@a13xu/lucid 1.19.0 → 1.21.0

package/build/index.js

@@ -28,6 +28,7 @@ import { handleCompressText, CompressTextSchema } from "./tools/compress.js";
 import { handleBackupFile, BackupFileSchema, handleRestoreFile, RestoreFileSchema, handleCheckTruncateRisk, CheckTruncateRiskSchema, } from "./tools/backup.js";
 import { handleSessionStatus, SessionStatusSchema } from "./tools/session.js";
 import { handleDelegateLocal, DelegateLocalSchema, handleLocalLlmStatus, LocalLlmStatusSchema, } from "./tools/delegate-local.js";
+import { handleIngestBook, IngestBookSchema, handleGenerateBookSkill, GenerateBookSkillSchema, handleListBooks, ListBooksSchema, runBookCli, } from "./tools/book.js";
 import { loadLocalConfig } from "./local-llm/config.js";
 // ---------------------------------------------------------------------------
 // CLI mode: lucid watch | lucid status | lucid stop
@@ -50,6 +51,13 @@ if (_cliCmd === "local") {
     const exitCode = await runLocalLlmCli(_cliArgs);
     process.exit(exitCode);
 }
+if (_cliCmd === "book") {
+    const { initDatabase, prepareStatements } = await import("./database.js");
+    const bookDb = initDatabase();
+    const bookStmts = prepareStatements(bookDb);
+    const exitCode = await runBookCli(_cliArgs, bookStmts);
+    process.exit(exitCode);
+}
 // ---------------------------------------------------------------------------
 // `lucid guard <subcmd>` — invoked from Claude Code hooks (PreToolUse, etc.)
 //
@@ -108,14 +116,13 @@ async function guardPreEdit(stmts, flagArgs) {
                 toolName = payload.tool_name ?? "Write";
                 const ti = payload.tool_input ?? {};
                 path = ti.file_path ?? ti.path;
-                if (typeof ti.content === "string")
+                // Only `Write` carries the full new file content. `Edit`/`MultiEdit`
+                // carry replacement fragments, not the full post-write state — assessing
+                // shrinkage on those produces false MAJOR_SHRINK blocks on every edit
+                // of a non-tiny file. Leave content=null so assessTruncate skips the
+                // size-based rules and only the cascade lock can apply.
+                if (toolName === "Write" && typeof ti.content === "string") {
                     content = ti.content;
-                else if (Array.isArray(ti.edits)) {
-                    // MultiEdit — sum up final state crudely: use new_strings concatenated.
-                    content = ti.edits.map((e) => e.new_string ?? "").join("\n");
-                }
-                else if (typeof ti.new_string === "string") {
-                    content = ti.new_string;
                 }
             }
             catch {
@@ -670,6 +677,29 @@ server.registerTool("plan_update_task", {
     inputSchema: PlanUpdateTaskSchema.shape,
 }, tx("plan_update_task", (args) => handlePlanUpdateTask(stmts, args)));
 // ---------------------------------------------------------------------------
+// Tools — Book Ingestion (PDF/EPUB/DOCX → markdown chunks → skill router)
+// ---------------------------------------------------------------------------
+server.registerTool("ingest_book", {
+    title: "Ingest Book",
+    description: "Convert a book (PDF, EPUB, DOCX, or Markdown) into chunked markdown files " +
+        "under ./books/<slug>/ and index every chunk into Lucid. Requires user-installed " +
+        "converter (pymupdf4llm for PDF, pandoc for EPUB/DOCX). Pair with generate_book_skill " +
+        "to make the corpus auto-load as a Claude Code skill.",
+    inputSchema: IngestBookSchema.shape,
+}, tx("ingest_book", (args) => handleIngestBook(stmts, args)));
+server.registerTool("generate_book_skill", {
+    title: "Generate Book Skill",
+    description: "Emit a thin SKILL.md router into ~/.claude/skills/book-<slug>/ (or .claude/skills/ " +
+        "for project scope). The skill auto-loads (~100 tokens) when its trigger topics come up " +
+        "and delegates retrieval to smart_context. Run after ingest_book.",
+    inputSchema: GenerateBookSkillSchema.shape,
+}, tx("generate_book_skill", (args) => handleGenerateBookSkill(args)));
+server.registerTool("list_books", {
+    title: "List Books",
+    description: "List ingested books under ./books/ with chunk counts and ingestion dates.",
+    inputSchema: ListBooksSchema.shape,
+}, tx("list_books", (args) => handleListBooks(args)));
+// ---------------------------------------------------------------------------
 // Tools — Updater
 // ---------------------------------------------------------------------------
 server.registerTool("update_lucid", {

package/build/tools/book.d.ts

@@ -0,0 +1,58 @@
+/**
+ * Book ingestion pipeline — PDF/EPUB → Markdown chunks → Lucid index → Claude
+ * Code Skill router. Conversion shells out to user-installed tools
+ * (pymupdf4llm for PDF, pandoc for EPUB); Lucid itself stays dependency-free.
+ *
+ * Three handlers:
+ *   - handleIngestBook       : convert + chunk + index
+ *   - handleGenerateBookSkill: emit ~/.claude/skills/book-<slug>/SKILL.md
+ *   - handleListBooks        : list indexed books
+ */
+import { z } from "zod";
+import type { Statements } from "../database.js";
+export declare const IngestBookSchema: z.ZodObject<{
+    path: z.ZodString;
+    title: z.ZodOptional<z.ZodString>;
+    out_dir: z.ZodOptional<z.ZodString>;
+    chunker: z.ZodDefault<z.ZodOptional<z.ZodEnum<["heading", "page", "none"]>>>;
+    index: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
+}, "strip", z.ZodTypeAny, {
+    path: string;
+    chunker: "none" | "page" | "heading";
+    index: boolean;
+    title?: string | undefined;
+    out_dir?: string | undefined;
+}, {
+    path: string;
+    title?: string | undefined;
+    out_dir?: string | undefined;
+    chunker?: "none" | "page" | "heading" | undefined;
+    index?: boolean | undefined;
+}>;
+export declare const GenerateBookSkillSchema: z.ZodObject<{
+    slug: z.ZodString;
+    title: z.ZodOptional<z.ZodString>;
+    topics: z.ZodOptional<z.ZodString>;
+    scope: z.ZodDefault<z.ZodOptional<z.ZodEnum<["user", "project"]>>>;
+}, "strip", z.ZodTypeAny, {
+    slug: string;
+    scope: "project" | "user";
+    title?: string | undefined;
+    topics?: string | undefined;
+}, {
+    slug: string;
+    title?: string | undefined;
+    topics?: string | undefined;
+    scope?: "project" | "user" | undefined;
+}>;
+export declare const ListBooksSchema: z.ZodObject<{
+    dir: z.ZodOptional<z.ZodString>;
+}, "strip", z.ZodTypeAny, {
+    dir?: string | undefined;
+}, {
+    dir?: string | undefined;
+}>;
+export declare function handleIngestBook(stmts: Statements, args: z.infer<typeof IngestBookSchema>): string;
+export declare function handleGenerateBookSkill(args: z.infer<typeof GenerateBookSkillSchema>): string;
+export declare function handleListBooks(args: z.infer<typeof ListBooksSchema>): string;
+export declare function runBookCli(args: string[], stmts: Statements): Promise<number>;

package/build/tools/book.js

@@ -0,0 +1,446 @@
+/**
+ * Book ingestion pipeline — PDF/EPUB → Markdown chunks → Lucid index → Claude
+ * Code Skill router. Conversion shells out to user-installed tools
+ * (pymupdf4llm for PDF, pandoc for EPUB); Lucid itself stays dependency-free.
+ *
+ * Three handlers:
+ *   - handleIngestBook       : convert + chunk + index
+ *   - handleGenerateBookSkill: emit ~/.claude/skills/book-<slug>/SKILL.md
+ *   - handleListBooks        : list indexed books
+ */
+import { z } from "zod";
+import { existsSync, mkdirSync, readFileSync, writeFileSync, readdirSync, } from "fs";
+import { join, resolve, extname, basename } from "path";
+import { homedir } from "os";
+import { spawnSync } from "child_process";
+import { indexFile, upsertFileIndex } from "../indexer/file.js";
+// ---------------------------------------------------------------------------
+// Schemas
+// ---------------------------------------------------------------------------
+export const IngestBookSchema = z.object({
+    path: z.string().min(1).describe("Path to the source book (.pdf, .epub, .docx, or .md)."),
+    title: z.string().optional().describe("Display title. Defaults to the filename."),
+    out_dir: z.string().optional().describe("Output directory for chunked markdown. Defaults to ./books/<slug>/"),
+    chunker: z.enum(["heading", "page", "none"]).optional().default("heading")
+        .describe("How to split: by H1/H2 headings (default), by source page, or single file."),
+    index: z.boolean().optional().default(true).describe("Index chunks into Lucid immediately."),
+});
+export const GenerateBookSkillSchema = z.object({
+    slug: z.string().min(1).describe("Book slug (the directory name produced by ingest_book)."),
+    title: z.string().optional().describe("Display title shown in the skill description."),
+    topics: z.string().optional().describe("Comma-separated topics that should trigger the skill."),
+    scope: z.enum(["user", "project"]).optional().default("user")
+        .describe("Where to install the skill: ~/.claude/skills/ (user) or .claude/skills/ (project)."),
+});
+export const ListBooksSchema = z.object({
+    dir: z.string().optional().describe("Root directory to scan. Defaults to ./books/"),
+});
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+function slugify(input) {
+    return input
+        .toLowerCase()
+        .normalize("NFKD")
+        .replace(/[̀-ͯ]/g, "")
+        .replace(/[^a-z0-9]+/g, "-")
+        .replace(/^-+|-+$/g, "")
+        .slice(0, 60) || "book";
+}
+function which(cmd) {
+    const probe = process.platform === "win32" ? "where" : "which";
+    const r = spawnSync(probe, [cmd], { stdio: "ignore" });
+    return r.status === 0;
+}
+// ---------------------------------------------------------------------------
+// Converters — shell out to user tools; never bundle binaries
+// ---------------------------------------------------------------------------
+function convertPdfToMarkdown(src) {
+    // Preferred: pymupdf4llm (fastest for native-text PDFs, no GPU, no ML)
+    if (which("python") || which("python3")) {
+        const py = which("python3") ? "python3" : "python";
+        const script = `import sys\n` +
+            `try:\n` +
+            `  import pymupdf4llm\n` +
+            `  sys.stdout.write(pymupdf4llm.to_markdown(sys.argv[1]))\n` +
+            `except ImportError:\n` +
+            `  sys.stderr.write("pymupdf4llm not installed\\n")\n` +
+            `  sys.exit(2)\n`;
+        const r = spawnSync(py, ["-c", script, src], { encoding: "utf-8", maxBuffer: 200 * 1024 * 1024 });
+        if (r.status === 0 && r.stdout)
+            return r.stdout;
+    }
+    // Fallback: marker_single (datalab-to/marker)
+    if (which("marker_single")) {
+        const tmpDir = join(homedir(), ".lucid", "tmp-marker", `${Date.now()}`);
+        mkdirSync(tmpDir, { recursive: true });
+        const r = spawnSync("marker_single", [src, "--output_format", "markdown", "--output_dir", tmpDir], { encoding: "utf-8" });
+        if (r.status === 0) {
+            const produced = readdirSync(tmpDir, { recursive: true })
+                .filter((p) => typeof p === "string" && p.endsWith(".md"));
+            if (produced.length > 0)
+                return readFileSync(join(tmpDir, produced[0]), "utf-8");
+        }
+    }
+    throw new Error(`Cannot convert PDF — install one of:\n` +
+        `  pip install pymupdf4llm   # fast, native PDFs (recommended)\n` +
+        `  pip install marker-pdf    # best quality, supports scans`);
+}
+function convertEpubToMarkdown(src) {
+    if (which("pandoc")) {
+        const r = spawnSync("pandoc", ["-f", "epub", "-t", "gfm", src], { encoding: "utf-8", maxBuffer: 200 * 1024 * 1024 });
+        if (r.status === 0 && r.stdout)
+            return r.stdout;
+    }
+    throw new Error(`Cannot convert EPUB — install pandoc: https://pandoc.org/installing.html`);
+}
+function convertDocxToMarkdown(src) {
+    if (which("pandoc")) {
+        const r = spawnSync("pandoc", ["-f", "docx", "-t", "gfm", src], { encoding: "utf-8", maxBuffer: 200 * 1024 * 1024 });
+        if (r.status === 0 && r.stdout)
+            return r.stdout;
+    }
+    throw new Error(`Cannot convert DOCX — install pandoc.`);
+}
+function loadSource(path) {
+    const ext = extname(path).toLowerCase();
+    if (!existsSync(path))
+        throw new Error(`Source not found: ${path}`);
+    if (ext === ".md" || ext === ".markdown" || ext === ".txt")
+        return readFileSync(path, "utf-8");
+    if (ext === ".pdf")
+        return convertPdfToMarkdown(path);
+    if (ext === ".epub")
+        return convertEpubToMarkdown(path);
+    if (ext === ".docx")
+        return convertDocxToMarkdown(path);
+    throw new Error(`Unsupported source format: ${ext}. Use .pdf, .epub, .docx, .md, or .txt.`);
+}
+// ---------------------------------------------------------------------------
+// Chunking
+// ---------------------------------------------------------------------------
+function chunkByHeading(md) {
+    // Split on H1 (#) primarily; if a single chunk grows beyond 80KB, fall back to H2.
+    const HARD_LIMIT = 80 * 1024;
+    const lines = md.split(/\r?\n/);
+    const out = [];
+    let cur = null;
+    const push = () => {
+        if (cur && cur.buf.length > 0) {
+            out.push({ index: out.length + 1, title: cur.title, content: cur.buf.join("\n").trim() });
+        }
+    };
+    for (const line of lines) {
+        const h1 = /^# +(.+)$/.exec(line);
+        if (h1) {
+            push();
+            cur = { title: h1[1].trim(), buf: [line] };
+            continue;
+        }
+        if (!cur)
+            cur = { title: "Preface", buf: [] };
+        cur.buf.push(line);
+        // Soft split on H2 if the current chunk is getting too big
+        const h2 = /^## +(.+)$/.exec(line);
+        if (h2 && cur.buf.join("\n").length > HARD_LIMIT) {
+            // Drop the trailing line, push, restart with the H2 as new chunk title.
+            cur.buf.pop();
+            push();
+            cur = { title: h2[1].trim(), buf: [line] };
+        }
+    }
+    push();
+    // If no headings were found at all → fall back to a single chunk
+    if (out.length === 0)
+        out.push({ index: 1, title: "Body", content: md.trim() });
+    return out;
+}
+function chunkByPage(md) {
+    // pymupdf4llm separates pages with -----\n; treat each as a chunk
+    const parts = md.split(/\n-{3,}\n/);
+    return parts
+        .map((p, i) => ({ index: i + 1, title: `Page ${i + 1}`, content: p.trim() }))
+        .filter((c) => c.content.length > 0);
+}
+function chunkAll(md, mode) {
+    if (mode === "none")
+        return [{ index: 1, title: "Full", content: md.trim() }];
+    if (mode === "page")
+        return chunkByPage(md);
+    return chunkByHeading(md);
+}
+// ---------------------------------------------------------------------------
+// handleIngestBook
+// ---------------------------------------------------------------------------
+export function handleIngestBook(stmts, args) {
+    const src = resolve(args.path);
+    const title = args.title ?? basename(src, extname(src));
+    const slug = slugify(title);
+    const outDir = resolve(args.out_dir ?? join(process.cwd(), "books", slug));
+    mkdirSync(outDir, { recursive: true });
+    const raw = loadSource(src);
+    const chunks = chunkAll(raw, args.chunker ?? "heading");
+    // Write a manifest so generate_book_skill knows what was produced
+    const manifest = {
+        slug, title,
+        source: src,
+        chunker: args.chunker ?? "heading",
+        chunks: chunks.map((c) => ({ index: c.index, title: c.title, file: chunkFilename(c) })),
+        created_at: new Date().toISOString(),
+    };
+    // Write chunks
+    for (const c of chunks) {
+        const file = join(outDir, chunkFilename(c));
+        const header = `---\nbook: ${title}\nslug: ${slug}\nchunk: ${c.index}\ntitle: ${c.title.replace(/"/g, '\\"')}\n---\n\n`;
+        writeFileSync(file, header + c.content, "utf-8");
+    }
+    writeFileSync(join(outDir, "_manifest.json"), JSON.stringify(manifest, null, 2), "utf-8");
+    // Index into Lucid via existing pipeline (sync_file is .ts/.py only — call indexer directly)
+    let indexed = 0;
+    if (args.index !== false) {
+        for (const c of chunks) {
+            const file = join(outDir, chunkFilename(c));
+            const idx = indexFile(file);
+            if (!idx)
+                continue;
+            const r = upsertFileIndex(idx, readFileSync(file, "utf-8"), stmts);
+            if (r.stored)
+                indexed++;
+        }
+    }
+    return [
+        `📚 Ingested: "${title}" → ${outDir}`,
+        `   chunks: ${chunks.length} (${args.chunker ?? "heading"})`,
+        args.index !== false ? `   indexed: ${indexed} new chunk(s) into Lucid` : `   indexing skipped`,
+        ``,
+        `Next: generate the auto-loaded skill router with:`,
+        `  lucid book skill ${slug} --topics "<comma,separated,topics>"`,
+    ].join("\n");
+}
+function chunkFilename(c) {
+    const safe = slugify(c.title);
+    return `${String(c.index).padStart(3, "0")}-${safe}.md`;
+}
+// ---------------------------------------------------------------------------
+// handleGenerateBookSkill
+// ---------------------------------------------------------------------------
+export function handleGenerateBookSkill(args) {
+    const slug = args.slug;
+    const booksRoot = join(process.cwd(), "books", slug);
+    const manifestPath = join(booksRoot, "_manifest.json");
+    let title = args.title ?? slug;
+    let chunkCount = 0;
+    let chunkTitles = [];
+    if (existsSync(manifestPath)) {
+        try {
+            const m = JSON.parse(readFileSync(manifestPath, "utf-8"));
+            title = args.title ?? m.title;
+            chunkCount = m.chunks.length;
+            chunkTitles = m.chunks.map((c) => c.title).slice(0, 12);
+        }
+        catch { /* ignore — manifest optional */ }
+    }
+    const topics = (args.topics ?? "").split(",").map((t) => t.trim()).filter(Boolean);
+    const skillRoot = args.scope === "project"
+        ? join(process.cwd(), ".claude", "skills", `book-${slug}`)
+        : join(homedir(), ".claude", "skills", `book-${slug}`);
+    mkdirSync(skillRoot, { recursive: true });
+    const description = topics.length > 0
+        ? `Use when the user asks about ${topics.join(", ")} — or any concept from "${title}". Retrieves passages via Lucid smart_context.`
+        : `Use for questions whose answers should cite "${title}". Retrieves passages via Lucid smart_context.`;
+    const tocPreview = chunkTitles.length > 0
+        ? chunkTitles.map((t, i) => `   ${String(i + 1).padStart(2, "0")}. ${t}`).join("\n")
+        : "   (run `lucid book ingest` first to populate the table of contents)";
+    const body = `---
+name: book-${slug}
+description: ${description}
+---
+
+# ${title} — Lucid skill router
+
+This skill does NOT contain the book text. It delegates retrieval to Lucid's
+indexed corpus so passages are pulled on demand instead of being loaded into
+every prompt.
+
+## When invoked
+
+1. Identify the user's question or the topic at hand.
+2. Call \`mcp__lucid__smart_context\` with:
+   - \`query\`: the user's question, or the topic phrased as a search
+   - \`task_type\`: "moderate" for explanations, "complex" for multi-chapter synthesis
+3. The retrieved chunks come from \`books/${slug}/\` (${chunkCount} indexed chunk${chunkCount === 1 ? "" : "s"}).
+4. Quote chapter / chunk titles when you cite, so the source stays verifiable.
+5. After answering, call \`mcp__lucid__reward\` if the passages were on-point
+   so future queries on this topic rank them higher (decay half-life ~14 days).
+
+## Table of contents (first ${Math.min(chunkTitles.length, 12)} chunks)
+
+${tocPreview}
+
+## Notes
+
+- If \`smart_context\` returns nothing, the corpus may not be indexed. Run
+  \`lucid book ingest <source>\` then \`sync_project\` to rebuild.
+- This skill is a thin router (~100 tokens at scan time). Adding more books
+  here does not bloat Claude Code's startup.
+`;
+    const skillFile = join(skillRoot, "SKILL.md");
+    writeFileSync(skillFile, body, "utf-8");
+    return [
+        `🧠 Skill generated: ${skillFile}`,
+        `   description trigger: ${description}`,
+        `   scope: ${args.scope ?? "user"}`,
+        chunkCount > 0 ? `   linked to ${chunkCount} chunk(s) in books/${slug}/` : `   ⚠️  no chunks indexed yet`,
+    ].join("\n");
+}
+// ---------------------------------------------------------------------------
+// handleListBooks
+// ---------------------------------------------------------------------------
+export function handleListBooks(args) {
+    const root = resolve(args.dir ?? join(process.cwd(), "books"));
+    if (!existsSync(root))
+        return `No books directory at ${root}. Run \`lucid book ingest\` first.`;
+    const entries = readdirSync(root, { withFileTypes: true })
+        .filter((d) => d.isDirectory())
+        .map((d) => d.name);
+    if (entries.length === 0)
+        return `No books indexed under ${root}.`;
+    const lines = [`📚 Books in ${root}:`];
+    for (const name of entries) {
+        const manifestPath = join(root, name, "_manifest.json");
+        if (existsSync(manifestPath)) {
+            try {
+                const m = JSON.parse(readFileSync(manifestPath, "utf-8"));
+                lines.push(`  • ${name} — "${m.title}" (${m.chunks.length} chunks, ingested ${m.created_at.slice(0, 10)})`);
+            }
+            catch {
+                lines.push(`  • ${name} — (manifest unreadable)`);
+            }
+        }
+        else {
+            const mdCount = readdirSync(join(root, name)).filter((f) => f.endsWith(".md")).length;
+            lines.push(`  • ${name} — ${mdCount} .md file(s), no manifest`);
+        }
+    }
+    return lines.join("\n");
+}
+// ---------------------------------------------------------------------------
+// CLI entry — `lucid book <subcmd>`
+// ---------------------------------------------------------------------------
+export async function runBookCli(args, stmts) {
+    const sub = args[0];
+    if (!sub || sub === "--help" || sub === "-h" || sub === "help") {
+        process.stdout.write(BOOK_HELP);
+        return 0;
+    }
+    if (sub === "ingest") {
+        const rest = args.slice(1);
+        if (rest.includes("--help") || rest.includes("-h") || rest.length === 0) {
+            process.stdout.write(INGEST_HELP);
+            return rest.length === 0 ? 64 : 0;
+        }
+        const path = rest.find((a) => !a.startsWith("--"));
+        if (!path) {
+            process.stderr.write("Missing source path.\n");
+            return 64;
+        }
+        const opts = {
+            path,
+            title: getFlag(rest, "--title"),
+            out_dir: getFlag(rest, "--out"),
+            chunker: getFlag(rest, "--chunker") ?? "heading",
+            index: !rest.includes("--no-index"),
+        };
+        process.stdout.write(handleIngestBook(stmts, IngestBookSchema.parse(opts)) + "\n");
+        return 0;
+    }
+    if (sub === "skill") {
+        const rest = args.slice(1);
+        if (rest.includes("--help") || rest.includes("-h") || rest.length === 0) {
+            process.stdout.write(SKILL_HELP);
+            return rest.length === 0 ? 64 : 0;
+        }
+        const slug = rest.find((a) => !a.startsWith("--"));
+        if (!slug) {
+            process.stderr.write("Missing book slug.\n");
+            return 64;
+        }
+        const opts = {
+            slug,
+            title: getFlag(rest, "--title"),
+            topics: getFlag(rest, "--topics"),
+            scope: getFlag(rest, "--scope") ?? "user",
+        };
+        process.stdout.write(handleGenerateBookSkill(GenerateBookSkillSchema.parse(opts)) + "\n");
+        return 0;
+    }
+    if (sub === "list") {
+        const rest = args.slice(1);
+        if (rest.includes("--help") || rest.includes("-h")) {
+            process.stdout.write(LIST_HELP);
+            return 0;
+        }
+        process.stdout.write(handleListBooks({ dir: getFlag(rest, "--dir") }) + "\n");
+        return 0;
+    }
+    process.stderr.write(`Unknown subcommand: ${sub}\n\n${BOOK_HELP}`);
+    return 64;
+}
+function getFlag(args, name) {
+    const i = args.indexOf(name);
+    return i >= 0 && i + 1 < args.length ? args[i + 1] : undefined;
+}
+// ---------------------------------------------------------------------------
+// Help text
+// ---------------------------------------------------------------------------
+const BOOK_HELP = `lucid book — convert books to indexed markdown + auto-load as Claude Code skills
+
+USAGE
+  lucid book <command> [options]
+
+COMMANDS
+  ingest <path>     Convert a PDF/EPUB/DOCX/MD into chunked markdown and index it.
+  skill  <slug>     Emit a SKILL.md router into ~/.claude/skills/ for a book.
+  list              List ingested books and their chunk counts.
+
+Run \`lucid book <command> --help\` for command-specific options.
+
+EXAMPLES
+  lucid book ingest ./clean-code.pdf --title "Clean Code"
+  lucid book skill clean-code --topics "naming,refactor,functions,code review"
+  lucid book list
+
+DEPENDENCIES (install only what you need)
+  PDF   pip install pymupdf4llm        # native-text PDFs, fastest
+        pip install marker-pdf         # OCR-quality, supports scans
+  EPUB  pandoc                         # https://pandoc.org/installing.html
+  DOCX  pandoc
+`;
+const INGEST_HELP = `lucid book ingest <path> [options]
+
+Convert a book into chunked markdown and index it in Lucid.
+
+OPTIONS
+  --title TEXT       Display title. Defaults to filename without extension.
+  --out  DIR         Output directory. Defaults to ./books/<slug>/
+  --chunker MODE     heading (default) | page | none
+  --no-index         Skip indexing into Lucid (just produce markdown).
+
+OUTPUT
+  ./books/<slug>/001-<title>.md, 002-<title>.md, ..., _manifest.json
+`;
+const SKILL_HELP = `lucid book skill <slug> [options]
+
+Generate a SKILL.md router that auto-loads when relevant topics come up.
+The skill itself is a thin router (~100 tokens) — retrieval happens via Lucid.
+
+OPTIONS
+  --title  TEXT      Override display title (defaults to manifest).
+  --topics LIST      Comma-separated trigger topics (e.g. "naming,functions").
+  --scope  SCOPE     user (default, ~/.claude/skills) | project (.claude/skills)
+`;
+const LIST_HELP = `lucid book list [options]
+
+OPTIONS
+  --dir DIR         Books root. Defaults to ./books/
+`;

package/build/tools/init.js

@@ -53,7 +53,7 @@ const LUCID_PRE_EDIT_HOOK = {
     hooks: [
         {
             type: "command",
-            command: `lucid guard pre-edit 2>&1 # ${LUCID_GUARD_MARKER}`,
+            command: `lucid guard pre-edit # ${LUCID_GUARD_MARKER}`,
         },
     ],
 };

package/build/tools/sync.js

@@ -8,7 +8,7 @@ import { decompress } from "../store/content.js";
 import { implicitRewardFromSync } from "../memory/experience.js";
 import { indexFileInQdrant } from "../retrieval/qdrant.js";
 import { loadConfig, getQdrantConfig } from "../config.js";
-const SUPPORTED_EXTS = new Set([".ts", ".tsx", ".js", ".jsx", ".vue", ".py", ".go", ".rs"]);
+const SUPPORTED_EXTS = new Set([".ts", ".tsx", ".js", ".jsx", ".vue", ".py", ".go", ".rs", ".md"]);
 // ---------------------------------------------------------------------------
 // sync_file
 // ---------------------------------------------------------------------------

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@a13xu/lucid",
-  "version": "1.19.0",
+  "version": "1.21.0",
   "description": "Token-efficient memory, code indexing, and validation for Claude Code agents — SQLite + FTS5, TF-IDF + Qdrant retrieval, AST skeleton pruning, diff-aware context, Logic Guardian drift detection",
   "type": "module",
   "bin": {