eminiar 0.10.0 → 0.10.1

package/dist/server.js

@@ -13518,7 +13518,7 @@ function chunkMarkdown(text2, baseTitle, opts = DEFAULT_CHUNKER) {
     if (/^\s*(```|~~~)/.test(line)) inFence = !inFence;
     const m = !inFence && /^(#{1,6})\s+(.+?)\s*#*\s*$/.exec(line);
     if (m) {
-      if (current.lines.length) sections.push(current);
+      sections.push(current);
       const level = m[1].length;
       while (stack.length && stack[stack.length - 1].level >= level) stack.pop();
       stack.push({ level, text: m[2] });
@@ -13527,14 +13527,23 @@ function chunkMarkdown(text2, baseTitle, opts = DEFAULT_CHUNKER) {
     }
     current.lines.push(line);
   });
-  if (current.lines.length) sections.push(current);
+  sections.push(current);
   const out = [];
-  for (const s of sections) {
+  sections.forEach((s, k) => {
     const title = s.path.length ? `${baseTitle} > ${s.path.join(" > ")}` : baseTitle;
     const body = s.lines.join("\n");
+    if (!body.trim()) {
+      const next = sections[k + 1];
+      const isParent = next !== void 0 && next.path.length > s.path.length && s.path.every((p, j) => next.path[j] === p);
+      if (s.path.length && !isParent) {
+        const heading = s.path[s.path.length - 1];
+        out.push({ title, body: heading, startLine: s.startLine, endLine: s.startLine });
+      }
+      return;
+    }
     if (body.length <= opts.maxChars) {
       out.push({ title, body, startLine: s.startLine, endLine: s.startLine + s.lines.length - 1 });
-      continue;
+      return;
     }
     let fence2 = false;
     const fenceState = s.lines.map((l) => {
@@ -13550,7 +13559,7 @@ function chunkMarkdown(text2, baseTitle, opts = DEFAULT_CHUNKER) {
         s.startLine - 1
       )
     );
-  }
+  });
   return out;
 }
 var init_markdown = __esm({
@@ -16953,7 +16962,20 @@ import { readFileSync as readFileSync6, statSync as statSync4 } from "fs";
 import { relative as relative5, resolve as resolve6 } from "path";
 function indexContent(db, input2) {
   const kind = input2.kind ?? (input2.contentType === "code" ? "code" : input2.contentType === "json" ? "json" : void 0);
-  const { chunks, kind: k } = input2.chunks ? { chunks: finalize2(input2.chunks, input2.contentType ?? "code"), kind: kind ?? "code" } : chunk(input2.content, input2.title, kind, input2.uri);
+  const cut = input2.chunks ? { chunks: finalize2(input2.chunks, input2.contentType ?? "code"), kind: kind ?? "code" } : chunk(input2.content, input2.title, kind, input2.uri);
+  const k = cut.kind;
+  const fallback = cut.chunks.length === 0 && input2.content.trim().length > 0;
+  const chunks = fallback ? finalize2(
+    [
+      {
+        title: input2.title,
+        body: input2.content,
+        startLine: 1,
+        endLine: input2.content.split("\n").length
+      }
+    ],
+    kindToContentType(k)
+  ) : cut.chunks;
   const contentHash = hashText(input2.content);
   const existing = input2.uri ? db.findDocumentByUri(input2.uri) : void 0;
   if (existing && existing.content_hash === contentHash) {
@@ -16963,7 +16985,8 @@ function indexContent(db, input2) {
         created: false,
         chunks: chunks.length,
         chars: input2.content.length,
-        kind: k
+        kind: k,
+        fallback
       };
     }
     db.deleteDocument(existing.id);
@@ -16980,11 +17003,25 @@ function indexContent(db, input2) {
     },
     chunks
   );
-  return { documentId: id, created, chunks: chunks.length, chars: input2.content.length, kind: k };
+  return {
+    documentId: id,
+    created,
+    chunks: chunks.length,
+    chars: input2.content.length,
+    kind: k,
+    fallback
+  };
 }
 function indexPath(db, root, target, opts = {}) {
   const abs = resolve6(root, target);
-  const summary = { files: 0, indexed: 0, unchanged: 0, skipped: [], chunks: 0 };
+  const summary = {
+    files: 0,
+    indexed: 0,
+    unchanged: 0,
+    skipped: [],
+    chunks: 0,
+    fallbacks: 0
+  };
   const st = statSync4(abs);
   const files = st.isDirectory() ? import_fast_glob.default.sync(opts.globs ?? ["**/*.md"], {
     cwd: abs,
@@ -17019,6 +17056,7 @@ function indexPath(db, root, target, opts = {}) {
       if (r.created) summary.indexed++;
       else summary.unchanged++;
       summary.chunks += r.chunks;
+      if (r.fallback) summary.fallbacks++;
     } catch (err) {
       summary.skipped.push({ path: rel, reason: err.message });
     }
@@ -49547,6 +49585,10 @@ function topTermsOf(bodies, limit) {
 
 // src/storage/sqlite/schema.ts
 init_esm_shims();
+function hasColumn(db, table, column) {
+  const cols = db.pragma(`table_info(${table})`);
+  return cols.some((c) => c.name === column);
+}
 var migrations = [
   {
     version: 1,
@@ -49881,7 +49923,6 @@ CREATE TABLE IF NOT EXISTS transcript_files (
   offset INTEGER NOT NULL DEFAULT 0,
   size INTEGER NOT NULL DEFAULT 0,
   mtime INTEGER NOT NULL DEFAULT 0,
-  head_hash TEXT,
   ingested_at INTEGER NOT NULL
 );
 
@@ -49909,6 +49950,16 @@ CREATE TABLE IF NOT EXISTS statusline_snapshots (
 CREATE INDEX IF NOT EXISTS statusline_snapshots_session_ts ON statusline_snapshots(session_id, ts);
 CREATE INDEX IF NOT EXISTS statusline_snapshots_ts ON statusline_snapshots(ts);
 `
+  },
+  {
+    // transcript_files.head_hash (fingerprint of the file head, so a rewrite with the same
+    // size is noticed) was first added to v5 after v5 had already run on real databases, so
+    // version-5 databases exist both with and without the column. Add it only when missing.
+    version: 6,
+    up(db) {
+      if (!hasColumn(db, "transcript_files", "head_hash"))
+        db.exec("ALTER TABLE transcript_files ADD COLUMN head_hash TEXT");
+    }
   }
 ];
 function vecTableSql(dims) {
@@ -50020,7 +50071,8 @@ var SqliteAdapter = class _SqliteAdapter {
     if (pending2.length) {
       this.transaction(() => {
         for (const m of pending2) {
-          this.db.exec(m.sql);
+          if (m.sql) this.db.exec(m.sql);
+          m.up?.(this.db);
           this.metaSet("schema_version", String(m.version));
         }
       });
@@ -50239,18 +50291,42 @@ var SqliteAdapter = class _SqliteAdapter {
   }
   vecSearch(embedding, k, filters) {
     if (!this.vectors) return [];
-    const raw = this.db.prepare(
-      "SELECT chunk_id id, distance FROM chunk_vec WHERE embedding MATCH ? AND k = ? ORDER BY distance"
-    ).all(embedding, BigInt(filters ? k * 4 : k));
-    if (!filters || !raw.length) return raw.map((r) => ({ chunkId: r.id, distance: r.distance }));
     const { where, params } = filterSql(filters);
-    const allowed = new Set(
-      this.db.prepare(
-        `SELECT c.id FROM chunks c JOIN documents d ON d.id = c.document_id
-             WHERE c.id IN (${raw.map(() => "?").join(",")})${where}`
-      ).all(...raw.map((r) => r.id), ...params).map((r) => r.id)
-    );
-    return raw.filter((r) => allowed.has(r.id)).slice(0, k).map((r) => ({ chunkId: r.id, distance: r.distance }));
+    if (!where) return this.knn(embedding, k);
+    try {
+      return this.knn(
+        embedding,
+        k,
+        `AND chunk_id IN (SELECT c.id FROM chunks c JOIN documents d ON d.id = c.document_id WHERE 1=1${where})`,
+        params
+      );
+    } catch {
+      return this.vecSearchWidening(embedding, k, where, params);
+    }
+  }
+  knn(embedding, k, extra = "", params = []) {
+    return this.db.prepare(
+      `SELECT chunk_id id, distance FROM chunk_vec WHERE embedding MATCH ? AND k = ? ${extra} ORDER BY distance`
+    ).all(embedding, BigInt(k), ...params).map((r) => ({ chunkId: r.id, distance: r.distance }));
+  }
+  /** Fallback filtered KNN: fetch k*4, k*16, ... nearest until k pass the filter or all were seen. */
+  vecSearchWidening(embedding, k, where, params) {
+    const total = this.vecCount();
+    for (let fetch2 = k * 4; ; fetch2 *= 4) {
+      const raw = this.knn(embedding, Math.min(fetch2, total));
+      if (!raw.length) return [];
+      const allowed = /* @__PURE__ */ new Set();
+      for (let i = 0; i < raw.length; i += 400) {
+        const ids = raw.slice(i, i + 400).map((r) => r.chunkId);
+        for (const r of this.db.prepare(
+          `SELECT c.id FROM chunks c JOIN documents d ON d.id = c.document_id
+             WHERE c.id IN (${ids.map(() => "?").join(",")})${where}`
+        ).all(...ids, ...params))
+          allowed.add(r.id);
+      }
+      const hits = raw.filter((r) => allowed.has(r.chunkId)).slice(0, k);
+      if (hits.length >= k || fetch2 >= total) return hits;
+    }
   }
   unembeddedChunkIds(limit) {
     return this.db.prepare("SELECT id FROM chunks WHERE embedded = 0 ORDER BY id LIMIT ?").all(limit).map((r) => r.id);
@@ -50529,6 +50605,26 @@ var SqliteAdapter = class _SqliteAdapter {
     }
     return out;
   }
+  graphDependentFileCounts(paths2, kinds) {
+    const out = /* @__PURE__ */ new Map();
+    if (!paths2.length || !kinds.length) return out;
+    const kph = kinds.map(() => "?").join(",");
+    for (let i = 0; i < paths2.length; i += 400) {
+      const batch = paths2.slice(i, i + 400);
+      const pph = batch.map(() => "?").join(",");
+      for (const r of this.db.prepare(
+        `SELECT d.path p, COUNT(DISTINCT s.path) n
+             FROM graph_nodes d
+             JOIN graph_edges e ON e.dst = d.id AND e.kind IN (${kph})
+             JOIN graph_nodes s ON s.id = e.src
+            WHERE d.path IN (${pph}) AND s.path IS NOT NULL AND s.path <> d.path
+            GROUP BY d.path`
+      ).all(...kinds, ...batch)) {
+        out.set(r.p, r.n);
+      }
+    }
+    return out;
+  }
   edgesWhere(col, ids, kinds) {
     if (!ids.length) return [];
     const out = [];
@@ -52664,6 +52760,25 @@ function findOnPath(bin) {
   }
   return void 0;
 }
+var REQUIRE_PREAMBLE = 'import { createRequire } from "node:module"; const require = createRequire(process.env.EMINIAR_PROJECT_ROOT ? process.env.EMINIAR_PROJECT_ROOT + "/package.json" : import.meta.url);';
+function requirePreamble(code) {
+  if (!/\brequire\s*\(/.test(code)) return void 0;
+  if (/\b(?:const|let|var|function)\s+require\b/.test(code) || /\bcreateRequire\b/.test(code)) {
+    return void 0;
+  }
+  return `${REQUIRE_PREAMBLE}
+`;
+}
+function prepareCode(lang, code) {
+  const pre = lang.preamble?.(code);
+  if (!pre) return code;
+  if (code.startsWith("#!")) {
+    const nl = code.indexOf("\n");
+    return nl === -1 ? `${code}
+${pre}` : code.slice(0, nl + 1) + pre + code.slice(nl + 1);
+  }
+  return pre + code;
+}
 function preferredShell() {
   const sh = process.env.SHELL ?? "";
   if (/\/(zsh|bash)$/.test(sh) && existsSync11(sh)) return sh;
@@ -52696,14 +52811,16 @@ var LANGUAGES = [
     aliases: ["js", "node", "mjs"],
     ext: ".mjs",
     detect: () => process.execPath,
-    argv: (bin, file3) => [bin, "--no-warnings", file3]
+    argv: (bin, file3) => [bin, "--no-warnings", file3],
+    preamble: requirePreamble
   },
   {
     id: "typescript",
     aliases: ["ts"],
     ext: ".ts",
     detect: () => process.execPath,
-    argv: (bin, file3) => [bin, "--no-warnings", tsxCli(), file3]
+    argv: (bin, file3) => [bin, "--no-warnings", tsxCli(), file3],
+    preamble: requirePreamble
   },
   {
     id: "python",
@@ -52889,7 +53006,7 @@ async function execute(ctx, input2) {
       );
     }
   }
-  const file3 = writeScratch(input2.code, lang.ext);
+  const file3 = writeScratch(prepareCode(lang, input2.code), lang.ext);
   const env = buildEnv(ctx.config.sandbox.envAllow, {
     EMINIAR_PROJECT_ROOT: ctx.root,
     NODE_PATH: [join17(ctx.root, "node_modules"), process.env.NODE_PATH].filter(Boolean).join(":"),
@@ -53339,10 +53456,22 @@ searchable terms: ${terms.join(", ")}`);
 }
 
 // src/tools/em_batch_execute.ts
-var INLINE_LIMIT = 1024;
+var INLINE_LIMIT_WITH_QUERIES = 1024;
+var DEFAULT_MAX_TOKENS = 2e3;
+function chooseInline(lengths, perOutputCap, budgetChars) {
+  const candidates2 = lengths.map((len, i) => ({ len, i })).filter((c) => c.len <= perOutputCap).sort((a, b) => a.len - b.len);
+  const inline = /* @__PURE__ */ new Set();
+  let remaining = budgetChars;
+  for (const c of candidates2) {
+    if (c.len > remaining) break;
+    inline.add(c.i);
+    remaining -= c.len;
+  }
+  return inline;
+}
 var emBatchExecute = defineTool({
   name: "em_batch_execute",
-  description: "Run several labelled commands in parallel, index every output under its label, and answer `queries` against all of them in the same round trip. Ideal for research: git status + recent commits + test output + dependency list in one call. Small outputs are returned inline; large ones only through the query hits.",
+  description: "Run several labelled commands in parallel, index every output under its label, and answer `queries` against all of them in the same round trip. Ideal for research: git status + recent commits + test output + dependency list in one call. Without `queries`/`intent`, each output up to the index threshold (5 KB) is returned inline, within `max_tokens` overall; with them, only outputs up to 1 KB are inlined and the query hits carry the rest. Every output is indexed; an elided one is fetched with em_search {document_ids: [N]}.",
   schema: {
     commands: external_exports.array(
       external_exports.object({
@@ -53363,6 +53492,9 @@ var emBatchExecute = defineTool({
     const concurrency = args.concurrency ?? 4;
     const results = [];
     let rawChars = 0;
+    const queries = [...args.queries ?? []];
+    if (args.intent && !queries.includes(args.intent)) queries.unshift(args.intent);
+    const maxTokens = args.max_tokens ?? DEFAULT_MAX_TOKENS;
     const queue = [...args.commands];
     const worker = async () => {
       for (; ; ) {
@@ -53384,7 +53516,7 @@ var emBatchExecute = defineTool({
           results.push({
             label: cmd.label,
             header: describeRun(r),
-            inline: r.stdout.length <= INLINE_LIMIT ? r.stdout : void 0,
+            stdout: r.stdout,
             docId: p.documentId,
             stderr: r.exitCode !== 0 ? r.stderr.trim().slice(-800) : void 0
           });
@@ -53396,15 +53528,21 @@ var emBatchExecute = defineTool({
     await Promise.all(Array.from({ length: Math.min(concurrency, queue.length) }, worker));
     const order = new Map(args.commands.map((c, i) => [c.label, i]));
     results.sort((a, b) => (order.get(a.label) ?? 0) - (order.get(b.label) ?? 0));
+    const perOutputCap = queries.length ? INLINE_LIMIT_WITH_QUERIES : ctx.config.sandbox.indexThreshold;
+    const inline = chooseInline(
+      results.map((r) => r.stdout?.length ?? Number.POSITIVE_INFINITY),
+      perOutputCap,
+      maxTokens * 4
+    );
     const lines = [];
-    for (const r of results) {
-      lines.push(`## ${r.label} \u2014 ${r.header}${r.docId ? ` \xB7 doc ${r.docId}` : ""}`);
-      if (r.inline !== void 0) lines.push(r.inline.trimEnd());
+    results.forEach((r, i) => {
+      const shown = inline.has(i) && r.stdout !== void 0;
+      const hint = r.docId && !shown ? ` \xB7 em_search {document_ids: [${r.docId}]} for the output` : "";
+      lines.push(`## ${r.label} \u2014 ${r.header}${r.docId ? ` \xB7 doc ${r.docId}` : ""}${hint}`);
+      if (shown) lines.push(r.stdout.trimEnd());
       if (r.stderr) lines.push(`stderr: ${r.stderr}`);
       lines.push("");
-    }
-    const queries = [...args.queries ?? []];
-    if (args.intent && !queries.includes(args.intent)) queries.unshift(args.intent);
+    });
     const docIds = results.map((r) => r.docId).filter((d) => d !== void 0);
     if (queries.length && docIds.length) {
       const responses = [];
@@ -53414,7 +53552,7 @@ var emBatchExecute = defineTool({
         );
       }
       lines.push("# query hits");
-      lines.push(formatResults(responses, args.max_tokens ?? 2e3));
+      lines.push(formatResults(responses, maxTokens));
     }
     const out = lines.join("\n");
     recordStat(ctx, "em_batch_execute", {
@@ -54476,6 +54614,9 @@ var NAME_RE2 = /^[a-z0-9][a-z0-9._-]{0,63}$/;
 function docsTag(name) {
   return `${DOCS_PREFIX}${name}`;
 }
+function nameFromTag(tag) {
+  return tag.startsWith(DOCS_PREFIX) ? tag.slice(DOCS_PREFIX.length) : void 0;
+}
 function isValidName2(name) {
   return NAME_RE2.test(name);
 }
@@ -56927,7 +57068,7 @@ function removeDocSource(ctx, db, name) {
   const removed = removeSource(ctx, name);
   const documents = db.deleteDocumentsBySource(docsTag(name));
   clearSyncState(db, name);
-  if (removed) syncDocsGraphQuietly(ctx, db);
+  if (removed || documents) syncDocsGraphQuietly(ctx, db);
   return { removed, documents };
 }
 function formatAdd(o) {
@@ -56998,6 +57139,26 @@ function ago(ts) {
   if (s < 172800) return `${Math.round(s / 3600)}h ago`;
   return `${Math.round(s / 86400)}d ago`;
 }
+function orphanedDocSources(db, registered, stats = db.documentSourceStats(DOCS_PREFIX)) {
+  const known = new Set(registered);
+  const out = [];
+  for (const s of stats) {
+    const name = nameFromTag(s.source);
+    if (name === void 0 || known.has(name)) continue;
+    out.push({ name, documents: s.documents, chunks: s.chunks, latest: s.latest });
+  }
+  return out;
+}
+function formatOrphans(orphans2) {
+  const names = orphans2.map((o) => o.name);
+  return [
+    `# Orphaned in the index (${orphans2.length}; pages without a manifest entry)`,
+    ...orphans2.map(
+      (o) => `- ${o.name} \xB7 ${o.documents} doc(s) \xB7 ${o.chunks} chunks \xB7 indexed ${ago(o.latest)}`
+    ),
+    `em_docs {action:"remove", names:${JSON.stringify(names)}} drops them; {action:"add", pick:"recommended"} or {action:"add", name, url|path} re-registers.`
+  ];
+}
 function formatDiscovery(d, max = 40) {
   const lines = [];
   lines.push(
@@ -57145,19 +57306,30 @@ var emDocs = defineTool({
         case "remove": {
           const names = [...args.name ? [args.name] : [], ...args.names ?? []];
           if (!names.length) return errorResult("remove needs `name` or `names`");
+          let unknown2 = 0;
           const lines = names.map((n) => {
             const r = removeDocSource(ctx, db, n);
-            return r.removed ? `removed ${n} (${r.documents} document(s) dropped)` : `${n}: not registered${r.documents ? ` (${r.documents} orphan document(s) dropped)` : ""}`;
+            if (r.removed) return `removed ${n} (${r.documents} document(s) dropped)`;
+            if (r.documents) return `removed ${n} (orphaned; ${r.documents} document(s) dropped)`;
+            unknown2++;
+            return `${n}: not registered and not in the index; em_docs {action:"list"} shows both`;
           });
-          return done(lines.join("\n"));
+          const out = lines.join("\n");
+          return unknown2 === names.length ? errorResult(out) : done(out);
         }
         case "list": {
           const m = readManifest(ctx);
-          if (!m.sources.length)
-            return done(
-              `no documentation sources registered (${manifestPath(ctx)}). em_docs {action:"discover"} finds local docs and the libraries, services and APIs this project uses.`
-            );
-          const stats = new Map(db.documentSourceStats(DOCS_PREFIX).map((s) => [s.source, s]));
+          const sourceStats = db.documentSourceStats(DOCS_PREFIX);
+          const orphans2 = orphanedDocSources(
+            db,
+            m.sources.map((s) => s.name),
+            sourceStats
+          );
+          if (!m.sources.length) {
+            const none = `no documentation sources registered (${manifestPath(ctx)}). em_docs {action:"discover"} finds local docs and the libraries, services and APIs this project uses.`;
+            return done(orphans2.length ? [none, "", ...formatOrphans(orphans2)].join("\n") : none);
+          }
+          const stats = new Map(sourceStats.map((s) => [s.source, s]));
           const catalog = mergedCatalog(ctx.config.docs.catalog);
           const categorised = m.sources.map((s) => ({
             s,
@@ -57188,9 +57360,10 @@ var emDocs = defineTool({
               `- ${s.name} [${s.kind}]${s.provider ? ` \u2190 ${s.provider}` : ""}${s.tags?.length ? ` (${s.tags.join(", ")})` : ""} \xB7 ${st?.documents ?? 0} doc(s) \xB7 ${st?.chunks ?? 0} chunks \xB7 ${state} \xB7 ${where}`
             );
           }
+          if (orphans2.length) lines.push("", ...formatOrphans(orphans2));
           lines.push(
             "",
-            `${m.sources.length} source(s); em_docs {action:"lookup", query} searches them all, {action:"sync"} refreshes.`
+            `${m.sources.length} source(s)${orphans2.length ? ` + ${orphans2.length} orphaned` : ""}; em_docs {action:"lookup", query} searches them all, {action:"sync"} refreshes.`
           );
           return done(lines.join("\n"));
         }
@@ -57236,10 +57409,18 @@ ${r.urls.map((u) => `  ${u}`).join("\n")}
             raw += r.hits.reduce((n, h) => n + h.chunk.body.length, 0);
             responses.push(r);
           }
-          const out = formatResults(
-            responses,
-            args.max_tokens ?? ctx.config.search.defaultMaxTokens
+          let out = formatResults(responses, args.max_tokens ?? ctx.config.search.defaultMaxTokens);
+          const hitSources = new Set(
+            responses.flatMap((r) => r.hits.map((h) => h.chunk.document.source))
           );
+          const orphaned = orphanedDocSources(
+            db,
+            m.sources.map((s) => s.name)
+          ).filter((o) => hitSources.has(docsTag(o.name)));
+          if (orphaned.length)
+            out += `
+
+note: ${orphaned.map((o) => o.name).join(", ")} ${orphaned.length === 1 ? "is" : "are"} not in the manifest (orphaned pages); em_docs {action:"list"} explains`;
           recordStat(ctx, "em_docs", {
             rawChars: raw,
             returnedChars: out.length,
@@ -57470,7 +57651,7 @@ var LANG_ENUM = external_exports.string().describe(
 );
 var emExecute = defineTool({
   name: "em_execute",
-  description: "Run code in a sandbox inside the project root and return only stdout. Output over the index threshold (5 KB) is indexed; pass `intent` to get back just the matching sections plus a vocabulary of searchable terms. Use this instead of Bash for anything that prints a lot: git log/diff, test runs, dependency trees, logs, API calls. Shell snippets honour Claude Code's Bash deny rules.",
+  description: "Run code in a sandbox inside the project root and return only stdout. Output over the index threshold (5 KB) is indexed; pass `intent` to get back just the matching sections plus a vocabulary of searchable terms. Use this instead of Bash for anything that prints a lot: git log/diff, test runs, dependency trees, logs, API calls. Shell snippets honour Claude Code's Bash deny rules. javascript and typescript run as ES modules; `import` and `require` both work.",
   schema: {
     language: LANG_ENUM,
     code: external_exports.string().min(1).describe("Program or shell script to run"),
@@ -57552,7 +57733,7 @@ function refuse(check3, target) {
 // src/tools/em_execute_file.ts
 var emExecuteFile = defineTool({
   name: "em_execute_file",
-  description: "Process a file in the sandbox without reading it into context. Your code receives the absolute path as $EM_FILE (shell) / process.env.EM_FILE (node) / os.environ['EM_FILE'] (python); print only what you need. The file must be inside the project root unless a Read(<path>) allow rule opts it in; a matching Read deny or ask rule refuses the path, inside the root or out. Omit `code` to get a quick summary (size, lines, head).",
+  description: "Process a file in the sandbox without reading it into context. Your code receives the absolute path as $EM_FILE (shell) / process.env.EM_FILE (node) / os.environ['EM_FILE'] (python); print only what you need. The file must be inside the project root unless a Read(<path>) allow rule opts it in; a matching Read deny or ask rule refuses the path, inside the root or out. Omit `code` to get a quick summary (size, lines, head). javascript and typescript run as ES modules; `import` and `require` both work.",
   schema: {
     path: external_exports.string().describe("File to process, relative to the project root or absolute"),
     language: LANG_ENUM.optional(),
@@ -58528,7 +58709,7 @@ var emIndex = defineTool({
       });
       kickEmbedding(ctx, db);
       return text(
-        `${r.created ? "indexed" : "unchanged"} "${args.title}" as doc ${r.documentId}: ${r.chunks} chunks, ${r.chars} chars (${r.kind})`
+        `${r.created ? "indexed" : "unchanged"} "${args.title}" as doc ${r.documentId}: ${r.chunks} chunks, ${r.chars} chars (${r.kind})${r.fallback ? " (no sections found; indexed as one chunk)" : ""}`
       );
     }
     if (!args.path) return errorResult("provide `path` or `content`");
@@ -58538,7 +58719,7 @@ var emIndex = defineTool({
       recursive: args.recursive ?? true
     });
     const lines = [
-      `${args.path}: ${s.files} file(s), ${s.indexed} indexed, ${s.unchanged} unchanged, ${s.chunks} chunks`
+      `${args.path}: ${s.files} file(s), ${s.indexed} indexed, ${s.unchanged} unchanged, ${s.chunks} chunks${s.fallbacks ? ` (${s.fallbacks} with no sections found; indexed as one chunk)` : ""}`
     ];
     for (const sk of s.skipped.slice(0, 10)) lines.push(`  skipped ${sk.path}: ${sk.reason}`);
     if (s.skipped.length > 10) lines.push(`  \u2026 ${s.skipped.length - 10} more skipped`);
@@ -58914,7 +59095,10 @@ var emSearch = defineTool({
       }
     }
     const out = formatResults(responses, args.max_tokens ?? ctx.config.search.defaultMaxTokens);
-    const note = mode !== "lexical" && mode !== "regex" && !vectorSearch && ctx.config.embeddings.enabled ? "\n\n(embeddings: model still loading \u2014 lexical results only this time)" : "";
+    let note = mode !== "lexical" && mode !== "regex" && !vectorSearch && ctx.config.embeddings.enabled ? "\n\n(embeddings: model still loading \u2014 lexical results only this time)" : "";
+    const sources = Array.isArray(args.source) ? args.source : args.source ? [args.source] : [];
+    if (mode === "semantic" && !ctx.config.docs.embed && sources.some((s) => s.startsWith("docs:")) && responses.every((r) => r.hits.length === 0))
+      note += '\n\n(docs:* pages are indexed without vectors because docs.embed is off; use mode "hybrid" or "lexical", or set docs.embed: true in eminiar.config.json and em_docs {action:"sync", force:true})';
     recordStat(ctx, "em_search", {
       rawChars: responses.reduce(
         (a, r) => a + r.hits.reduce((b, h) => b + h.chunk.body.length, 0),
@@ -58933,6 +59117,7 @@ init_esm_shims();
 // src/session/recall.ts
 init_esm_shims();
 init_ids();
+init_metrics();
 init_model();
 var RECALL_CATEGORY = "recall";
 function recallStats(events) {
@@ -60875,12 +61060,17 @@ function usageLine(ctx, sessionId) {
   if (!usageEnabled(ctx)) return void 0;
   try {
     const db = getStorage(ctx);
-    ingestSession(db, ctx.root, sessionId, void 0, { overrides: ctx.config.usage.pricing });
+    let failed;
+    try {
+      ingestSession(db, ctx.root, sessionId, void 0, { overrides: ctx.config.usage.pricing });
+    } catch (err) {
+      failed = `usage: ingest failed: ${err.message}`;
+    }
     const b = usageBrief(db, ctx, { sessionId });
-    if (!b.todayUsd && !b.sessionUsd) return void 0;
-    return `usage: today ${formatUsd(b.todayUsd)} \xB7 session ${formatUsd(b.sessionUsd)} \xB7 see em_usage`;
-  } catch {
-    return void 0;
+    const line = b.todayUsd || b.sessionUsd ? `usage: today ${formatUsd(b.todayUsd)} \xB7 session ${formatUsd(b.sessionUsd)} \xB7 see em_usage` : void 0;
+    return [line, failed].filter((s) => s).join("\n") || void 0;
+  } catch (err) {
+    return `usage: unavailable: ${err.message}`;
   }
 }
 function mergeByTool(rows) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "eminiar",
-  "version": "0.10.0",
+  "version": "0.10.1",
   "description": "Local-first context, memory and knowledge-graph plugin for Claude Code. Sandboxed execution, FTS5 + vector search, session continuity, and a code graph — nothing leaves your machine.",
   "type": "module",
   "license": "MIT",

package/skills/em-docs/SKILL.md

@@ -20,6 +20,8 @@ Act on `$ARGUMENTS` with `em_docs`; with no arguments, run `list` and, when it i
   one source with `name`. Cite the `doc N` ids. Only when the answer is not in the index should
   you fall back to `em_fetch_and_index` for a specific page, then `{action: "add", url}` if it is
   worth keeping.
-- **list** / **remove** → the matching action.
+- **list** / **remove** → the matching action. `list` also shows sources *orphaned in the
+  index* (pages indexed under `docs:<name>` with no manifest entry, e.g. the manifest lives on
+  another branch); `remove` drops their pages, `add` with the same name re-registers them.
 
 The manifest `.eminiar/docs.json` is meant to be committed; say so once when sources were added.