖 Pi 0.83+ 专有 API（例如 `ctx.scopedModels` 或 bundled TypeBox 1.3 aliases）。
 
 ## 命令
 
@@ -108,7 +108,7 @@ Pi 0.79.7 及之后，`pi update` 默认只更新 Pi 本体。若要更新已安
 
 LiteLLM / OneAPI / NewAPI / 类 OpenRouter 渠道等第三方 `openai-completions` 代理，常会把同一个 session 分散到多个上游后端，导致 provider 侧 prompt cache 被拆散。
 
-Pi 0.81+ 也内置了使用 OpenAI-shaped transport 的 `llama.cpp` provider。Pi 0.82+ core 在启用 cache retention 时会为它生成 session `prompt_cache_key`，因此本扩展会保留该 key，并在缺失时使用同样的保守 fallback。只有符合 Pi 内置 provider 明确 compat 指纹的模型会跳过通用 proxy 路由 / session-affinity 建议；仅复用 `llama.cpp` id 的自定义或覆盖 provider 仍按普通 OpenAI-compatible 渠道处理。`prompt_cache_retention` 继续遵循统一安全规则：仅官方 OpenAI 或 `models.json` 中有效配置为 `supportsLongCacheRetention: true` 时保留，否则发送前移除。有效值遵循 Pi 的优先级：先看 `modelOverrides[modelId].compat`，再看匹配的 `models[].compat`，最后看 provider 级 `compat`；高层显式 `false` 会覆盖低层的 `true`。
+Pi 0.84.1 还修复了内置 Fireworks 渠道对拒绝 `prompt_cache_retention` 的模型兼容性；本扩展继续依据 Pi 暴露的有效 model compat 做诊断，不按 provider 名称增加特殊分支。Pi 0.81+ 也内置了使用 OpenAI-shaped transport 的 `llama.cpp` provider。Pi 0.82+ core 在启用 cache retention 时会为它生成 session `prompt_cache_key`，因此本扩展会保留该 key，并在缺失时使用同样的保守 fallback。只有符合 Pi 内置 provider 明确 compat 指纹的模型会跳过通用 proxy 路由 / session-affinity 建议；仅复用 `llama.cpp` id 的自定义或覆盖 provider 仍按普通 OpenAI-compatible 渠道处理。`prompt_cache_retention` 继续遵循统一安全规则：仅官方 OpenAI 或 `models.json` 中有效配置为 `supportsLongCacheRetention: true` 时保留，否则发送前移除。有效值遵循 Pi 的优先级：先看 `modelOverrides[modelId].compat`，再看匹配的 `models[].compat`，最后看 provider 级 `compat`；高层显式 `false` 会覆盖低层的 `true`。
 
 对真正的代理，建议先启用 session affinity：
 
@@ -149,7 +149,7 @@ Anthropic 按 `tools → system → messages` 顺序处理 cache breakpoint，
 
 ## Adaptive thinking 模型
 
-Claude 从 opus-4.6 / sonnet-4.6（含 Opus 5、Sonnet 5）/ fable-5 开始需要在 compat 中设置 `forceAdaptiveThinking: true`。Kimi Coding K3（`k3`）和 `kimi-for-coding` 也使用 adaptive thinking，并需要 `allowEmptySignature: true`，以正确重放空 signature 的 thinking block。缺少这些 compat 时，Pi 可能发送旧版 thinking payload 或错误重放 thinking。Pi 0.83.0 的原生 Opus 5 catalog 已覆盖在同一 adaptive-thinking 检测中；如果自定义 `anthropic-messages` 渠道没有继承该 compat，仍需手动设置。
+Claude 从 opus-4.6 / sonnet-4.6（含 Opus 5、Sonnet 5）/ fable-5 开始需要在 compat 中设置 `forceAdaptiveThinking: true`。Kimi Coding K3（`k3`）和 `kimi-for-coding` 也使用 adaptive thinking，并需要 `allowEmptySignature: true`，以正确重放空 signature 的 thinking block。缺少这些 compat 时，Pi 可能发送旧版 thinking payload 或错误重放 thinking。Pi 0.83+ 的原生 Opus 5 catalog 已覆盖在同一 adaptive-thinking 检测中；如果自定义 `anthropic-messages` 渠道没有继承该 compat，仍需手动设置。
 
 Pi 内置 catalog 已为官方模型设置此 flag。`models.json` 中覆盖这些模型的自定义渠道必须包含该 flag：
 
@@ -282,10 +282,10 @@ Pi 0.79+ 已内置 footer `CH` 标记，用于显示最近一次 prompt cache hi
 示例 footer：
 
 ```text
-· OpenAI cache 3/10 · 0.002M/0.005M tok (40%) ⚠️ compat
+· OpenAI cache 3/10·0.002M/0.005M 40.0% ⚠️ compat
 ```
 
-开头的 `· ` 由本扩展负责，用于把本扩展状态与同一 footer 中其他扩展发布的状态隔开。普通、disabled、router 恢复以及带 warning 的状态都会保留此前缀。统计格式仍是 `<label> <命中请求数>/<总请求数> · <cached input tokens>/<total input tokens> tok (<token 命中率>)`；`/cache-optimizer stats` 的输出不变。部分 adapter 还可能追加 `· write <tokens> tok`，运行时诊断可能追加 `⚠️ compat` 或 `⚠️ integrity`。
+开头的 `· ` 由本扩展负责，用于把本扩展状态与同一 footer 中其他扩展发布的状态隔开。普通、disabled、router 恢复以及带 warning 的状态都会保留此前缀。紧凑 footer 格式为 `<label> <命中请求数>/<总请求数>·<cached input tokens>/<total input tokens> <token 命中率>`；token 命中率保留一位小数，并去掉多余的 `tok` 后缀。`/cache-optimizer stats` 的输出不变。部分 adapter 还可能追加 `·write <tokens>`，运行时诊断可能追加 `⚠️ compat` 或 `⚠️ integrity`。
 
 支持的 footer label 包括：DS、Claude、OpenAI、Gemini、Kimi、Qwen、GLM、MiniMax、Mimo、Hunyuan、Mistral、Grok、Llama、Nemotron、Cohere、Yi、Doubao、ERNIE、Baichuan、StepFun、Spark、InternLM、Gemma、Phi、Jamba、Solar、Sonar、Nova、Reka、Falcon、DBRX、MPT、StableLM、Aquila、EXAONE、HyperCLOVA、Luminous、Hermes、Granite、Arctic、Pangu、SenseNova、Zhinao、MiniCPM、XVERSE、Orion、OpenChat、Vicuna、Wizard、Zephyr、Dolphin、OpenOrca、Starling、BLOOM、RWKV、Aya。
 

package/index.ts

@@ -3733,13 +3733,13 @@ function formatTokenCount(value: number): string {
 
 function formatCacheStats(adapter: CacheProviderAdapter, stats: CacheStats): string {
   const percent = stats.totalInputTokens > 0
-    ? ` (${Math.round((stats.cachedInputTokens / stats.totalInputTokens) * 100)}%)`
-    : "";
+    ? (stats.cachedInputTokens / stats.totalInputTokens) * 100
+    : 0;
   const writeText = adapter.showCacheWrite && stats.cacheWriteInputTokens > 0
-    ? ` · write ${formatTokenCount(stats.cacheWriteInputTokens)} tok`
+    ? `·write ${formatTokenCount(stats.cacheWriteInputTokens)}`
     : "";
 
-  return `${adapter.label} ${stats.hitRequests}/${stats.totalRequests} · ${formatTokenCount(stats.cachedInputTokens)}/${formatTokenCount(stats.totalInputTokens)} tok${percent}${writeText}`;
+  return `${adapter.label} ${stats.hitRequests}/${stats.totalRequests}·${formatTokenCount(stats.cachedInputTokens)}/${formatTokenCount(stats.totalInputTokens)} ${percent.toFixed(1)}%${writeText}`;
 }
 
 function prefixFooterStatus(statusText: string | undefined): string | undefined {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-cache-optimizer",
-  "version": "2.8.1",
+  "version": "2.8.2",
   "description": "Improve Pi prompt/KV cache hit rates with stable prompts, OpenAI-compatible cache keys, proxy compat warnings, and footer cache stats.",
   "keywords": [
     "pi-package",
@@ -35,6 +35,7 @@
     "check": "npm run typecheck && npm test && npm run check:diff && npm run check:pack"
   },
   "devDependencies": {
+    "@earendil-works/pi-coding-agent": "^0.84.1",
     "@types/node": "^22.0.0",
     "jiti": "^2.7.0",
     "typescript": "^5.8.3"