billion-context 0.1.117 → 0.1.118

package/README.md

@@ -320,6 +320,18 @@ Then send one message from your client and watch the log
 should see a `processTurn` line per request, and once the conversation grows,
 `[acp-usage] round N input=X cached=Y (cache hit Z%)` + a `compress` event.
 
+### Client uses `http.proxy` (CONNECT) but nothing compresses
+
+Some clients (VS Code-based IDEs: CodeBuddy, Cursor, Windsurf, …) only offer an HTTP **proxy** setting (`http.proxy`, `codingcopilot.httpProxyURL`, …) — no model base-URL to rewrite. Such clients send `CONNECT <model-host>:443` through the proxy instead of plain `/bili/…` requests. That path is only decrypted when the model host is on bili's **MITM whitelist**; otherwise bili blind-tunnels the TLS bytes (opaque relay) and can never see — or compress — the model requests (#897).
+
+This failure mode is now loud instead of silent:
+
+- a one-time `BLIND TUNNEL WARNING` per target host in the log, with the fix steps;
+- `blindTunnels` (count + exact target hosts) in `curl -s http://localhost:8787/__bili/health` and `/__bili/stats` (loopback-only);
+- an `UNDECRYPTED TRAFFIC (instance-level)` section in `acp_status` output while such tunnels exist.
+
+To actually compress such a client: add its model domain to `"mitm".domains` in `billion-context.json` (e.g. `"mitm": { "domains": ["copilot.tencent.com"] }`) or via `BILI_MITM_DOMAINS`, restart bili, and make the client trust bili's root CA (`NODE_EXTRA_CA_CERTS=~/.local/share/billion-context/ca/root-ca.pem` for Node-based clients, or the client's own CA-path setting). The `/bili/` prefix trick does not apply here — there is no URL to change. Details: [CONFIGURATION.md → MITM](CONFIGURATION.md#mitm-transparent-proxy-login-clients).
+
 ## Running the proxy
 
 ### Flags

package/README.zh-CN.md

@@ -159,6 +159,18 @@ curl -s http://localhost:8787/__bili/stats
 同时也打到 stderr)。每个请求应该看到一行 `processTurn`,等对话变长后
 会出现 `[acp-usage] round N input=X cached=Y (cache hit Z%)` + `compress` 事件。
 
+### 客户端用 `http.proxy`(CONNECT)接入但从不压缩
+
+部分客户端(VS Code 系 IDE:CodeBuddy、Cursor、Windsurf……)只提供一个 HTTP **代理**设置(`http.proxy`、`codingcopilot.httpProxyURL` 等),没有可改写的模型 base-URL。这类客户端不走普通的 `/bili/…` 请求,而是把 `CONNECT <模型域名>:443` 发给代理。只有当模型域名在 bili 的 **MITM 白名单**里时这条路径才会被解密;否则 bili 只做盲隧道(不透明转发),永远看不到——也就无法压缩——模型请求(#897)。
+
+该失效模式现在不再静默:
+
+- 日志里对每个目标域名打一次 `BLIND TUNNEL WARNING`,附修复步骤;
+- `curl -s http://localhost:8787/__bili/health` 与 `/__bili/stats` 输出 `blindTunnels`(计数 + 精确目标域名,仅 loopback);
+- 存在此类隧道时,`acp_status` 输出会多一节 `UNDECRYPTED TRAFFIC (instance-level)`。
+
+要真正压缩这类客户端:把它的模型域名加进 `billion-context.json` 的 `"mitm".domains`(如 `"mitm": { "domains": ["copilot.tencent.com"] }`)或环境变量 `BILI_MITM_DOMAINS`,重启 bili,并让客户端信任 bili 的根 CA(Node 系客户端用 `NODE_EXTRA_CA_CERTS=~/.local/share/billion-context/ca/root-ca.pem`,有 CA 路径设置的用其设置)。`/bili/` 前缀方案在这里不适用——没有 URL 可改。详见 [CONFIGURATION.zh-CN.md → MITM](CONFIGURATION.zh-CN.md#mitm-透明代理登录客户端)。
+
 ## 运行代理
 
 ### 命令行参数

package/dist/agent/opencode.js

@@ -25599,7 +25599,7 @@ ${captureLines}` : capture.stack;
   }
 });
 
-// node_modules/acp-kernel/dist/chunk-QPVKTDYF.js
+// node_modules/acp-kernel/dist/chunk-VPRKYTHW.js
 import { createRequire } from "module";
 var require2 = createRequire(import.meta.url);
 function thinkingTokenValue(thinking) {
@@ -25726,6 +25726,7 @@ var COMPRESS_TOOL_NAME = "compress";
 var DECOMPRESS_TOOL_NAME = "decompress";
 var SEARCH_CONTEXT_TOOL_NAME = "search_context";
 var ACP_STATUS_TOOL_NAME = "acp_status";
+var ACP_CACHE_TOOL_NAME = "acp_cache";
 var ABSORB_TOOL_NAME = "absorb";
 var ACP_TEXT_OPEN = "<acp_compress>";
 var ACP_TEXT_CLOSE = "</acp_compress>";
@@ -25829,10 +25830,15 @@ var COMPRESS_TOOL_OPENAI = {
   }
 };
 var TEXT_PROMPT_SECTIONS = [
-  ["acpTags", `ACP TAGS
+  [
+    "acpTags",
+    `ACP TAGS
 
-Each message in the conversation is annotated with a <acp tokens="2.1K" type="tool:bash">m00175</acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`],
-  ["textProtocol", `COMPRESSION PROTOCOL (TEXT)
+Each message in the conversation is annotated with a <acp tokens="2.1K" type="tool:bash">m00175</acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`
+  ],
+  [
+    "textProtocol",
+    `COMPRESSION PROTOCOL (TEXT)
 
 You manage context by emitting a special trigger in your text output. When you decide a range of conversation is genuinely consumed and should be compressed into a summary, output EXACTLY this marker (the proxy intercepts and executes it; the marker is stripped from what the user sees):
 
@@ -25843,8 +25849,11 @@ Rules for the trigger:
 - JSON shape matches the compress tool: {"content":[{startId,endId,summary,topic?}]}. Batch multiple ranges in one trigger.
 - After emitting the marker, STOP your turn. Do not continue with other text \u2014 the proxy will execute the compression and return the result, then you continue fresh.
 - Do NOT wrap the marker in code fences, quotes, or commentary.
-- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`],
-  ["textTools", `ACP TOOLS (TEXT TRIGGERS)
+- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`
+  ],
+  [
+    "textTools",
+    `ACP TOOLS (TEXT TRIGGERS)
 
 Since host tools cannot coexist with a declared tools field, ALL ACP tools use text triggers. Emit the marker; the proxy intercepts and executes it; the marker is stripped from what the user sees.
 
@@ -25864,13 +25873,19 @@ Since host tools cannot coexist with a declared tools field, ALL ACP tools use t
 Rules for ALL triggers:
 - Output on its own, NO surrounding prose. Just the raw marker.
 - After emitting, STOP your turn. The proxy executes and returns the result.
-- Do NOT wrap in code fences, quotes, or commentary.`]
+- Do NOT wrap in code fences, quotes, or commentary.`
+  ]
 ];
 var HYBRID_PROMPT_SECTIONS = [
-  ["acpTags", `ACP TAGS
+  [
+    "acpTags",
+    `ACP TAGS
 
-Each message in the conversation is annotated with a <acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`],
-  ["textProtocol", `COMPRESSION PROTOCOL (TEXT)
+Each message in the conversation is annotated with a <acp> tag showing its reference ID, approximate token size, and content type. These tags are system metadata. NEVER echo these history tags. Use only the ref ID (e.g. m00005), never the XML wrapper.`
+  ],
+  [
+    "textProtocol",
+    `COMPRESSION PROTOCOL (TEXT)
 
 You manage context by emitting a special trigger in your text output. When you decide a range of conversation is genuinely consumed and should be compressed into a summary, output EXACTLY this marker (the proxy intercepts and executes it; the marker is stripped from what the user sees):
 
@@ -25881,8 +25896,11 @@ Rules for the trigger:
 - JSON shape: {"content":[{startId,endId,summary,topic?}]}. Batch multiple ranges in one trigger.
 - After emitting the marker, STOP your turn. Do not continue with other text \u2014 the proxy will execute the compression and return the result, then you continue fresh.
 - Do NOT wrap the marker in code fences, quotes, or commentary.
-- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`],
-  ["functionTools", `ACP TOOLS (FUNCTION CALLS)
+- NEVER compress on short conversations or when context is small (well below the window limit). Only compress when context is genuinely large.`
+  ],
+  [
+    "functionTools",
+    `ACP TOOLS (FUNCTION CALLS)
 
 The proxy also provides these as real function tools you can call directly (they appear in your tool list). Call them like any other function; the proxy executes them and returns the result, then you continue.
 
@@ -25890,7 +25908,8 @@ The proxy also provides these as real function tools you can call directly (they
 - search_context \u2014 search compressed block summaries by keyword. Arguments: {"query":"...","limit":5}.
 - decompress \u2014 restore compressed content for exact details. Arguments: {"blockId":"b5"} (optional "toFile":"/tmp/x.txt", "full":true).
 
-Note: compress is ONLY available via the text marker above (it needs batch ranges + an immediate stop), NOT as a function tool.`]
+Note: compress is ONLY available via the text marker above (it needs batch ranges + an immediate stop), NOT as a function tool.`
+  ]
 ];
 var DECOMPRESS_TOOL_OPENAI = {
   type: "function",
@@ -25943,11 +25962,24 @@ var ACP_STATUS_TOOL_OPENAI = {
     }
   }
 };
+var ACP_CACHE_TOOL_DESCRIPTION = "Prompt-cache reconciliation: grand ledger (total input/cached/output, overall hit rate) with every request's miss split into new content / compression re-pay / TTL expiry, plus per-fold economics (breakeven turns vs measured cadence). Read-only. Call when asked about cache hits, cache invalidation, or what compression costs.";
+var ACP_CACHE_TOOL_OPENAI = {
+  type: "function",
+  function: {
+    name: ACP_CACHE_TOOL_NAME,
+    description: ACP_CACHE_TOOL_DESCRIPTION,
+    parameters: {
+      type: "object",
+      properties: {}
+    }
+  }
+};
 var ACP_TOOLS_OPENAI = [
   COMPRESS_TOOL_OPENAI,
   DECOMPRESS_TOOL_OPENAI,
   SEARCH_CONTEXT_TOOL_OPENAI,
-  ACP_STATUS_TOOL_OPENAI
+  ACP_STATUS_TOOL_OPENAI,
+  ACP_CACHE_TOOL_OPENAI
 ];
 var DECOMPRESS_TOOL = {
   name: DECOMPRESS_TOOL_NAME,
@@ -25964,6 +25996,18 @@ var ACP_STATUS_TOOL = {
   description: ACP_STATUS_TOOL_OPENAI.function.description,
   input_schema: ACP_STATUS_TOOL_OPENAI.function.parameters
 };
+var ACP_CACHE_TOOL = {
+  name: ACP_CACHE_TOOL_NAME,
+  description: ACP_CACHE_TOOL_DESCRIPTION,
+  input_schema: ACP_CACHE_TOOL_OPENAI.function.parameters
+};
+var ACP_TOOLS_ANTHROPIC = [
+  COMPRESS_TOOL,
+  DECOMPRESS_TOOL,
+  SEARCH_CONTEXT_TOOL,
+  ACP_STATUS_TOOL,
+  ACP_CACHE_TOOL
+];
 var COMPRESS_TOOL_RESPONSES = {
   type: "function",
   name: COMPRESS_TOOL_NAME,
@@ -25988,6 +26032,32 @@ var ACP_STATUS_TOOL_RESPONSES = {
   description: ACP_STATUS_TOOL_OPENAI.function.description,
   parameters: ACP_STATUS_TOOL_OPENAI.function.parameters
 };
+var ACP_CACHE_TOOL_RESPONSES = {
+  type: "function",
+  name: ACP_CACHE_TOOL_NAME,
+  description: ACP_CACHE_TOOL_DESCRIPTION,
+  parameters: ACP_CACHE_TOOL_OPENAI.function.parameters
+};
+var ACP_TOOLS_RESPONSES = [
+  COMPRESS_TOOL_RESPONSES,
+  DECOMPRESS_TOOL_RESPONSES,
+  SEARCH_CONTEXT_TOOL_RESPONSES,
+  ACP_STATUS_TOOL_RESPONSES,
+  ACP_CACHE_TOOL_RESPONSES
+];
+var ACP_READONLY_TOOLS_RESPONSES = [
+  DECOMPRESS_TOOL_RESPONSES,
+  SEARCH_CONTEXT_TOOL_RESPONSES,
+  ACP_STATUS_TOOL_RESPONSES,
+  ACP_CACHE_TOOL_RESPONSES
+];
+var ACP_TOOL_NAMES = /* @__PURE__ */ new Set([
+  COMPRESS_TOOL_NAME,
+  DECOMPRESS_TOOL_NAME,
+  SEARCH_CONTEXT_TOOL_NAME,
+  ACP_STATUS_TOOL_NAME,
+  ACP_CACHE_TOOL_NAME
+]);
 var ABSORB_TOOL_DESCRIPTION = "Distill a tool result into a compact summary you write. REQUIRED immediately after a tool result ends with an [ACP absorb] instruction: pass its ref and the distilled essentials (outcome, key values, paths:lines, errors, decisions). The original output is then removed from context; your summary is the durable record.";
 var ABSORB_PARAMETERS = {
   type: "object",
@@ -26011,6 +26081,62 @@ var ABSORB_TOOL_OPENAI = {
     parameters: ABSORB_PARAMETERS
   }
 };
+var COMPRESS_TOOL_GOOGLE = {
+  name: COMPRESS_TOOL_NAME,
+  description: COMPRESS_TOOL.description,
+  // `anyOf` is rejected by older API revisions, so `content` declares the
+  // object form; a JSON-encoded string of that array is still accepted by
+  // parseCompressInput, and the line form (a summary whose first line is
+  // 'm00150–m00220 optional topic') is documented in the description.
+  parameters: {
+    type: "object",
+    properties: {
+      topic: {
+        type: "string",
+        description: "Optional short title for the compressed range"
+      },
+      content: {
+        type: "array",
+        description: "One or more ranges to compress into separate summary blocks. Object form: {startId,endId,summary,topic?}. A JSON-encoded string of that array is also accepted; in the line form the summary begins with its own first line 'm00150\u2013m00220 optional topic', the rest being the summary markdown verbatim. REQUIRED \u2014 compress without content is invalid.",
+        items: {
+          type: "object",
+          properties: {
+            topic: { type: "string" },
+            startId: {
+              type: "string",
+              description: "mNNNNN ref at the start of the range"
+            },
+            endId: {
+              type: "string",
+              description: "mNNNNN ref at the end of the range"
+            },
+            summary: {
+              type: "string",
+              description: "Self-contained summary replacing the range"
+            }
+          },
+          required: ["startId", "endId", "summary"]
+        }
+      }
+    },
+    required: ["content"]
+  }
+};
+var DECOMPRESS_TOOL_GOOGLE = {
+  name: DECOMPRESS_TOOL_NAME,
+  description: DECOMPRESS_TOOL_OPENAI.function.description,
+  parameters: DECOMPRESS_TOOL_OPENAI.function.parameters
+};
+var SEARCH_CONTEXT_TOOL_GOOGLE = {
+  name: SEARCH_CONTEXT_TOOL_NAME,
+  description: SEARCH_CONTEXT_TOOL_OPENAI.function.description,
+  parameters: SEARCH_CONTEXT_TOOL_OPENAI.function.parameters
+};
+var ACP_STATUS_TOOL_GOOGLE = {
+  name: ACP_STATUS_TOOL_NAME,
+  description: ACP_STATUS_TOOL_OPENAI.function.description,
+  parameters: ACP_STATUS_TOOL_OPENAI.function.parameters
+};
 function formatTokens(tokens) {
   if (tokens < 1e3) return String(tokens);
   if (tokens < 1e4) return (tokens / 1e3).toFixed(1) + "K";
@@ -26368,6 +26494,38 @@ var MAX_REQUEST_BYTES = 100 * 1024 * 1024;
 var UPSTREAM_TIMEOUT_MS = 12 * 60 * 1e3;
 
 // src/compress-tool.ts
+var UNADOPTED_KERNEL_TOOLS = /* @__PURE__ */ new Set([ACP_CACHE_TOOL_NAME]);
+var PROXY_TOOL_NAMES = new Set([...ACP_TOOL_NAMES].filter((n) => !UNADOPTED_KERNEL_TOOLS.has(n)));
+var SEARCH_CONTEXT_CONVERSATION_ID_PARAM = {
+  type: "string",
+  description: "Target bili conversation id. Defaults to the current conversation. May reference another historical pfa-* session for read-only search."
+};
+function withConversationId(schema) {
+  return { ...schema, properties: { ...schema.properties, conversation_id: SEARCH_CONTEXT_CONVERSATION_ID_PARAM } };
+}
+var BILI_SEARCH_CONTEXT_TOOL = {
+  name: SEARCH_CONTEXT_TOOL.name,
+  description: SEARCH_CONTEXT_TOOL.description,
+  input_schema: withConversationId(SEARCH_CONTEXT_TOOL.input_schema)
+};
+var BILI_SEARCH_CONTEXT_TOOL_OPENAI = {
+  type: "function",
+  function: {
+    name: SEARCH_CONTEXT_TOOL_OPENAI.function.name,
+    description: SEARCH_CONTEXT_TOOL_OPENAI.function.description,
+    parameters: withConversationId(SEARCH_CONTEXT_TOOL_OPENAI.function.parameters)
+  }
+};
+var BILI_SEARCH_CONTEXT_TOOL_RESPONSES = {
+  type: "function",
+  name: SEARCH_CONTEXT_TOOL_RESPONSES.name,
+  description: SEARCH_CONTEXT_TOOL_RESPONSES.description,
+  parameters: withConversationId(SEARCH_CONTEXT_TOOL_RESPONSES.parameters)
+};
+var BILI_ACP_TOOLS_ANTHROPIC = ACP_TOOLS_ANTHROPIC.filter((t) => !UNADOPTED_KERNEL_TOOLS.has(t.name)).map((t) => t.name === SEARCH_CONTEXT_TOOL_NAME ? BILI_SEARCH_CONTEXT_TOOL : t);
+var BILI_ACP_TOOLS_OPENAI = ACP_TOOLS_OPENAI.filter((t) => !UNADOPTED_KERNEL_TOOLS.has(t.function.name)).map((t) => t.function.name === SEARCH_CONTEXT_TOOL_NAME ? BILI_SEARCH_CONTEXT_TOOL_OPENAI : t);
+var BILI_ACP_TOOLS_RESPONSES = ACP_TOOLS_RESPONSES.filter((t) => !UNADOPTED_KERNEL_TOOLS.has(t.name)).map((t) => t.name === SEARCH_CONTEXT_TOOL_NAME ? BILI_SEARCH_CONTEXT_TOOL_RESPONSES : t);
+var BILI_ACP_READONLY_TOOLS_RESPONSES = ACP_READONLY_TOOLS_RESPONSES.filter((t) => !UNADOPTED_KERNEL_TOOLS.has(t.name)).map((t) => t.name === SEARCH_CONTEXT_TOOL_NAME ? BILI_SEARCH_CONTEXT_TOOL_RESPONSES : t);
 var ABSORB_TOOL_RESPONSES = {
   type: "function",
   name: ABSORB_TOOL_OPENAI.function.name,