muonroi-cli 1.4.1 → 1.5.0

package/dist/src/mcp/smart-filter.js

@@ -47,6 +47,37 @@ function hasDocsSignal(message) {
     return (/https?:\/\/\S+/i.test(message) ||
         /\b(docs?|documentation|api|sdk|library|libraries|framework|package|npm|pip|cargo|crate|gem|maven|nuget|install|migrat\w*|changelog|release\s*notes?|reference|usage|fetch|download|http|web|google|search\s+(the\s+)?web|news|weather|headlines|look\s*up|online|internet)\b/i.test(message));
 }
+/**
+ * Matches a question ABOUT the Muonroi ecosystem — where muonroi-docs (the
+ * authoritative ecosystem source: BB/.NET recipes, package docs, open-core
+ * boundary) is exactly what's needed, even though the message carries no generic
+ * docs/api keyword. Deliberately ecosystem-specific so it only ever KEEPS
+ * muonroi-docs (never other docs servers). EN + VI.
+ */
+function hasEcosystemSignal(message) {
+    return /\bmuonroi\b|\becosystem\b|hệ\s*sinh\s*thái|he\s*sinh\s*thai|building[-\s]?block|\bbb\b|open[-\s]?core/i.test(message);
+}
+/**
+ * Explicit "use a tool / MCP tool" intent. The filter only sees server *ids*,
+ * not their tool lists (MCP tools are fetched lazily at build time), so when the
+ * user asks to call a specific tool by name we cannot tell which server owns it.
+ * Dropping any optional server then risks stripping the exact tool requested.
+ *
+ * Live miss (session f6f7881a5fae): "bạn thử call tool setup_guide ... ( call
+ * tool chứ không phải đọc code )" carried no docs keyword, so `muonroi-docs`
+ * (id matches /docs/) was dropped — the model had no `setup_guide` tool and
+ * resorted to driving the server by hand over bash JSON-RPC, fabricating output.
+ *
+ * When this fires we keep ALL optional servers for the turn. Over-keeping costs
+ * tokens but never removes capability — this module's documented safe direction.
+ * EN + VI: "call/use/invoke/run the X tool", "tool call", "gọi/dùng/chạy tool".
+ */
+function hasExplicitToolIntent(message) {
+    return (/\b(?:call|use|invoke|run|exercise|trigger|try)\s+(?:the\s+)?(?:mcp\s+)?(?:[a-z0-9_.-]+\s+)?tool(?:s)?\b/i.test(message) ||
+        /\btool[\s_-]?call\b/i.test(message) ||
+        /\bmcp\b[^\n]*\btool/i.test(message) ||
+        /\b(?:gọi|dùng|chạy|thử)\s+(?:tool|mcp)\b/i.test(message));
+}
 /**
  * Filesystem-MCP tool names that 1:1 duplicate a first-class BUILTIN file tool.
  * The builtin `read_file`/`write_file`/`edit_file` are strictly better (read-
@@ -102,9 +133,27 @@ export function dropRedundantFsMcpTools(mcpTools, builtinToolNames) {
 export function filterMcpServersByMessage(servers, userMessage, opts = {}) {
     if (opts.disabled)
         return servers;
+    // Explicit "call the X tool" intent → keep every server this turn. We can't map
+    // a named tool back to its server from config alone (tool lists load lazily),
+    // so dropping any optional server risks stripping the exact tool requested.
+    if (hasExplicitToolIntent(userMessage))
+        return servers;
     const browser = hasBrowserSignal(userMessage);
     const docs = hasDocsSignal(userMessage);
+    const ecosystem = hasEcosystemSignal(userMessage);
+    const lower = userMessage.toLowerCase();
     return servers.filter((s) => {
+        // A server named outright in the message ("check the muonroi-docs MCP") is
+        // always relevant — never let a category skip override an explicit mention.
+        if (s.id && lower.includes(s.id.toLowerCase()))
+            return true;
+        // muonroi-docs is the AUTHORITATIVE ecosystem source. A question about the
+        // Muonroi ecosystem ("hệ sinh thái muonroi", "building-block", "bb rule
+        // engine") matches no generic docs/api keyword, so SKIP_WHEN_NO_DOCS would
+        // wrongly drop it and the agent falls back to guessing from files (live
+        // session dbe408937a3d turn 1). Keep it whenever the turn is ecosystem-about.
+        if (ecosystem && /(^|[-_])docs([-_]|$)/.test(s.id) && /muonroi/i.test(s.id))
+            return true;
         if (!browser && SKIP_WHEN_NO_BROWSER.test(s.id))
             return false;
         if (!docs && SKIP_WHEN_NO_DOCS.test(s.id))

package/dist/src/mcp/smoke.test.js

@@ -86,49 +86,49 @@ describe("MCP smoke test — buildMcpToolSet", () => {
     it.skipIf(process.platform === "win32" || !!process.env.CI)("discovers tools from stdio MCP echo stub", async () => {
         // Inline MCP echo server script — handles initialize, notifications/initialized,
         // and tools/list using Content-Length framing per the MCP JSON-RPC spec.
-        const echoServerScript = `
-const { stdin, stdout } = require('process');
-let buf = '';
-stdin.setEncoding('utf8');
-stdin.on('data', (chunk) => {
-  buf += chunk;
-  while (true) {
-    const headerEnd = buf.indexOf('\\r\\n\\r\\n');
-    if (headerEnd === -1) break;
-    const header = buf.slice(0, headerEnd);
-    const match = header.match(/Content-Length:\\s*(\\d+)/i);
-    if (!match) { buf = buf.slice(headerEnd + 4); continue; }
-    const len = parseInt(match[1], 10);
-    const bodyStart = headerEnd + 4;
-    if (buf.length < bodyStart + len) break;
-    const body = buf.slice(bodyStart, bodyStart + len);
-    buf = buf.slice(bodyStart + len);
-    handleMessage(JSON.parse(body));
-  }
-});
-function send(obj) {
-  const s = JSON.stringify(obj);
-  stdout.write('Content-Length: ' + Buffer.byteLength(s) + '\\r\\n\\r\\n' + s);
-}
-function handleMessage(msg) {
-  if (msg.method === 'initialize') {
-    send({ jsonrpc: '2.0', id: msg.id, result: {
-      protocolVersion: '2024-11-05',
-      capabilities: { tools: { listChanged: false } },
-      serverInfo: { name: 'echo-stub', version: '1.0.0' }
-    }});
-  } else if (msg.method === 'notifications/initialized') {
-    // no response needed
-  } else if (msg.method === 'tools/list') {
-    send({ jsonrpc: '2.0', id: msg.id, result: {
-      tools: [{
-        name: 'echo',
-        description: 'Echoes input',
-        inputSchema: { type: 'object', properties: { message: { type: 'string' } }, required: ['message'] }
-      }]
-    }});
-  }
-}
+        const echoServerScript = `
+const { stdin, stdout } = require('process');
+let buf = '';
+stdin.setEncoding('utf8');
+stdin.on('data', (chunk) => {
+  buf += chunk;
+  while (true) {
+    const headerEnd = buf.indexOf('\\r\\n\\r\\n');
+    if (headerEnd === -1) break;
+    const header = buf.slice(0, headerEnd);
+    const match = header.match(/Content-Length:\\s*(\\d+)/i);
+    if (!match) { buf = buf.slice(headerEnd + 4); continue; }
+    const len = parseInt(match[1], 10);
+    const bodyStart = headerEnd + 4;
+    if (buf.length < bodyStart + len) break;
+    const body = buf.slice(bodyStart, bodyStart + len);
+    buf = buf.slice(bodyStart + len);
+    handleMessage(JSON.parse(body));
+  }
+});
+function send(obj) {
+  const s = JSON.stringify(obj);
+  stdout.write('Content-Length: ' + Buffer.byteLength(s) + '\\r\\n\\r\\n' + s);
+}
+function handleMessage(msg) {
+  if (msg.method === 'initialize') {
+    send({ jsonrpc: '2.0', id: msg.id, result: {
+      protocolVersion: '2024-11-05',
+      capabilities: { tools: { listChanged: false } },
+      serverInfo: { name: 'echo-stub', version: '1.0.0' }
+    }});
+  } else if (msg.method === 'notifications/initialized') {
+    // no response needed
+  } else if (msg.method === 'tools/list') {
+    send({ jsonrpc: '2.0', id: msg.id, result: {
+      tools: [{
+        name: 'echo',
+        description: 'Echoes input',
+        inputSchema: { type: 'object', properties: { message: { type: 'string' } }, required: ['message'] }
+      }]
+    }});
+  }
+}
 `;
         const tmpDir = await mkdtemp(join(tmpdir(), "mcp-echo-stub-"));
         const scriptPath = join(tmpDir, "echo-server.js");

package/dist/src/mcp/tools-server.d.ts

@@ -6,9 +6,16 @@
  * immediately; poll with selfverify.status; fetch selfverify.result when done.
  *
  * Lives at the app layer (NOT agent-harness-core) so it may import src/self-qa.
+ *
+ * NOTE: this server is for EXTERNAL agents (Claude Code etc.). The CLI's OWN
+ * inner agent now exposes the same capabilities as NATIVE in-process builtins
+ * (src/tools/native-tools.ts) — it no longer self-spawns this server. The two
+ * surfaces share their cores (self-verify-runner.ts, setup-guide-text.ts, the
+ * ee/forensics/lsp modules) so behaviour is identical.
  */
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { JobManager, type Runner } from "./self-verify-jobs.js";
 export declare function registerSelfVerifyTools(server: McpServer, jm: JobManager): void;
+export declare function registerSetupGuideTool(server: McpServer): void;
 export declare function createToolsServer(runner?: Runner): McpServer;
 export declare function runToolsMcpServer(): Promise<void>;

package/dist/src/mcp/tools-server.js

@@ -6,6 +6,12 @@
  * immediately; poll with selfverify.status; fetch selfverify.result when done.
  *
  * Lives at the app layer (NOT agent-harness-core) so it may import src/self-qa.
+ *
+ * NOTE: this server is for EXTERNAL agents (Claude Code etc.). The CLI's OWN
+ * inner agent now exposes the same capabilities as NATIVE in-process builtins
+ * (src/tools/native-tools.ts) — it no longer self-spawns this server. The two
+ * surfaces share their cores (self-verify-runner.ts, setup-guide-text.ts, the
+ * ee/forensics/lsp modules) so behaviour is identical.
  */
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
@@ -14,29 +20,9 @@ import { registerEETools } from "./ee-tools.js";
 import { registerForensicsTools } from "./forensics-tools.js";
 import { registerLspTools } from "./lsp-tools.js";
 import { JobManager } from "./self-verify-jobs.js";
+import { defaultRunner } from "./self-verify-runner.js";
+import { SETUP_GUIDE_TEXT } from "./setup-guide-text.js";
 const LOG_TAIL = 40;
-/** Default runner: drives the real self-verify functions in-process. */
-const defaultRunner = {
-    async tier1(opts, log) {
-        // signal intentionally not forwarded: runSelfVerify/runAgenticLoop do not yet
-        // accept an AbortSignal. cancel() marks the job and discards the late result.
-        const { runSelfVerify } = await import("../self-qa/index.js");
-        return runSelfVerify({
-            baseRef: opts.since,
-            maxScenarios: opts.max,
-            emitSpecs: opts.emit,
-            specOutDir: opts.out,
-            log,
-        });
-    },
-    async agentic(opts, log) {
-        // signal intentionally not forwarded: runSelfVerify/runAgenticLoop do not yet
-        // accept an AbortSignal. cancel() marks the job and discards the late result.
-        const { createLLMBrain, runAgenticLoop } = await import("../self-qa/agentic-loop.js");
-        const brain = await createLLMBrain({ modelId: opts.llm });
-        return runAgenticLoop({ goal: opts.goal, brain, maxTurns: opts.turns ?? 20, log });
-    },
-};
 function ok(data) {
     return { content: [{ type: "text", text: JSON.stringify(data) }] };
 }
@@ -129,6 +115,16 @@ export function registerSelfVerifyTools(server, jm) {
     });
     server.registerTool("selfverify_cancel", { description: "Cancel a running self-verify run (best-effort).", inputSchema: { runId: z.string() } }, async ({ runId }) => ok({ cancelled: jm.cancel(runId) }));
 }
+export function registerSetupGuideTool(server) {
+    server.registerTool("setup_guide", {
+        description: "Returns a concise, up-to-date setup, install, first-run, MCP wiring, and verification guide for muonroi-cli. " +
+            "Call this directly (setup_guide) when the user asks for setup instructions, onboarding, or 'how do I start' — " +
+            "instead of guessing, reading files, or shelling commands. Keeps agents on the happy path.",
+        inputSchema: {},
+    }, async () => {
+        return { content: [{ type: "text", text: SETUP_GUIDE_TEXT }] };
+    });
+}
 export function createToolsServer(runner = defaultRunner) {
     const server = new McpServer({ name: "muonroi-tools", version: "0.1.0" });
     const jm = new JobManager(runner);
@@ -136,6 +132,7 @@ export function createToolsServer(runner = defaultRunner) {
     registerEETools(server);
     registerForensicsTools(server);
     registerLspTools(server);
+    registerSetupGuideTool(server);
     return server;
 }
 export async function runToolsMcpServer() {