context-mode 0.8.0 → 0.9.0

package/hooks/pretooluse.mjs

@@ -0,0 +1,157 @@
+#!/usr/bin/env node
+/**
+ * Unified PreToolUse hook for context-mode
+ * Redirects data-fetching tools to context-mode MCP tools
+ *
+ * Cross-platform (Windows/macOS/Linux) — no bash/jq dependency.
+ */
+
+let raw = "";
+process.stdin.setEncoding("utf-8");
+for await (const chunk of process.stdin) raw += chunk;
+
+const input = JSON.parse(raw);
+const tool = input.tool_name ?? "";
+const toolInput = input.tool_input ?? {};
+
+// ─── Bash: redirect data-fetching commands via updatedInput ───
+if (tool === "Bash") {
+  const command = toolInput.command ?? "";
+
+  // curl/wget → replace with echo redirect
+  if (/(^|\s|&&|\||\;)(curl|wget)\s/i.test(command)) {
+    console.log(JSON.stringify({
+      hookSpecificOutput: {
+        hookEventName: "PreToolUse",
+        updatedInput: {
+          command: 'echo "context-mode: curl/wget blocked. You MUST use mcp__context-mode__fetch_and_index(url, source) to fetch URLs, or mcp__context-mode__execute(language, code) to run HTTP calls in sandbox. Do NOT retry with curl/wget."',
+        },
+      },
+    }));
+    process.exit(0);
+  }
+
+  // inline fetch (node -e, python -c, etc.) → replace with echo redirect
+  if (
+    /fetch\s*\(\s*['"](https?:\/\/|http)/i.test(command) ||
+    /requests\.(get|post|put)\s*\(/i.test(command) ||
+    /http\.(get|request)\s*\(/i.test(command)
+  ) {
+    console.log(JSON.stringify({
+      hookSpecificOutput: {
+        hookEventName: "PreToolUse",
+        updatedInput: {
+          command: 'echo "context-mode: Inline HTTP blocked. Use mcp__context-mode__execute(language, code) to run HTTP calls in sandbox, or mcp__context-mode__fetch_and_index(url, source) for web pages. Do NOT retry with Bash."',
+        },
+      },
+    }));
+    process.exit(0);
+  }
+
+  // allow all other Bash commands
+  process.exit(0);
+}
+
+// ─── Read: nudge toward execute_file ───
+if (tool === "Read") {
+  console.log(JSON.stringify({
+    hookSpecificOutput: {
+      hookEventName: "PreToolUse",
+      additionalContext:
+        "CONTEXT TIP: If this file is large (>50 lines), prefer mcp__context-mode__execute_file(path, language, code) — processes in sandbox, only stdout enters context.",
+    },
+  }));
+  process.exit(0);
+}
+
+// ─── Grep: nudge toward execute ───
+if (tool === "Grep") {
+  console.log(JSON.stringify({
+    hookSpecificOutput: {
+      hookEventName: "PreToolUse",
+      additionalContext:
+        'CONTEXT TIP: If results may be large, prefer mcp__context-mode__execute(language: "shell", code: "grep ...") — runs in sandbox, only stdout enters context.',
+    },
+  }));
+  process.exit(0);
+}
+
+// ─── Glob: passthrough ───
+if (tool === "Glob") {
+  process.exit(0);
+}
+
+// ─── WebFetch: deny + redirect to sandbox ───
+if (tool === "WebFetch") {
+  const url = toolInput.url ?? "";
+  console.log(JSON.stringify({
+    hookSpecificOutput: {
+      hookEventName: "PreToolUse",
+      permissionDecision: "deny",
+      reason: `context-mode: WebFetch blocked. Use mcp__context-mode__fetch_and_index(url: "${url}", source: "...") to fetch this URL in sandbox. Then use mcp__context-mode__search(queries: [...]) to query results. Do NOT use curl/wget — they are also blocked.`,
+    },
+  }));
+  process.exit(0);
+}
+
+// ─── WebSearch: passthrough ───
+if (tool === "WebSearch") {
+  process.exit(0);
+}
+
+// ─── Task: inject context-mode routing into subagent prompts ───
+if (tool === "Task") {
+  const subagentType = toolInput.subagent_type ?? "";
+  const prompt = toolInput.prompt ?? "";
+
+  const ROUTING_BLOCK = `
+
+---
+CONTEXT WINDOW PROTECTION — USE CONTEXT-MODE MCP TOOLS
+
+Raw Bash/Read/WebFetch output floods your context. You have context-mode tools that keep data in sandbox.
+
+STEP 1 — GATHER: mcp__context-mode__batch_execute(commands, queries)
+  commands: [{label: "Name", command: "shell cmd"}, ...]
+  queries: ["query1", "query2", ...] — put 5-8 queries covering everything you need.
+  Runs all commands, indexes output, returns search results. ONE call, no follow-ups.
+
+STEP 2 — FOLLOW-UP: mcp__context-mode__search(queries: ["q1", "q2", "q3", ...])
+  Pass ALL follow-up questions as queries array. ONE call, not separate calls.
+
+OTHER: execute(language, code) | execute_file(path, language, code) | fetch_and_index(url) + search
+
+FORBIDDEN: Bash for output, Read for files, WebFetch. Bash is ONLY for git/mkdir/rm/mv.
+
+OUTPUT FORMAT — KEEP YOUR FINAL RESPONSE UNDER 500 WORDS:
+The parent agent context window is precious. Your full response gets injected into it.
+
+1. ARTIFACTS (PRDs, configs, code files) → Write to FILES, never return as inline text.
+   Return only: file path + 1-line description.
+2. DETAILED FINDINGS → Index into knowledge base:
+   mcp__context-mode__index(content: "...", source: "descriptive-label")
+   The parent agent shares the SAME knowledge base and can search() your indexed content.
+3. YOUR RESPONSE must be a concise summary:
+   - What you did (2-3 bullets)
+   - File paths created/modified (if any)
+   - Source labels you indexed (so parent can search)
+   - Key findings in bullet points
+   Do NOT return raw data, full file contents, or lengthy explanations.
+---`;
+
+  const updatedInput =
+    subagentType === "Bash"
+      ? { ...toolInput, prompt: prompt + ROUTING_BLOCK, subagent_type: "general-purpose" }
+      : { ...toolInput, prompt: prompt + ROUTING_BLOCK };
+
+  console.log(JSON.stringify({
+    hookSpecificOutput: {
+      hookEventName: "PreToolUse",
+      updatedInput,
+    },
+  }));
+  process.exit(0);
+}
+
+// Unknown tool — pass through
+process.exit(0);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "context-mode",
-  "version": "0.8.0",
+  "version": "0.9.0",
   "type": "module",
   "description": "Claude Code MCP plugin that saves 98% of your context window. Sandboxed code execution, FTS5 knowledge base, and intent-driven search.",
   "author": "Mert Koseoğlu",
@@ -32,7 +32,7 @@
     "skills",
     ".claude-plugin",
     ".mcp.json",
-    "start.sh",
+    "start.mjs",
     "README.md",
     "LICENSE"
   ],