opencode-gbk-tools 0.1.5 → 0.1.7

package/dist/agents/gbk-engine.md

@@ -1,5 +1,5 @@
 ---
-description: 处理 GBK/GB18030 编码文件的专用代理，只使用 gbk_read、gbk_write、gbk_edit
+description: 处理 GBK/GB18030 编码文件的专用代理，只使用 gbk_read、gbk_write、gbk_edit、gbk_search
 mode: primary
 permission:
   read: deny
@@ -18,6 +18,7 @@ permission:
 - 读取文件时优先使用 `gbk_read`
 - 创建或覆盖文件时优先使用 `gbk_write`
 - 修改已有文件时优先使用 `gbk_edit`
+- 在大文件中搜索内容时使用 `gbk_search`
 - 大文件读取优先使用 `gbk_read` 的分页或 `tail` 能力，避免一次读取过多内容
 - 修改时优先缩小编辑范围：能用 `startLine/endLine` 或 `startAnchor/endAnchor` 就不要全文替换
 - 文件发现可使用 `glob`，但文件内容读取必须使用 `gbk_read`
@@ -25,3 +26,14 @@ permission:
 - `edit: deny` 同时覆盖内置 `write`、`patch`、`multiedit`
 - 如果用户请求涉及 UTF-8 文件或二进制文件，先明确说明不适用
 - 若目标文件编码不确定，先提醒用户确认是 `gbk` 还是 `gb18030`
+
+## 大文件编辑工作流（必须遵守）
+
+当 `gbk_read` 返回 `truncated: true` 时，说明文件超出了读取窗口。
+此时若需要编辑文件，**必须按以下步骤操作，禁止直接猜测 oldString**：
+
+1. `gbk_search(filePath, pattern)` — 找到目标内容的精确行号和上下文
+2. `gbk_read(filePath, offset=<lineNumber>, limit=<N>)` — 读取目标行的精确内容
+3. `gbk_edit(filePath, oldString=<精确内容>, newString=<新内容>)` — 用精确内容替换
+
+**禁止**在未确认精确内容的情况下构造 `oldString`，否则 `gbk_edit` 必然报 `GBK_NO_MATCH`。

package/dist/opencode-tools/gbk_edit.js

@@ -16457,14 +16457,14 @@ function buildNoMatchMessage(content, oldString) {
     getNearestContext(content, oldString)
   ].join("\n");
 }
-function tryLooseBlockReplace(content, oldString, newString) {
-  const normalizedContent = normalizeNewlines(content);
-  const contentLines = splitNormalizedLines(content);
-  const oldLines = trimTrailingEmptyLines(splitNormalizedLines(oldString));
-  const newLines = splitNormalizedLines(newString);
-  if (oldLines.length === 0) {
-    return null;
-  }
+function hasLineNumberPrefixes(lines) {
+  const nonEmpty = lines.filter((l) => l.trim().length > 0);
+  return nonEmpty.length > 0 && nonEmpty.every((l) => /^\d+: /.test(l));
+}
+function stripLineNumberPrefixes(lines) {
+  return lines.map((l) => l.replace(/^\d+: /, ""));
+}
+function matchLooseBlock(contentLines, oldLines, newLines, newlineStyle, content) {
   for (let start = 0; start < contentLines.length; start += 1) {
     let contentIndex = start;
     let oldIndex = 0;
@@ -16494,12 +16494,32 @@ function tryLooseBlockReplace(content, oldString, newString) {
       ].join("\n");
       return {
         occurrencesBefore: 1,
-        content: detectNewlineStyle(content) === "crlf" ? replacedNormalized.replace(/\n/g, "\r\n") : replacedNormalized
+        content: newlineStyle === "crlf" ? replacedNormalized.replace(/\n/g, "\r\n") : replacedNormalized
       };
     }
   }
   return null;
 }
+function tryLooseBlockReplace(content, oldString, newString) {
+  const contentLines = splitNormalizedLines(content);
+  const oldLines = trimTrailingEmptyLines(splitNormalizedLines(oldString));
+  const newLines = splitNormalizedLines(newString);
+  const newlineStyle = detectNewlineStyle(content);
+  if (oldLines.length === 0) {
+    return null;
+  }
+  const result = matchLooseBlock(contentLines, oldLines, newLines, newlineStyle, content);
+  if (result !== null) {
+    return result;
+  }
+  if (hasLineNumberPrefixes(oldLines)) {
+    const strippedOldLines = trimTrailingEmptyLines(stripLineNumberPrefixes(oldLines));
+    if (strippedOldLines.length > 0) {
+      return matchLooseBlock(contentLines, strippedOldLines, newLines, newlineStyle, content);
+    }
+  }
+  return null;
+}
 function countOccurrences(text, target) {
   if (target.length === 0) {
     throw createGbkError("GBK_EMPTY_OLD_STRING", "oldString \u4E0D\u80FD\u4E3A\u7A7A");
@@ -16725,12 +16745,22 @@ var gbk_edit_default = tool({
 Reads the FULL file content regardless of file size \u2014 not limited by gbk_read's line window.
 Safe to use on files with more than 2000 lines.
 
+CRITICAL \u2014 do NOT include line number prefixes in oldString or newString:
+gbk_read output looks like "3787: SENDMSG 0 content". The "3787: " is a navigation prefix, NOT file content.
+oldString must be the raw file content: "SENDMSG 0 content" (no line number prefix).
+Including line numbers in oldString will cause GBK_NO_MATCH even if the content exists.
+
+Recommended workflow for large files (when gbk_read returned truncated=true):
+1. gbk_search(pattern) \u2192 find exact lineNumber
+2. gbk_read(offset=<lineNumber>, limit=20) \u2192 get the exact block (strip "N: " prefixes)
+3. gbk_edit(oldString=<content without prefixes>, newString=<new content>)
+
 For large files, use 'startLine'/'endLine' or 'startAnchor'/'endAnchor' to narrow the search scope
 and avoid false matches. Scoped edits also improve performance on very large files.`,
   args: {
     filePath: tool.schema.string().describe("Target file path"),
-    oldString: tool.schema.string().describe("Exact text to replace (must match file content, not gbk_read output with line numbers)"),
-    newString: tool.schema.string().describe("Replacement text"),
+    oldString: tool.schema.string().describe("Exact text to replace \u2014 raw file content only, no 'N: ' line number prefixes from gbk_read output"),
+    newString: tool.schema.string().describe("Replacement text \u2014 raw content only, no line number prefixes"),
     replaceAll: tool.schema.boolean().optional().describe("Replace all occurrences (default: false, requires unique match)"),
     startLine: tool.schema.union([tool.schema.number().int().positive(), tool.schema.literal(-1)]).optional().describe("Restrict edit scope to 1-based start line (inclusive)"),
     endLine: tool.schema.union([tool.schema.number().int().positive(), tool.schema.literal(-1)]).optional().describe("Restrict edit scope to 1-based end line (inclusive)"),

package/dist/opencode-tools/gbk_read.js

@@ -16609,14 +16609,23 @@ var gbk_read_default = tool({
 Returns up to 'limit' lines (default 2000) starting from 'offset'.
 When the file has more lines than the window, 'truncated' is true and 'totalLines' shows the full count.
 
-IMPORTANT: If 'truncated' is true, the returned content is incomplete.
-DO NOT use the returned content as 'oldString' for gbk_edit on a truncated file.
-To edit content beyond the visible window, use gbk_edit with 'startLine'/'endLine' to target the exact range,
-or read the specific range first with 'offset' set to the desired line number.`,
+IMPORTANT \u2014 line number format: each output line is prefixed with "N: " (e.g. "3787: content").
+These prefixes are for navigation only. Strip them before using any line as 'oldString' in gbk_edit.
+
+IMPORTANT \u2014 do NOT use large limits to read the whole file:
+- Setting limit=40000 or similar to "read everything" is WRONG. It produces an enormous response
+  that is unreliable to process and may be truncated by the protocol.
+- To find content in a large file, use gbk_search instead.
+- To read a specific section, set offset=<lineNumber> and limit=<small number like 10-30>.
+
+Workflow when truncated=true:
+1. gbk_search(pattern) \u2192 get exact lineNumber
+2. gbk_read(offset=<lineNumber>, limit=20) \u2192 get the exact block
+3. gbk_edit(oldString=<exact content without line prefixes>) \u2192 edit reliably`,
   args: {
     filePath: tool.schema.string().describe("Target file path"),
     offset: tool.schema.union([tool.schema.number().int().positive(), tool.schema.literal(-1)]).optional().describe("1-based start line (default: 1)"),
-    limit: tool.schema.union([tool.schema.number().int().positive(), tool.schema.literal(-1)]).optional().describe("Number of lines to read (default: 2000). Use -1 to apply the default."),
+    limit: tool.schema.union([tool.schema.number().int().positive(), tool.schema.literal(-1)]).optional().describe("Number of lines to read (default: 2000, max recommended: 200). Use -1 to apply the default."),
     tail: tool.schema.boolean().optional().describe("Read last N lines instead of offset-based window"),
     encoding: tool.schema.enum(["gbk", "gb18030"]).optional().describe("Text encoding (default: gbk)"),
     allowExternal: tool.schema.boolean().optional().describe("Allow paths outside workspace root")