modelfunction-cli 0.2.0 → 0.4.0

package/lib/mcp-server.js

@@ -57,6 +57,7 @@ export async function startMcpServer() {
             source: mcpServer.source,
             origin: "mcp-probe",
             confidence: "high",
+            replaceExisting: true,
             note: `server ${mcpServer.name}`
           });
           report.probes.push({ server: mcpServer.name, ok: true, toolCount: added.length, tools: added });

package/lib/scan.js

@@ -1,5 +1,6 @@
 import { readdir, readFile, stat } from "node:fs/promises";
 import { basename, extname, join, relative, resolve } from "node:path";
+import { extractJsTools } from "./js-tools.js";
 
 const SKIP_DIRS = new Set([
   "node_modules",
@@ -19,7 +20,10 @@ const SKIP_DIRS = new Set([
 ]);
 
 const MAX_DEPTH = 6;
-const MAX_FILES = 5000;
+// 5k was too small: a real agent monorepo (openclaw) has 16k files after the
+// skip list, so the walk stopped a third of the way in and reported 26 of its
+// 53 skills. Silent partial results are worse than a slow scan.
+const MAX_FILES = 25000;
 const MAX_FILE_BYTES = 200_000;
 
 const FUNCTION_SPEC_NAMES = new Set([
@@ -81,7 +85,8 @@ const JS_FRAMEWORKS = {
 export async function scanProject(root = ".") {
   const absRoot = resolve(root);
   const files = [];
-  await walk(absRoot, 0, files);
+  const state = { truncated: false };
+  await walk(absRoot, 0, files, state);
 
   const report = {
     root: absRoot,
@@ -97,7 +102,16 @@ export async function scanProject(root = ".") {
     warnings: []
   };
 
+  if (state.truncated) {
+    report.warnings.push(`扫描在 ${MAX_FILES} 个文件处截断，结果可能不完整；可把 --root 指向子目录`);
+  }
+
   const seenTools = new Set();
+  // Tool names are frequently constants defined in a sibling module
+  // (`name: CONFIG_TOOL_NAME`). Files are visited in path order, so the
+  // constant may not be known yet; such files are retried after the walk.
+  const constIndex = new Map();
+  const retryFiles = new Set();
   const dependencies = new Set();
   const frameworks = new Set();
 
@@ -224,6 +238,23 @@ export async function scanProject(root = ".") {
           });
         }
       }
+      if (!name.endsWith(".py") && !TEST_PATH.test(rel)) {
+        for (const match of text.matchAll(/(?:export\s+)?const\s+([A-Z][A-Z0-9_]{2,})\s*=\s*(['"])([^'"\n]{1,200})\2/g)) {
+          if (!constIndex.has(match[1])) constIndex.set(match[1], match[3]);
+        }
+        const jsTools = extractJsTools(text, {
+          externals: constIndex,
+          onUnresolvedName: () => retryFiles.add(file)
+        });
+        if (jsTools.length > 0) {
+          addTools(report, jsTools, {
+            source: rel,
+            origin: "code-literal",
+            confidence: "medium",
+            note: "从 JS/TS 声明或 TypeBox/zod schema 静态提取（不执行代码），需人工确认"
+          });
+        }
+      }
       if (TEST_PATH.test(rel)) continue;
       // Only matches that carry a tool name are actionable; a bare `input_schema:`
       // in API plumbing is not a tool and only inflates the report.
@@ -249,19 +280,46 @@ export async function scanProject(root = ".") {
     }
   }
 
+  if (retryFiles.size > 0 && constIndex.size > 0) {
+    for (const file of retryFiles) {
+      const text = await readText(file, report);
+      if (!text) continue;
+      const tools = extractJsTools(text, { externals: constIndex });
+      if (tools.length === 0) continue;
+      addTools(report, tools, {
+        source: relative(absRoot, file),
+        origin: "code-literal",
+        confidence: "medium",
+        note: "名称常量在其他模块定义，扫描完整仓库后再解析"
+      });
+    }
+  }
+
   report.dependencies = [...dependencies].sort();
   report.frameworks = [...frameworks].sort();
   return report;
 }
 
-export function addTools(report, tools, { source, origin, confidence = "high", note, meta }) {
-  const existing = new Set(report.tools.map((tool) => tool.name));
+export function addTools(
+  report,
+  tools,
+  { source, origin, confidence = "high", note, meta, replaceExisting = false }
+) {
+  const index = new Map(report.tools.map((tool, position) => [tool.name, position]));
   const added = [];
   for (const tool of tools) {
-    if (!tool?.name || existing.has(tool.name)) continue;
-    existing.add(tool.name);
-    report.tools.push({ ...tool, source, origin, ...(meta ?? {}) });
-    added.push(tool.name);
+    if (!tool?.name) continue;
+    const existing = index.get(tool.name);
+    if (existing === undefined) {
+      index.set(tool.name, report.tools.length);
+      report.tools.push({ ...tool, source, origin, ...(meta ?? {}) });
+      added.push(tool.name);
+    } else if (replaceExisting) {
+      // A probed tool is the runtime truth; a static guess with the same name
+      // should be replaced, not silently kept.
+      report.tools[existing] = { ...tool, source, origin, ...(meta ?? {}) };
+      added.push(tool.name);
+    }
   }
   report.sources.push({
     type: origin,
@@ -294,7 +352,9 @@ export function buildSuiteDraft(report) {
     })),
     cases: []
   };
-  if (report.skills.length > 0) draft.skills = report.skills.map((skill) => skill.name);
+  if (report.skills.length > 0) {
+    draft.skills = report.skills.map((skill) => ({ name: skill.name, description: skill.description }));
+  }
   // Where a tool came from is the only mount hint the scan can observe without
   // guessing: one MCP server or one spec file is normally one thing agents mount.
   const profiles = groupBySource(report.tools);
@@ -354,7 +414,7 @@ export function formatScanSummary(report) {
   lines.push(
     "",
     `tools:       ${report.tools.length}`,
-    `skills:      ${report.skills.length}（不进入工具评测）`,
+    `skills:      ${report.skills.length}（仅静态检查）`,
     `mcp servers: ${report.mcpServers.length}${probed > 0 ? `（已 probe ${probed} 个）` : "（需 --probe 才能拿到工具清单）"}`,
     `code match:  ${report.codeMatches.length}（low confidence，需人工确认）`
   );
@@ -385,8 +445,11 @@ export function formatScanSummary(report) {
   return lines.join("\n");
 }
 
-async function walk(dir, depth, files) {
-  if (depth > MAX_DEPTH || files.length >= MAX_FILES) return;
+async function walk(dir, depth, files, state) {
+  if (depth > MAX_DEPTH || files.length >= MAX_FILES) {
+    if (files.length >= MAX_FILES) state.truncated = true;
+    return;
+  }
   let entries;
   try {
     entries = await readdir(dir, { withFileTypes: true });
@@ -395,11 +458,14 @@ async function walk(dir, depth, files) {
   }
   entries.sort((a, b) => a.name.localeCompare(b.name));
   for (const entry of entries) {
-    if (files.length >= MAX_FILES) return;
+    if (files.length >= MAX_FILES) {
+      state.truncated = true;
+      return;
+    }
     const full = join(dir, entry.name);
     if (entry.isDirectory()) {
       if (SKIP_DIRS.has(entry.name)) continue;
-      await walk(full, depth + 1, files);
+      await walk(full, depth + 1, files, state);
     } else if (entry.isFile()) {
       files.push(full);
     }

package/lib/suite.js

@@ -3,10 +3,10 @@ import { dirname, resolve } from "node:path";
 
 const EMPTY_SCHEMA = { type: "object", properties: {} };
 
-export async function loadSuite(suitePath, { allowEmptyCases = false } = {}) {
+export async function loadSuite(suitePath, { allowEmptyCases = false, allowEmptyTools = false } = {}) {
   const abs = resolve(suitePath);
   const raw = JSON.parse(await readFile(abs, "utf8"));
-  if (!Array.isArray(raw.tools) || raw.tools.length === 0) {
+  if (!Array.isArray(raw.tools) || (raw.tools.length === 0 && !allowEmptyTools)) {
     throw new Error(`suite ${abs}: "tools" must be a non-empty array`);
   }
   if (!Array.isArray(raw.cases) || (raw.cases.length === 0 && !allowEmptyCases)) {
@@ -29,7 +29,7 @@ export async function loadSuite(suitePath, { allowEmptyCases = false } = {}) {
     tools,
     cases,
     profiles: raw.profiles ?? null,
-    skills: Array.isArray(raw.skills) ? raw.skills.map(String) : [],
+    skills: Array.isArray(raw.skills) ? raw.skills.map(normalizeSkill) : [],
     execution:
       raw.execution && typeof raw.execution === "object" && !Array.isArray(raw.execution)
         ? { ...raw.execution }
@@ -58,6 +58,23 @@ async function normalizeTool(entry, baseDir) {
   };
 }
 
+// Skills carry no parameter schema in the prompt-injected form (SKILL.md
+// frontmatter), so the only checkable surface is name + description. Both the
+// bare-name shorthand and the full form are accepted.
+function normalizeSkill(entry, index) {
+  if (typeof entry === "string") {
+    const name = entry.trim();
+    if (!name) throw new Error(`skill #${index + 1}: name must be a non-empty string`);
+    return { name, description: "" };
+  }
+  if (entry && typeof entry === "object") {
+    const name = typeof entry.name === "string" ? entry.name.trim() : "";
+    if (!name) throw new Error(`skill #${index + 1}: "name" must be a non-empty string`);
+    return { name, description: typeof entry.description === "string" ? entry.description : "" };
+  }
+  throw new Error(`skill #${index + 1}: must be a name or {name, description}`);
+}
+
 function normalizeCase(entry, index, toolNames) {
   const id = entry.id ?? `case_${index + 1}`;
   if (typeof entry.input !== "string" || entry.input.trim() === "") {

package/mf.js

@@ -50,10 +50,10 @@ Usage:
   mf scan [options]
   mf case add --suite <file> (--input <file> | stdin) [--source agent] [--reviewed] [--dry-run]
   mf coverage [--suite <file>] [--json]
-  mf lint [--suite <file>] [--json]
+  mf lint [--suite <file>] [--json]  # 静态检查 tools 与 skills，不调用模型
   mf compare <before.json> <after.json> [--json]
   mf ablate --tool <name> [--suite <file>] [--profile <group>] [--models <ids>] [--json]
-  mf cluster [--suite <file>] [--threshold <0.5>] [--max-size <10>] [--out <file>] [--json]  # 建议挂载分组
+  mf cluster [--suite <file>] [--threshold <0.5>] [--max-size <10>] [--out <file>] [--json]  # 建议挂载分组 + 同族候选
   mf exec --entry <file> (--input <json> | --input-json <json>) [--runtime node|python] [--python <cmd>] [--export run] [--allow-read <paths>] [--allow-write <paths>] [--allow-net <hosts>] [--timeout <ms>] [--warmup <n>] [--json]
   mf sandbox check [--json]   # probe available isolation engines on this machine
   mf mcp                # start stdio MCP server for coding agents
@@ -344,6 +344,7 @@ export async function runScan(options, io = console) {
         source: server.source,
         origin: "mcp-probe",
         confidence: "high",
+        replaceExisting: true,
         note: `server ${server.name}`,
         meta: { provider: server.name }
       });
@@ -408,8 +409,11 @@ export async function runCaseAdd(options, io = console) {
 
 export async function runCoverage(options, io = console) {
   // A scan draft has no cases yet; "what is still missing" is exactly the view
-  // a user needs right after scanning.
-  const suite = await loadSuite(resolve(options.suite ?? DEFAULT_SUITE), { allowEmptyCases: true });
+  // a user needs right after scanning. A skills-only suite is also valid.
+  const suite = await loadSuite(resolve(options.suite ?? DEFAULT_SUITE), {
+    allowEmptyCases: true,
+    allowEmptyTools: true
+  });
   const report = computeCoverage(suite);
   io.log(options.json ? JSON.stringify(report, null, 2) : formatCoverage(report));
   return 0;
@@ -422,7 +426,10 @@ export async function runSandboxCheck(options, io = console) {
 }
 
 export async function runLint(options, io = console) {
-  const suite = await loadSuite(resolve(options.suite ?? DEFAULT_SUITE), { allowEmptyCases: true });
+  const suite = await loadSuite(resolve(options.suite ?? DEFAULT_SUITE), {
+    allowEmptyCases: true,
+    allowEmptyTools: true
+  });
   const report = lintSuite(suite);
   io.log(options.json ? JSON.stringify(report, null, 2) : formatLint(report));
   return report.counts.error > 0 ? 1 : 0;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "modelfunction-cli",
-  "version": "0.2.0",
+  "version": "0.4.0",
   "type": "module",
   "license": "MIT",
   "description": "mf — function-calling accuracy scorecard for AI agents",