claude-brain 0.30.2 → 0.30.3

package/src/intelligence/hf-downloader.ts

@@ -1,222 +1,222 @@
-/**
- * Hugging Face Hub Downloader — streams ONNX models from HF with SHA256 verification.
- * Atomic writes via temp file + rename. Inline retry with exponential backoff.
- */
-
-import { createHash } from 'node:crypto'
-import { existsSync, mkdirSync, renameSync, unlinkSync } from 'node:fs'
-import { join } from 'node:path'
-import type { ModelTask } from '@/intelligence/model-manager'
-
-export interface HfManifestEntry {
-  file: string
-  metaFile: string
-  sha256: string
-  metaSha256: string
-  size: number
-  version: string
-  params: string
-  accuracy: number | null
-  labels: string[]
-  maxSeqLen: number
-}
-
-export interface HfManifest {
-  hfRepo: string
-  hfBranch: string
-  models: Record<string, HfManifestEntry>
-}
-
-export interface DownloadOptions {
-  /** Target directory to write models into */
-  destDir: string
-  /** Tasks to download (subset of manifest keys) */
-  tasks: ModelTask[]
-  /** HF repo override (default from manifest) */
-  hfRepo?: string
-  /** HF branch override (default from manifest) */
-  hfBranch?: string
-  /** Progress callback: task name, bytes downloaded so far, total bytes */
-  onProgress?: (task: string, downloaded: number, total: number) => void
-  /** Called when a task completes */
-  onComplete?: (task: string, bytes: number) => void
-  /** Called on error */
-  onError?: (task: string, error: string) => void
-}
-
-export interface DownloadResult {
-  task: string
-  success: boolean
-  bytes: number
-  error?: string
-}
-
-const MAX_RETRIES = 3
-const RETRY_DELAYS = [2000, 4000, 8000]
-const DOWNLOAD_TIMEOUT_MS = 300_000 // 5 minutes per file
-
-/**
- * Build HF resolve URL for a file.
- * Pattern: https://huggingface.co/{repo}/resolve/{branch}/{filename}
- */
-function hfUrl(repo: string, branch: string, filename: string): string {
-  return `https://huggingface.co/${repo}/resolve/${branch}/${filename}`
-}
-
-/**
- * Stream-download a single file with SHA256 verification and atomic write.
- * Returns the number of bytes written.
- */
-async function downloadFile(
-  url: string,
-  destPath: string,
-  expectedSha256: string,
-  expectedSize: number,
-  onProgress?: (downloaded: number, total: number) => void,
-): Promise<number> {
-  const tempPath = `${destPath}.download`
-
-  // Clean up any leftover temp file
-  if (existsSync(tempPath)) {
-    unlinkSync(tempPath)
-  }
-
-  const response = await fetch(url, {
-    signal: AbortSignal.timeout(DOWNLOAD_TIMEOUT_MS),
-  })
-
-  if (!response.ok) {
-    throw new Error(`HTTP ${response.status}: ${response.statusText}`)
-  }
-
-  if (!response.body) {
-    throw new Error('Response body is null')
-  }
-
-  const contentLength = parseInt(response.headers.get('content-length') || '0', 10) || expectedSize
-  const hash = createHash('sha256')
-  const writer = Bun.file(tempPath).writer()
-  let downloaded = 0
-
-  const reader = response.body.getReader()
-
-  try {
-    while (true) {
-      const { done, value } = await reader.read()
-      if (done) break
-
-      writer.write(value)
-      hash.update(value)
-      downloaded += value.byteLength
-
-      if (onProgress) {
-        onProgress(downloaded, contentLength)
-      }
-    }
-  } finally {
-    await writer.end()
-  }
-
-  // Verify SHA256
-  const actualSha256 = hash.digest('hex')
-  if (actualSha256 !== expectedSha256) {
-    // Clean up bad file
-    if (existsSync(tempPath)) unlinkSync(tempPath)
-    throw new Error(
-      `SHA256 mismatch: expected ${expectedSha256.slice(0, 12)}..., got ${actualSha256.slice(0, 12)}...`
-    )
-  }
-
-  // Atomic rename
-  renameSync(tempPath, destPath)
-  return downloaded
-}
-
-/**
- * Download a single file with retry logic.
- */
-async function downloadWithRetry(
-  url: string,
-  destPath: string,
-  expectedSha256: string,
-  expectedSize: number,
-  onProgress?: (downloaded: number, total: number) => void,
-): Promise<number> {
-  let lastError: Error | null = null
-
-  for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
-    try {
-      return await downloadFile(url, destPath, expectedSha256, expectedSize, onProgress)
-    } catch (err) {
-      lastError = err instanceof Error ? err : new Error(String(err))
-
-      // Clean up temp file on failure
-      const tempPath = `${destPath}.download`
-      if (existsSync(tempPath)) {
-        try { unlinkSync(tempPath) } catch { /* ignore */ }
-      }
-
-      if (attempt < MAX_RETRIES) {
-        await new Promise(resolve => setTimeout(resolve, RETRY_DELAYS[attempt]))
-      }
-    }
-  }
-
-  throw lastError!
-}
-
-/**
- * Main entry point: download models from Hugging Face Hub.
- */
-export async function downloadFromHuggingFace(
-  manifest: HfManifest,
-  options: DownloadOptions,
-): Promise<DownloadResult[]> {
-  const repo = options.hfRepo || manifest.hfRepo
-  const branch = options.hfBranch || manifest.hfBranch
-  const results: DownloadResult[] = []
-
-  // Ensure dest directory exists
-  if (!existsSync(options.destDir)) {
-    mkdirSync(options.destDir, { recursive: true })
-  }
-
-  for (const task of options.tasks) {
-    const entry = manifest.models[task]
-    if (!entry) {
-      const err = `No manifest entry for task "${task}"`
-      options.onError?.(task, err)
-      results.push({ task, success: false, bytes: 0, error: err })
-      continue
-    }
-
-    try {
-      // Download ONNX model
-      const modelUrl = hfUrl(repo, branch, entry.file)
-      const modelDest = join(options.destDir, entry.file)
-
-      const bytes = await downloadWithRetry(
-        modelUrl,
-        modelDest,
-        entry.sha256,
-        entry.size,
-        (downloaded, total) => options.onProgress?.(task, downloaded, total),
-      )
-
-      // Download metadata JSON
-      const metaUrl = hfUrl(repo, branch, entry.metaFile)
-      const metaDest = join(options.destDir, entry.metaFile)
-
-      await downloadWithRetry(metaUrl, metaDest, entry.metaSha256, 0)
-
-      options.onComplete?.(task, bytes)
-      results.push({ task, success: true, bytes })
-    } catch (err) {
-      const msg = err instanceof Error ? err.message : String(err)
-      options.onError?.(task, msg)
-      results.push({ task, success: false, bytes: 0, error: msg })
-    }
-  }
-
-  return results
-}
+/**
+ * Hugging Face Hub Downloader — streams ONNX models from HF with SHA256 verification.
+ * Atomic writes via temp file + rename. Inline retry with exponential backoff.
+ */
+
+import { createHash } from 'node:crypto'
+import { existsSync, mkdirSync, renameSync, unlinkSync } from 'node:fs'
+import { join } from 'node:path'
+import type { ModelTask } from '@/intelligence/model-manager'
+
+export interface HfManifestEntry {
+  file: string
+  metaFile: string
+  sha256: string
+  metaSha256: string
+  size: number
+  version: string
+  params: string
+  accuracy: number | null
+  labels: string[]
+  maxSeqLen: number
+}
+
+export interface HfManifest {
+  hfRepo: string
+  hfBranch: string
+  models: Record<string, HfManifestEntry>
+}
+
+export interface DownloadOptions {
+  /** Target directory to write models into */
+  destDir: string
+  /** Tasks to download (subset of manifest keys) */
+  tasks: ModelTask[]
+  /** HF repo override (default from manifest) */
+  hfRepo?: string
+  /** HF branch override (default from manifest) */
+  hfBranch?: string
+  /** Progress callback: task name, bytes downloaded so far, total bytes */
+  onProgress?: (task: string, downloaded: number, total: number) => void
+  /** Called when a task completes */
+  onComplete?: (task: string, bytes: number) => void
+  /** Called on error */
+  onError?: (task: string, error: string) => void
+}
+
+export interface DownloadResult {
+  task: string
+  success: boolean
+  bytes: number
+  error?: string
+}
+
+const MAX_RETRIES = 3
+const RETRY_DELAYS = [2000, 4000, 8000]
+const DOWNLOAD_TIMEOUT_MS = 300_000 // 5 minutes per file
+
+/**
+ * Build HF resolve URL for a file.
+ * Pattern: https://huggingface.co/{repo}/resolve/{branch}/{filename}
+ */
+function hfUrl(repo: string, branch: string, filename: string): string {
+  return `https://huggingface.co/${repo}/resolve/${branch}/${filename}`
+}
+
+/**
+ * Stream-download a single file with SHA256 verification and atomic write.
+ * Returns the number of bytes written.
+ */
+async function downloadFile(
+  url: string,
+  destPath: string,
+  expectedSha256: string,
+  expectedSize: number,
+  onProgress?: (downloaded: number, total: number) => void,
+): Promise<number> {
+  const tempPath = `${destPath}.download`
+
+  // Clean up any leftover temp file
+  if (existsSync(tempPath)) {
+    unlinkSync(tempPath)
+  }
+
+  const response = await fetch(url, {
+    signal: AbortSignal.timeout(DOWNLOAD_TIMEOUT_MS),
+  })
+
+  if (!response.ok) {
+    throw new Error(`HTTP ${response.status}: ${response.statusText}`)
+  }
+
+  if (!response.body) {
+    throw new Error('Response body is null')
+  }
+
+  const contentLength = parseInt(response.headers.get('content-length') || '0', 10) || expectedSize
+  const hash = createHash('sha256')
+  const writer = Bun.file(tempPath).writer()
+  let downloaded = 0
+
+  const reader = response.body.getReader()
+
+  try {
+    while (true) {
+      const { done, value } = await reader.read()
+      if (done) break
+
+      writer.write(value)
+      hash.update(value)
+      downloaded += value.byteLength
+
+      if (onProgress) {
+        onProgress(downloaded, contentLength)
+      }
+    }
+  } finally {
+    await writer.end()
+  }
+
+  // Verify SHA256
+  const actualSha256 = hash.digest('hex')
+  if (actualSha256 !== expectedSha256) {
+    // Clean up bad file
+    if (existsSync(tempPath)) unlinkSync(tempPath)
+    throw new Error(
+      `SHA256 mismatch: expected ${expectedSha256.slice(0, 12)}..., got ${actualSha256.slice(0, 12)}...`
+    )
+  }
+
+  // Atomic rename
+  renameSync(tempPath, destPath)
+  return downloaded
+}
+
+/**
+ * Download a single file with retry logic.
+ */
+async function downloadWithRetry(
+  url: string,
+  destPath: string,
+  expectedSha256: string,
+  expectedSize: number,
+  onProgress?: (downloaded: number, total: number) => void,
+): Promise<number> {
+  let lastError: Error | null = null
+
+  for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
+    try {
+      return await downloadFile(url, destPath, expectedSha256, expectedSize, onProgress)
+    } catch (err) {
+      lastError = err instanceof Error ? err : new Error(String(err))
+
+      // Clean up temp file on failure
+      const tempPath = `${destPath}.download`
+      if (existsSync(tempPath)) {
+        try { unlinkSync(tempPath) } catch { /* ignore */ }
+      }
+
+      if (attempt < MAX_RETRIES) {
+        await new Promise(resolve => setTimeout(resolve, RETRY_DELAYS[attempt]))
+      }
+    }
+  }
+
+  throw lastError!
+}
+
+/**
+ * Main entry point: download models from Hugging Face Hub.
+ */
+export async function downloadFromHuggingFace(
+  manifest: HfManifest,
+  options: DownloadOptions,
+): Promise<DownloadResult[]> {
+  const repo = options.hfRepo || manifest.hfRepo
+  const branch = options.hfBranch || manifest.hfBranch
+  const results: DownloadResult[] = []
+
+  // Ensure dest directory exists
+  if (!existsSync(options.destDir)) {
+    mkdirSync(options.destDir, { recursive: true })
+  }
+
+  for (const task of options.tasks) {
+    const entry = manifest.models[task]
+    if (!entry) {
+      const err = `No manifest entry for task "${task}"`
+      options.onError?.(task, err)
+      results.push({ task, success: false, bytes: 0, error: err })
+      continue
+    }
+
+    try {
+      // Download ONNX model
+      const modelUrl = hfUrl(repo, branch, entry.file)
+      const modelDest = join(options.destDir, entry.file)
+
+      const bytes = await downloadWithRetry(
+        modelUrl,
+        modelDest,
+        entry.sha256,
+        entry.size,
+        (downloaded, total) => options.onProgress?.(task, downloaded, total),
+      )
+
+      // Download metadata JSON
+      const metaUrl = hfUrl(repo, branch, entry.metaFile)
+      const metaDest = join(options.destDir, entry.metaFile)
+
+      await downloadWithRetry(metaUrl, metaDest, entry.metaSha256, 0)
+
+      options.onComplete?.(task, bytes)
+      results.push({ task, success: true, bytes })
+    } catch (err) {
+      const msg = err instanceof Error ? err.message : String(err)
+      options.onError?.(task, msg)
+      results.push({ task, success: false, bytes: 0, error: msg })
+    }
+  }
+
+  return results
+}

package/src/intelligence/hf-manifest.json

@@ -1,78 +1,78 @@
-{
-  "hfRepo": "demgun101/claude-brain-models",
-  "hfBranch": "main",
-  "models": {
-    "intent": {
-      "file": "intent.onnx",
-      "metaFile": "intent.json",
-      "sha256": "f276ec091712f53ceeccfdca70d468d0b5aa6da7ee5c4bd7867a7464a9732dd2",
-      "metaSha256": "d5702fd45e5685187e74810da75e41be9d12d9fd82b7ccd7244a9f0d33076c65",
-      "size": 120073283,
-      "version": "0.1.0",
-      "params": "nanoGPT-intent",
-      "accuracy": 0.9904,
-      "labels": ["session_start", "context_needed", "decision_made", "store_this", "pattern_found", "mistake_learned", "progress_update", "question", "comparison", "exploration", "list_all", "update_memory", "delete_memory", "detail_request", "timeline", "no_action"],
-      "maxSeqLen": 128
-    },
-    "entity": {
-      "file": "entity.onnx",
-      "metaFile": "entity.json",
-      "sha256": "d15706b78112e8dda3caa7e054237106b67200f90a1e6e0e4336526468dea8b5",
-      "metaSha256": "1de0636bc3bd9de23c0ce767a148d6becad7b6add6a0a717449a7ac185296d55",
-      "size": 120058476,
-      "version": "0.1.0",
-      "params": "nanoGPT-entity",
-      "accuracy": 0.9795,
-      "labels": ["O", "B-TECH", "I-TECH", "B-PROJECT", "I-PROJECT", "B-CONCEPT", "I-CONCEPT"],
-      "maxSeqLen": 128
-    },
-    "query": {
-      "file": "query.onnx",
-      "metaFile": "query.json",
-      "sha256": "802446105ad873b4a172dc875c07e245882f48691f72ae0abb0bda5934fa084a",
-      "metaSha256": "32393875ff58a87da83480673868e4ac14c3fd1a59e16836d7c1f2fa02a7f25e",
-      "size": 120057880,
-      "version": "0.1.0",
-      "params": "nanoGPT-query",
-      "accuracy": 0.985,
-      "labels": ["factual", "procedural", "comparative", "temporal", "exploratory", "decision"],
-      "maxSeqLen": 128
-    },
-    "knowledge": {
-      "file": "knowledge.onnx",
-      "metaFile": "knowledge.json",
-      "sha256": "a786eefc5ef5c6f2fc132c6de7f0972891057f5a993c9c2d9264207d36165035",
-      "metaSha256": "6aa77194cbd8c10a2451958ec5a3e43328df97ee8aea46abf53e3e432f3f3c4d",
-      "size": 120056340,
-      "version": "0.1.0",
-      "params": "nanoGPT-knowledge",
-      "accuracy": 0.998,
-      "labels": ["fact", "preference", "constraint", "goal", "definition"],
-      "maxSeqLen": 128
-    },
-    "compress": {
-      "file": "compress.onnx",
-      "metaFile": "compress.json",
-      "sha256": "2d950a0e0a2cdc5dc90b7c44803c5fb81c76d824bd208c64a7104c5845e1c237",
-      "metaSha256": "e91500c665ec47083bb6b3ff8c83d529f263792d83f7f169401267ceb1e8d031",
-      "size": 357902441,
-      "version": "0.1.0",
-      "params": "nanoGPT-compress",
-      "accuracy": null,
-      "labels": [],
-      "maxSeqLen": 256
-    },
-    "pattern": {
-      "file": "pattern.onnx",
-      "metaFile": "pattern.json",
-      "sha256": "3b44371eaef11fb8ccc4c1636d6cfbdcdf62ae8bc6ffc808ff0ff45b24824fde",
-      "metaSha256": "1126f3fa9a115b4d26063e006e7d33b9e82b7740194ac86b5dae9f25d5dfd1a2",
-      "size": 254526620,
-      "version": "0.1.0",
-      "params": "nanoGPT-pattern",
-      "accuracy": 0.8667,
-      "labels": ["solution", "anti-pattern", "best-practice", "common-issue"],
-      "maxSeqLen": 128
-    }
-  }
-}
+{
+  "hfRepo": "demgun101/claude-brain-models",
+  "hfBranch": "main",
+  "models": {
+    "intent": {
+      "file": "intent.onnx",
+      "metaFile": "intent.json",
+      "sha256": "f276ec091712f53ceeccfdca70d468d0b5aa6da7ee5c4bd7867a7464a9732dd2",
+      "metaSha256": "d5702fd45e5685187e74810da75e41be9d12d9fd82b7ccd7244a9f0d33076c65",
+      "size": 120073283,
+      "version": "0.1.0",
+      "params": "nanoGPT-intent",
+      "accuracy": 0.9904,
+      "labels": ["session_start", "context_needed", "decision_made", "store_this", "pattern_found", "mistake_learned", "progress_update", "question", "comparison", "exploration", "list_all", "update_memory", "delete_memory", "detail_request", "timeline", "no_action"],
+      "maxSeqLen": 128
+    },
+    "entity": {
+      "file": "entity.onnx",
+      "metaFile": "entity.json",
+      "sha256": "d15706b78112e8dda3caa7e054237106b67200f90a1e6e0e4336526468dea8b5",
+      "metaSha256": "1de0636bc3bd9de23c0ce767a148d6becad7b6add6a0a717449a7ac185296d55",
+      "size": 120058476,
+      "version": "0.1.0",
+      "params": "nanoGPT-entity",
+      "accuracy": 0.9795,
+      "labels": ["O", "B-TECH", "I-TECH", "B-PROJECT", "I-PROJECT", "B-CONCEPT", "I-CONCEPT"],
+      "maxSeqLen": 128
+    },
+    "query": {
+      "file": "query.onnx",
+      "metaFile": "query.json",
+      "sha256": "802446105ad873b4a172dc875c07e245882f48691f72ae0abb0bda5934fa084a",
+      "metaSha256": "32393875ff58a87da83480673868e4ac14c3fd1a59e16836d7c1f2fa02a7f25e",
+      "size": 120057880,
+      "version": "0.1.0",
+      "params": "nanoGPT-query",
+      "accuracy": 0.985,
+      "labels": ["factual", "procedural", "comparative", "temporal", "exploratory", "decision"],
+      "maxSeqLen": 128
+    },
+    "knowledge": {
+      "file": "knowledge.onnx",
+      "metaFile": "knowledge.json",
+      "sha256": "a786eefc5ef5c6f2fc132c6de7f0972891057f5a993c9c2d9264207d36165035",
+      "metaSha256": "6aa77194cbd8c10a2451958ec5a3e43328df97ee8aea46abf53e3e432f3f3c4d",
+      "size": 120056340,
+      "version": "0.1.0",
+      "params": "nanoGPT-knowledge",
+      "accuracy": 0.998,
+      "labels": ["fact", "preference", "constraint", "goal", "definition"],
+      "maxSeqLen": 128
+    },
+    "compress": {
+      "file": "compress.onnx",
+      "metaFile": "compress.json",
+      "sha256": "2d950a0e0a2cdc5dc90b7c44803c5fb81c76d824bd208c64a7104c5845e1c237",
+      "metaSha256": "e91500c665ec47083bb6b3ff8c83d529f263792d83f7f169401267ceb1e8d031",
+      "size": 357902441,
+      "version": "0.1.0",
+      "params": "nanoGPT-compress",
+      "accuracy": null,
+      "labels": [],
+      "maxSeqLen": 256
+    },
+    "pattern": {
+      "file": "pattern.onnx",
+      "metaFile": "pattern.json",
+      "sha256": "3b44371eaef11fb8ccc4c1636d6cfbdcdf62ae8bc6ffc808ff0ff45b24824fde",
+      "metaSha256": "1126f3fa9a115b4d26063e006e7d33b9e82b7740194ac86b5dae9f25d5dfd1a2",
+      "size": 254526620,
+      "version": "0.1.0",
+      "params": "nanoGPT-pattern",
+      "accuracy": 0.8667,
+      "labels": ["solution", "anti-pattern", "best-practice", "common-issue"],
+      "maxSeqLen": 128
+    }
+  }
+}

package/src/intelligence/index.ts

@@ -1,24 +1,24 @@
-/**
- * Intelligence Module
- * Phase 22: Consolidates advanced intelligence features under one namespace.
- *
- * Features are lazy-loaded to avoid startup cost when not needed.
- * Gate on config.intelligence?.enabled for opt-in activation.
- */
-
-import type { Config } from '@/config'
-
-export async function loadIntelligence(config: Config) {
-  if (!(config as any).intelligence?.enabled) return null
-  return {
-    prediction: await import('./prediction'),
-    reasoning: await import('./reasoning'),
-    temporal: await import('./temporal'),
-    optimization: await import('./optimization'),
-    crossProject: await import('./cross-project')
-  }
-}
-
-// Re-export submodule paths for direct imports
-export * from './optimization/semantic-cache'
-export * from './optimization/precompute'
+/**
+ * Intelligence Module
+ * Phase 22: Consolidates advanced intelligence features under one namespace.
+ *
+ * Features are lazy-loaded to avoid startup cost when not needed.
+ * Gate on config.intelligence?.enabled for opt-in activation.
+ */
+
+import type { Config } from '@/config'
+
+export async function loadIntelligence(config: Config) {
+  if (!(config as Record<string, unknown>).intelligence) return null
+  return {
+    prediction: await import('./prediction'),
+    reasoning: await import('./reasoning'),
+    temporal: await import('./temporal'),
+    optimization: await import('./optimization'),
+    crossProject: await import('./cross-project')
+  }
+}
+
+// Re-export submodule paths for direct imports
+export * from './optimization/semantic-cache'
+export * from './optimization/precompute'