@cmflow/source-pack 0.0.0-stage → 3.20.0

package/dist/buildSourcePack-DIp8G_uL.mjs

@@ -0,0 +1,3 @@
+import { t as buildSourcePack } from "./buildSourcePack-CaaMZ_tw.mjs";
+
+export { buildSourcePack };

package/dist/constants-DkiLCuIK.mjs

@@ -0,0 +1,25 @@
+//#region src/constants.ts
+const PACKAGE_NAME = "@cmflow/source-pack";
+/** Version of the archive layout (manifest, chunks, embeddings, attachments). */
+const SOURCE_PACK_FORMAT_VERSION = 2;
+/** Version of the chunking algorithm. Bump it whenever chunk boundaries or ids change. */
+const CHUNKING_FORMAT_VERSION = 1;
+const CHUNK_ID_ALGORITHM = "sha256(path:startLine:text)";
+const EMBEDDING_INPUT_FORMAT = "path\\ntext";
+const MANIFEST_FILE = "manifest.json";
+const CHUNKS_FILE = "chunks.jsonl.gz";
+const EMBEDDINGS_FILE = "embeddings.f32";
+const ATTACHMENTS_DIR = "attachments";
+const DEFAULT_CONFIG_FILE = "source-pack.config.ts";
+const DEFAULT_OUTPUT_DIR = "./tmp/source-pack";
+const DEFAULT_TAG_TEMPLATE = "version-{version}";
+const DEFAULT_ASSET_NAME_TEMPLATE = "source-pack-{version}.tar.gz";
+const DEFAULT_GITLAB_PACKAGE_NAME = "source-pack";
+const DEFAULT_GITHUB_API_URL = "https://api.github.com";
+const DEFAULT_GITLAB_API_URL = "https://gitlab.com/api/v4";
+const GITHUB_TOKEN_ENV = "GITHUB_TOKEN";
+const GITLAB_TOKEN_ENV = "GITLAB_TOKEN";
+
+//#endregion
+export { PACKAGE_NAME as _, DEFAULT_ASSET_NAME_TEMPLATE as a, DEFAULT_GITLAB_API_URL as c, DEFAULT_TAG_TEMPLATE as d, EMBEDDINGS_FILE as f, MANIFEST_FILE as g, GITLAB_TOKEN_ENV as h, CHUNK_ID_ALGORITHM as i, DEFAULT_GITLAB_PACKAGE_NAME as l, GITHUB_TOKEN_ENV as m, CHUNKING_FORMAT_VERSION as n, DEFAULT_CONFIG_FILE as o, EMBEDDING_INPUT_FORMAT as p, CHUNKS_FILE as r, DEFAULT_GITHUB_API_URL as s, ATTACHMENTS_DIR as t, DEFAULT_OUTPUT_DIR as u, SOURCE_PACK_FORMAT_VERSION as v };
+//# sourceMappingURL=constants-DkiLCuIK.mjs.map

package/dist/constants-DkiLCuIK.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"constants-DkiLCuIK.mjs","names":[],"sources":["../src/constants.ts"],"sourcesContent":["export const PACKAGE_NAME = \"@cmflow/source-pack\";\n\n/** Version of the archive layout (manifest, chunks, embeddings, attachments). */\nexport const SOURCE_PACK_FORMAT_VERSION = 2;\n\n/** Version of the chunking algorithm. Bump it whenever chunk boundaries or ids change. */\nexport const CHUNKING_FORMAT_VERSION = 1;\n\nexport const CHUNK_ID_ALGORITHM = \"sha256(path:startLine:text)\";\n\nexport const EMBEDDING_INPUT_FORMAT = \"path\\\\ntext\";\n\nexport const MANIFEST_FILE = \"manifest.json\";\nexport const CHUNKS_FILE = \"chunks.jsonl.gz\";\nexport const EMBEDDINGS_FILE = \"embeddings.f32\";\nexport const ATTACHMENTS_DIR = \"attachments\";\n\nexport const DEFAULT_CONFIG_FILE = \"source-pack.config.ts\";\nexport const DEFAULT_OUTPUT_DIR = \"./tmp/source-pack\";\nexport const DEFAULT_TAG_TEMPLATE = \"version-{version}\";\nexport const DEFAULT_ASSET_NAME_TEMPLATE = \"source-pack-{version}.tar.gz\";\nexport const DEFAULT_GITLAB_PACKAGE_NAME = \"source-pack\";\nexport const DEFAULT_GITHUB_API_URL = \"https://api.github.com\";\nexport const DEFAULT_GITLAB_API_URL = \"https://gitlab.com/api/v4\";\nexport const GITHUB_TOKEN_ENV = \"GITHUB_TOKEN\";\nexport const GITLAB_TOKEN_ENV = \"GITLAB_TOKEN\";\n"],"mappings":";AAAA,MAAa,eAAe;;AAG5B,MAAa,6BAA6B;;AAG1C,MAAa,0BAA0B;AAEvC,MAAa,qBAAqB;AAElC,MAAa,yBAAyB;AAEtC,MAAa,gBAAgB;AAC7B,MAAa,cAAc;AAC3B,MAAa,kBAAkB;AAC/B,MAAa,kBAAkB;AAE/B,MAAa,sBAAsB;AACnC,MAAa,qBAAqB;AAClC,MAAa,uBAAuB;AACpC,MAAa,8BAA8B;AAC3C,MAAa,8BAA8B;AAC3C,MAAa,yBAAyB;AACtC,MAAa,yBAAyB;AACtC,MAAa,mBAAmB;AAChC,MAAa,mBAAmB"}

package/dist/defaults-BJs7a37e.mjs

@@ -0,0 +1,52 @@
+//#region src/config/defaults.ts
+const DEFAULT_INCLUDE = ["**/*.{ts,tsx,mts,cts,js,jsx,mjs,cjs,json,yaml,yml,md,mdx,feature,graphql,gql,sql,sh,py,java,kt,kts,go,rb,php,cs,css,scss,html,vue,svelte,toml,xml,properties,tf,gradle}", "**/Dockerfile"];
+const LOCKFILES_BY_ECOSYSTEM = {
+	node: [
+		"**/package-lock.json",
+		"**/npm-shrinkwrap.json",
+		"**/yarn.lock",
+		"**/pnpm-lock.yaml",
+		"**/bun.lock",
+		"**/bun.lockb"
+	],
+	python: [
+		"**/poetry.lock",
+		"**/Pipfile.lock",
+		"**/uv.lock"
+	],
+	rust: ["**/Cargo.lock"],
+	go: ["**/go.sum"],
+	php: ["**/composer.lock"],
+	ruby: ["**/Gemfile.lock"],
+	jvm: ["**/gradle.lockfile"]
+};
+const LOCKFILES = Object.values(LOCKFILES_BY_ECOSYSTEM).flat();
+const TEST_FILES_BY_ECOSYSTEM = {
+	node: ["**/*.spec.*", "**/*.test.*"],
+	python: ["**/test_*.py", "**/*_test.py"],
+	go: ["**/*_test.go"],
+	jvm: ["**/src/test/**"]
+};
+const TEST_FILES = Object.values(TEST_FILES_BY_ECOSYSTEM).flat();
+/** Installed dependencies of a Node.js project. */
+const NODE_DEPENDENCY_DIRS = ["**/node_modules/**", "**/.yarn/**"];
+/** Generated build and coverage output, whatever the ecosystem. */
+const BUILD_OUTPUT_DIRS = [
+	"**/dist/**",
+	"**/build/**",
+	"**/coverage/**"
+];
+const DEFAULT_EXCLUDE = [
+	...NODE_DEPENDENCY_DIRS,
+	...BUILD_OUTPUT_DIRS,
+	...LOCKFILES,
+	...TEST_FILES
+];
+const DEFAULT_MAX_FILE_BYTES = 1048576;
+const DEFAULT_DIMENSIONS = 1536;
+const DEFAULT_MAX_INPUT_CHARS = 12e3;
+const DEFAULT_API_KEY_ENV = "LITELLM_API_KEY";
+
+//#endregion
+export { DEFAULT_INCLUDE as a, LOCKFILES_BY_ECOSYSTEM as c, DEFAULT_EXCLUDE as i, NODE_DEPENDENCY_DIRS as l, DEFAULT_API_KEY_ENV as n, DEFAULT_MAX_FILE_BYTES as o, DEFAULT_DIMENSIONS as r, DEFAULT_MAX_INPUT_CHARS as s, BUILD_OUTPUT_DIRS as t, TEST_FILES_BY_ECOSYSTEM as u };
+//# sourceMappingURL=defaults-BJs7a37e.mjs.map

package/dist/defaults-BJs7a37e.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"defaults-BJs7a37e.mjs","names":[],"sources":["../src/config/defaults.ts"],"sourcesContent":["export const DEFAULT_INCLUDE = [\n  \"**/*.{ts,tsx,mts,cts,js,jsx,mjs,cjs,json,yaml,yml,md,mdx,feature,graphql,gql,sql,sh,py,java,kt,kts,go,rb,php,cs,css,scss,html,vue,svelte,toml,xml,properties,tf,gradle}\",\n  \"**/Dockerfile\"\n];\n\nexport const LOCKFILES_BY_ECOSYSTEM = {\n  node: [\"**/package-lock.json\", \"**/npm-shrinkwrap.json\", \"**/yarn.lock\", \"**/pnpm-lock.yaml\", \"**/bun.lock\", \"**/bun.lockb\"],\n  python: [\"**/poetry.lock\", \"**/Pipfile.lock\", \"**/uv.lock\"],\n  rust: [\"**/Cargo.lock\"],\n  go: [\"**/go.sum\"],\n  php: [\"**/composer.lock\"],\n  ruby: [\"**/Gemfile.lock\"],\n  jvm: [\"**/gradle.lockfile\"]\n};\n\nexport const LOCKFILES = Object.values(LOCKFILES_BY_ECOSYSTEM).flat();\n\n// Test files are left out of the index: they double its size without describing production behavior.\nexport const TEST_FILES_BY_ECOSYSTEM = {\n  node: [\"**/*.spec.*\", \"**/*.test.*\"],\n  python: [\"**/test_*.py\", \"**/*_test.py\"],\n  go: [\"**/*_test.go\"],\n  jvm: [\"**/src/test/**\"]\n};\n\nexport const TEST_FILES = Object.values(TEST_FILES_BY_ECOSYSTEM).flat();\n\n/** Installed dependencies of a Node.js project. */\nexport const NODE_DEPENDENCY_DIRS = [\"**/node_modules/**\", \"**/.yarn/**\"];\n\n/** Generated build and coverage output, whatever the ecosystem. */\nexport const BUILD_OUTPUT_DIRS = [\"**/dist/**\", \"**/build/**\", \"**/coverage/**\"];\n\nexport const DEFAULT_EXCLUDE = [...NODE_DEPENDENCY_DIRS, ...BUILD_OUTPUT_DIRS, ...LOCKFILES, ...TEST_FILES];\n\nexport const DEFAULT_MAX_FILE_BYTES = 1024 * 1024;\nexport const DEFAULT_CHUNK_LINES = 100;\nexport const DEFAULT_CHUNK_OVERLAP = 15;\nexport const DEFAULT_DIMENSIONS = 1536;\nexport const DEFAULT_BATCH_SIZE = 64;\nexport const DEFAULT_EMBEDDING_CONCURRENCY = 4;\nexport const DEFAULT_MAX_RETRIES = 3;\nexport const DEFAULT_MAX_INPUT_CHARS = 12_000;\nexport const DEFAULT_API_KEY_ENV = \"LITELLM_API_KEY\";\n"],"mappings":";AAAA,MAAa,kBAAkB,CAC7B,2KACA,eACF;AAEA,MAAa,yBAAyB;CACpC,MAAM;EAAC;EAAwB;EAA0B;EAAgB;EAAqB;EAAe;CAAc;CAC3H,QAAQ;EAAC;EAAkB;EAAmB;CAAY;CAC1D,MAAM,CAAC,eAAe;CACtB,IAAI,CAAC,WAAW;CAChB,KAAK,CAAC,kBAAkB;CACxB,MAAM,CAAC,iBAAiB;CACxB,KAAK,CAAC,oBAAoB;AAC5B;AAEA,MAAa,YAAY,OAAO,OAAO,sBAAsB,CAAC,CAAC,KAAK;AAGpE,MAAa,0BAA0B;CACrC,MAAM,CAAC,eAAe,aAAa;CACnC,QAAQ,CAAC,gBAAgB,cAAc;CACvC,IAAI,CAAC,cAAc;CACnB,KAAK,CAAC,gBAAgB;AACxB;AAEA,MAAa,aAAa,OAAO,OAAO,uBAAuB,CAAC,CAAC,KAAK;;AAGtE,MAAa,uBAAuB,CAAC,sBAAsB,aAAa;;AAGxE,MAAa,oBAAoB;CAAC;CAAc;CAAe;AAAgB;AAE/E,MAAa,kBAAkB;CAAC,GAAG;CAAsB,GAAG;CAAmB,GAAG;CAAW,GAAG;AAAU;AAE1G,MAAa,yBAAyB;AAGtC,MAAa,qBAAqB;AAIlC,MAAa,0BAA0B;AACvC,MAAa,sBAAsB"}

package/dist/index.d.mts

@@ -0,0 +1,397 @@
+//#region src/config/interfaces/UserConfig.d.ts
+type RepositoryProvider = "github" | "gitlab";
+interface RepositoryConfig {
+  provider: RepositoryProvider;
+  /** GitHub owner or GitLab namespace (group/subgroup). */
+  owner?: string;
+  /** Repository / project name. */
+  name?: string;
+  /** GitLab numeric id or full path. Takes precedence over `owner`/`name` for GitLab. */
+  projectId?: string | number;
+  /** REST API base URL. Defaults to https://api.github.com or $CI_API_V4_URL / https://gitlab.com/api/v4. */
+  apiUrl?: string;
+}
+interface SourcesConfig {
+  /** Globs (minimatch, dot files included) of Git-tracked files to index. */
+  include?: string[];
+  /** Globs excluded from `include`. */
+  exclude?: string[];
+  /** Files larger than this are skipped. Default: 1 MiB. */
+  maxFileBytes?: number;
+}
+interface ChunkingConfig {
+  /** Lines per chunk. Default: 100. */
+  lines?: number;
+  /** Lines shared between two consecutive chunks. Default: 15. */
+  overlap?: number;
+}
+interface EmbeddingConfig {
+  /** OpenAI-compatible base URL (LiteLLM). `/v1` is appended when missing. */
+  baseUrl: string;
+  /** Primary model alias. */
+  model: string;
+  /** Aliases tried, in order, when the primary model fails. They must produce identical vectors. */
+  fallbackModels?: string[];
+  /** Model recorded in the manifest (the public model behind the gateway aliases). Default: `model`. */
+  referenceModel?: string;
+  /** Default: 1536. */
+  dimensions?: number;
+  /** Inputs per request. Default: 64. */
+  batchSize?: number;
+  /** Batches sent in parallel. Default: 4. */
+  concurrency?: number;
+  /** Retries per model and batch before switching to the next fallback. Default: 3. */
+  maxRetries?: number;
+  /** Embedded text is truncated to this length. Default: 12000. */
+  maxInputChars?: number;
+  /** Name of the environment variable holding the API key. Default: LITELLM_API_KEY. */
+  apiKeyEnv?: string;
+}
+/** Published on the release of the repository's provider (GitHub release asset or GitLab release link). */
+interface PublishConfig {
+  /** Release tag template. Default: `version-{version}`. */
+  tag?: string;
+  /** Archive name template. Default: `source-pack-{version}.tar.gz`. */
+  assetName?: string;
+  /** GitLab Generic Package name. Default: `source-pack`. */
+  packageName?: string;
+  /** Create the release when the tag has none. Default: false (the release tool owns releases). */
+  createRelease?: boolean;
+}
+interface UserConfig {
+  /** Output directory, relative to the command's working directory. Default: `./tmp/source-pack`. */
+  outputDir?: string;
+  repository: RepositoryConfig;
+  sources?: SourcesConfig;
+  chunking?: ChunkingConfig;
+  /** Omit to build a lexical-only pack (no embeddings.f32). */
+  embedding?: EmbeddingConfig;
+  /** Git-tracked files copied verbatim into the pack under `attachments/`. */
+  attachments?: string[];
+  publish?: PublishConfig;
+}
+//#endregion
+//#region src/fn/defineConfig.d.ts
+declare function defineConfig(config: UserConfig): UserConfig;
+//#endregion
+//#region src/interfaces/SourceChunk.d.ts
+interface SourceChunk {
+  /** sha256 of `${path}:${startLine}:${text}` */
+  id: string;
+  /** POSIX path relative to the repository root */
+  path: string;
+  /** 1-based, inclusive */
+  startLine: number;
+  /** 1-based, inclusive */
+  endLine: number;
+  text: string;
+}
+//#endregion
+//#region src/interfaces/SourcePackManifest.d.ts
+interface SourcePackProject {
+  provider: RepositoryProvider;
+  /** Stable project identifier, e.g. `github:ClubMediterranee/digital-api` or `gitlab:1234`. */
+  id: string;
+  owner?: string;
+  name?: string;
+  projectId?: string;
+  apiUrl: string;
+}
+interface SourcePackAttachment {
+  /** Path in the repository */
+  path: string;
+  /** Path inside the archive */
+  file: string;
+  bytes: number;
+  sha256: string;
+}
+interface SourcePackEmbeddingInfo {
+  /** Reference model (never the regional alias actually called) */
+  model: string;
+  dimensions: number;
+  /** How a chunk is turned into the embedded text */
+  input: string;
+  maxInputChars: number;
+  /** Little-endian float32, one vector per chunk, in chunk order */
+  encoding: "float32-le";
+  file: string;
+}
+interface SourcePackManifest {
+  formatVersion: number;
+  generator: {
+    name: string;
+    version: string;
+  };
+  project: SourcePackProject;
+  version: string;
+  tag: string;
+  ref: string;
+  sha: string;
+  committedAt: string;
+  createdAt: string;
+  /** Chunking algorithm; its parameters (`lines`, `overlap`) are in `config.chunking`. */
+  chunking: {
+    formatVersion: number;
+    idAlgorithm: string;
+  };
+  config: {
+    /** sha256 of the canonical JSON of `sources` + `chunking` (+ chunking format version) */
+    fingerprint: string;
+    sources: {
+      include: string[];
+      exclude: string[];
+      maxFileBytes: number;
+    };
+    chunking: {
+      lines: number;
+      overlap: number;
+    };
+  };
+  embedding: SourcePackEmbeddingInfo | null;
+  files: {
+    indexed: number;
+    skippedBinary: number;
+    skippedTooLarge: number;
+  };
+  chunks: {
+    count: number;
+    file: string;
+  };
+  attachments: SourcePackAttachment[];
+  publish: {
+    assetName: string;
+  };
+}
+//#endregion
+//#region src/interfaces/SourcePack.d.ts
+interface SourcePack {
+  manifest: SourcePackManifest;
+  chunks: SourceChunk[];
+  /** All vectors, concatenated in chunk order (`chunks.length * dimensions` floats), or null for a pack built without embeddings. */
+  embeddings: Float32Array | null;
+  /** Attachments keyed by their repository path. */
+  attachments: Map<string, Buffer>;
+  /** Vector of the chunk at `index` (a view on `embeddings`), or null. */
+  getEmbedding(index: number): Float32Array | null;
+}
+//#endregion
+//#region src/pack/readSourcePack.d.ts
+/**
+ * Reads a `source-pack-<version>.tar.gz` (file path or bytes) and checks its consistency
+ * (format version, chunk count, embedding size).
+ */
+declare function readSourcePack(source: string | Uint8Array): Promise<SourcePack>;
+//#endregion
+//#region src/config/interfaces/ResolvedConfig.d.ts
+interface ResolvedRepositoryConfig {
+  provider: RepositoryProvider;
+  owner?: string;
+  name?: string;
+  projectId?: string;
+  apiUrl: string;
+}
+interface ResolvedEmbeddingConfig {
+  baseUrl: string;
+  model: string;
+  fallbackModels: string[];
+  referenceModel: string;
+  dimensions: number;
+  batchSize: number;
+  concurrency: number;
+  maxRetries: number;
+  maxInputChars: number;
+  apiKeyEnv: string;
+}
+interface ResolvedPublishConfig {
+  tag: string;
+  assetName: string;
+  packageName: string;
+  createRelease: boolean;
+}
+interface ResolvedConfig {
+  /** Absolute path of the loaded config file (empty when built in memory). */
+  configPath: string;
+  /** Directory the command runs from; relative paths are resolved against it. */
+  cwd: string;
+  outputDir: string;
+  repository: ResolvedRepositoryConfig;
+  sources: {
+    include: string[];
+    exclude: string[];
+    maxFileBytes: number;
+  };
+  chunking: {
+    lines: number;
+    overlap: number;
+  };
+  embedding: ResolvedEmbeddingConfig | null;
+  attachments: string[];
+  publish: ResolvedPublishConfig;
+}
+//#endregion
+//#region src/utils/logger.d.ts
+interface Logger {
+  info(message: string): void;
+  warn(message: string): void;
+  error(message: string): void;
+}
+//#endregion
+//#region src/pack/buildSourcePack.d.ts
+interface BuildSourcePackOptions {
+  config: ResolvedConfig;
+  /** Commit-ish to index. Default: `tag` when given, otherwise HEAD. */
+  ref?: string;
+  /** Release tag. Default: `publish.tag` rendered with the version. */
+  tag?: string;
+  /** Release version. Default: extracted from `tag`, otherwise `version` of package.json at the ref. */
+  version?: string;
+  /** Build the pack without embeddings.f32 (no LiteLLM call). */
+  skipEmbeddings?: boolean;
+  /** Compute chunks and sizes without calling LiteLLM nor writing anything. */
+  dryRun?: boolean;
+  env?: NodeJS.ProcessEnv;
+  fetch?: typeof globalThis.fetch;
+  logger?: Logger;
+  now?: () => Date;
+}
+interface BuildSourcePackResult {
+  manifest: SourcePackManifest;
+  chunks: SourceChunk[];
+  dryRun: boolean;
+  archivePath?: string;
+  /** Archive size (or estimate in dry-run: gzipped chunks + raw embeddings + attachments). */
+  bytes: number;
+  chunksBytes: number;
+  embeddingsBytes: number;
+}
+/** Builds the source pack of `config.repository` at a commit: chunks + embeddings + attachments + manifest. */
+declare function buildSourcePack(options: BuildSourcePackOptions): Promise<BuildSourcePackResult>;
+//#endregion
+//#region src/pack/writeSourcePack.d.ts
+interface WriteSourcePackOptions {
+  archivePath: string;
+  manifest: SourcePackManifest;
+  chunks: SourceChunk[];
+  embeddings: Float32Array | null;
+  /** Keyed by the archive path declared in `manifest.attachments[].file` */
+  attachments: Map<string, Buffer>;
+}
+interface WrittenSourcePack {
+  archivePath: string;
+  bytes: number;
+  chunksBytes: number;
+  embeddingsBytes: number;
+}
+/** Writes `manifest.json`, `chunks.jsonl.gz`, `embeddings.f32` and attachments into a deterministic `.tar.gz`. */
+declare function writeSourcePack({ archivePath, manifest, chunks, embeddings, attachments }: WriteSourcePackOptions): Promise<WrittenSourcePack>;
+//#endregion
+//#region src/publish/interfaces/PublishContext.d.ts
+interface PublishResult {
+  target: string;
+  dryRun: boolean;
+  /** Download URL of the published asset (planned URL in dry-run). */
+  url: string;
+  /** True when an asset with the same name was replaced. */
+  replaced: boolean;
+}
+//#endregion
+//#region src/publish/publishSourcePack.d.ts
+interface PublishSourcePackOptions {
+  config: ResolvedConfig;
+  /** Archive to publish. Default: the archive of the last `build` in `outputDir`. */
+  archivePath?: string;
+  dryRun?: boolean;
+  env?: NodeJS.ProcessEnv;
+  fetch?: typeof globalThis.fetch;
+  logger?: Logger;
+}
+/** Publishes a built source pack on the release of its tag, on the provider of `repository`. Idempotent. */
+declare function publishSourcePack(options: PublishSourcePackOptions): Promise<PublishResult>;
+//#endregion
+//#region src/config/loadConfig.d.ts
+/**
+ * Loads `source-pack.config.ts` (through tsx) or a `.mjs`/`.js` config, then validates it.
+ * Relative paths of the config are resolved against `cwd` (the command's directory).
+ */
+declare function loadConfig(configPath: string, { cwd, env }?: {
+  cwd?: string;
+  env?: NodeJS.ProcessEnv;
+}): Promise<ResolvedConfig>;
+//#endregion
+//#region src/config/resolveConfig.d.ts
+interface ResolveConfigOptions {
+  cwd?: string;
+  configPath?: string;
+  env?: NodeJS.ProcessEnv;
+}
+/**
+ * Validates a raw user config and fills every default. Throws an explicit, field-level error message on invalid input.
+ */
+declare function resolveConfig(raw: unknown, { cwd, configPath, env }?: ResolveConfigOptions): ResolvedConfig;
+//#endregion
+//#region src/chunking/chunkText.d.ts
+/**
+ * Splits a file into overlapping line windows. Identical to the state-control indexer:
+ * windows of `lines` lines advancing by `lines - overlap`, blank windows skipped,
+ * id = sha256(`${path}:${startLine}:${text}`).
+ */
+declare function chunkText(path: string, text: string, lines?: number, overlap?: number): SourceChunk[];
+//#endregion
+//#region src/chunking/embeddingInput.d.ts
+/** Text sent to the embedding model for a chunk (`path\ntext`, truncated). Reuse it when re-embedding chunks. */
+declare function embeddingInput(chunk: Pick<SourceChunk, "path" | "text">, maxChars?: number): string;
+//#endregion
+//#region src/chunking/configFingerprint.d.ts
+interface FingerprintInput {
+  sources: {
+    include: string[];
+    exclude: string[];
+    maxFileBytes: number;
+  };
+  chunking: {
+    lines: number;
+    overlap: number;
+  };
+}
+/** Fingerprint of everything that determines the chunk set for a given commit. */
+declare function configFingerprint({ sources, chunking }: FingerprintInput): string;
+//#endregion
+//#region src/embedding/embedTexts.d.ts
+interface EmbedTextsOptions {
+  config: ResolvedEmbeddingConfig;
+  apiKey: string;
+  fetch?: typeof globalThis.fetch;
+  logger?: Logger;
+}
+/**
+ * Embeds `inputs` in batches through an OpenAI-compatible `/v1/embeddings` endpoint, `concurrency` batches at a time.
+ * Each batch is retried `maxRetries` times per model; when the current model keeps failing,
+ * the next `fallbackModels` alias is used for this batch and the following ones.
+ * Returns the vectors concatenated in input order.
+ */
+declare function embedTexts(inputs: string[], { config, apiKey, fetch, logger }: EmbedTextsOptions): Promise<Float32Array>;
+//#endregion
+//#region src/constants.d.ts
+declare const PACKAGE_NAME = "@cmflow/source-pack";
+/** Version of the archive layout (manifest, chunks, embeddings, attachments). */
+declare const SOURCE_PACK_FORMAT_VERSION = 2;
+/** Version of the chunking algorithm. Bump it whenever chunk boundaries or ids change. */
+declare const CHUNKING_FORMAT_VERSION = 1;
+declare const CHUNK_ID_ALGORITHM = "sha256(path:startLine:text)";
+declare const EMBEDDING_INPUT_FORMAT = "path\\ntext";
+declare const MANIFEST_FILE = "manifest.json";
+declare const CHUNKS_FILE = "chunks.jsonl.gz";
+declare const EMBEDDINGS_FILE = "embeddings.f32";
+declare const ATTACHMENTS_DIR = "attachments";
+declare const DEFAULT_CONFIG_FILE = "source-pack.config.ts";
+declare const DEFAULT_OUTPUT_DIR = "./tmp/source-pack";
+declare const DEFAULT_TAG_TEMPLATE = "version-{version}";
+declare const DEFAULT_ASSET_NAME_TEMPLATE = "source-pack-{version}.tar.gz";
+declare const DEFAULT_GITLAB_PACKAGE_NAME = "source-pack";
+declare const DEFAULT_GITHUB_API_URL = "https://api.github.com";
+declare const DEFAULT_GITLAB_API_URL = "https://gitlab.com/api/v4";
+declare const GITHUB_TOKEN_ENV = "GITHUB_TOKEN";
+declare const GITLAB_TOKEN_ENV = "GITLAB_TOKEN";
+//#endregion
+export { ATTACHMENTS_DIR, type BuildSourcePackOptions, type BuildSourcePackResult, CHUNKING_FORMAT_VERSION, CHUNKS_FILE, CHUNK_ID_ALGORITHM, type ChunkingConfig, DEFAULT_ASSET_NAME_TEMPLATE, DEFAULT_CONFIG_FILE, DEFAULT_GITHUB_API_URL, DEFAULT_GITLAB_API_URL, DEFAULT_GITLAB_PACKAGE_NAME, DEFAULT_OUTPUT_DIR, DEFAULT_TAG_TEMPLATE, EMBEDDINGS_FILE, EMBEDDING_INPUT_FORMAT, type EmbeddingConfig, GITHUB_TOKEN_ENV, GITLAB_TOKEN_ENV, MANIFEST_FILE, PACKAGE_NAME, type PublishConfig, type PublishResult, type PublishSourcePackOptions, type RepositoryConfig, type RepositoryProvider, type ResolvedConfig, type ResolvedEmbeddingConfig, type ResolvedPublishConfig, type ResolvedRepositoryConfig, SOURCE_PACK_FORMAT_VERSION, type SourceChunk, type SourcePack, type SourcePackAttachment, type SourcePackEmbeddingInfo, type SourcePackManifest, type SourcePackProject, type SourcesConfig, type UserConfig, buildSourcePack, chunkText, configFingerprint, defineConfig, embedTexts, embeddingInput, loadConfig, publishSourcePack, readSourcePack, resolveConfig, writeSourcePack };
+//# sourceMappingURL=index.d.mts.map

package/dist/index.mjs

@@ -0,0 +1,73 @@
+import { _ as PACKAGE_NAME, a as DEFAULT_ASSET_NAME_TEMPLATE, c as DEFAULT_GITLAB_API_URL, d as DEFAULT_TAG_TEMPLATE, f as EMBEDDINGS_FILE, g as MANIFEST_FILE, h as GITLAB_TOKEN_ENV, i as CHUNK_ID_ALGORITHM, l as DEFAULT_GITLAB_PACKAGE_NAME, m as GITHUB_TOKEN_ENV, n as CHUNKING_FORMAT_VERSION, o as DEFAULT_CONFIG_FILE, p as EMBEDDING_INPUT_FORMAT, r as CHUNKS_FILE, s as DEFAULT_GITHUB_API_URL, t as ATTACHMENTS_DIR, u as DEFAULT_OUTPUT_DIR, v as SOURCE_PACK_FORMAT_VERSION } from "./constants-DkiLCuIK.mjs";
+import { n as resolveConfig, t as loadConfig } from "./loadConfig-CNrL90dq.mjs";
+import { n as readArchiveEntries, t as publishSourcePack } from "./publishSourcePack-xNr9RBgv.mjs";
+import { a as configFingerprint, i as embeddingInput, n as writeSourcePack, o as chunkText, r as embedTexts, t as buildSourcePack } from "./buildSourcePack-CaaMZ_tw.mjs";
+import { promisify } from "node:util";
+import { gunzip } from "node:zlib";
+
+//#region src/fn/defineConfig.ts
+function defineConfig(config) {
+	return config;
+}
+
+//#endregion
+//#region src/pack/decodeEmbeddings.ts
+const LITTLE_ENDIAN_HOST = new Uint8Array(new Uint16Array([1]).buffer)[0] === 1;
+/**
+* Inverse of `encodeEmbeddings`. On a little-endian host the result is a view on the bytes when they are 4-byte aligned
+* (one copy otherwise); a big-endian host decodes every value.
+*/
+function decodeEmbeddings(buffer, littleEndianHost = LITTLE_ENDIAN_HOST) {
+	if (buffer.byteLength % 4 !== 0) throw new Error(`Invalid embeddings file: ${buffer.byteLength} bytes is not a multiple of 4`);
+	if (littleEndianHost) {
+		const aligned = buffer.byteOffset % 4 === 0 ? buffer : new Uint8Array(buffer);
+		return new Float32Array(aligned.buffer, aligned.byteOffset, aligned.byteLength / 4);
+	}
+	const view = new DataView(buffer.buffer, buffer.byteOffset, buffer.byteLength);
+	const vectors = new Float32Array(buffer.byteLength / 4);
+	for (let index = 0; index < vectors.length; index += 1) vectors[index] = view.getFloat32(index * 4, true);
+	return vectors;
+}
+
+//#endregion
+//#region src/pack/readSourcePack.ts
+const gunzipAsync = promisify(gunzip);
+function required(files, name) {
+	const file = files.get(name);
+	if (!file) throw new Error(`Invalid source pack: ${name} is missing`);
+	return file;
+}
+function readEmbeddings(files, manifest) {
+	const { embedding, chunks } = manifest;
+	if (!embedding) return null;
+	const embeddings = decodeEmbeddings(required(files, embedding.file));
+	if (embeddings.length !== chunks.count * embedding.dimensions) throw new Error(`Invalid source pack: expected ${chunks.count} x ${embedding.dimensions} floats, found ${embeddings.length}`);
+	return embeddings;
+}
+/**
+* Reads a `source-pack-<version>.tar.gz` (file path or bytes) and checks its consistency
+* (format version, chunk count, embedding size).
+*/
+async function readSourcePack(source) {
+	const files = await readArchiveEntries(source);
+	const manifest = JSON.parse(required(files, MANIFEST_FILE).toString("utf8"));
+	if (manifest.formatVersion !== 2) throw new Error(`Unsupported source pack format ${manifest.formatVersion} (this reader supports ${2})`);
+	const chunks = (await gunzipAsync(required(files, manifest.chunks.file))).toString("utf8").split("\n").filter(Boolean).map((line) => JSON.parse(line));
+	if (chunks.length !== manifest.chunks.count) throw new Error(`Invalid source pack: manifest declares ${manifest.chunks.count} chunks, found ${chunks.length}`);
+	const embeddings = readEmbeddings(files, manifest);
+	const dimensions = manifest.embedding?.dimensions ?? 0;
+	return {
+		manifest,
+		chunks,
+		embeddings,
+		attachments: new Map(manifest.attachments.map((attachment) => [attachment.path, required(files, attachment.file)])),
+		getEmbedding(index) {
+			if (!embeddings || index < 0 || index >= chunks.length) return null;
+			return embeddings.subarray(index * dimensions, (index + 1) * dimensions);
+		}
+	};
+}
+
+//#endregion
+export { ATTACHMENTS_DIR, CHUNKING_FORMAT_VERSION, CHUNKS_FILE, CHUNK_ID_ALGORITHM, DEFAULT_ASSET_NAME_TEMPLATE, DEFAULT_CONFIG_FILE, DEFAULT_GITHUB_API_URL, DEFAULT_GITLAB_API_URL, DEFAULT_GITLAB_PACKAGE_NAME, DEFAULT_OUTPUT_DIR, DEFAULT_TAG_TEMPLATE, EMBEDDINGS_FILE, EMBEDDING_INPUT_FORMAT, GITHUB_TOKEN_ENV, GITLAB_TOKEN_ENV, MANIFEST_FILE, PACKAGE_NAME, SOURCE_PACK_FORMAT_VERSION, buildSourcePack, chunkText, configFingerprint, defineConfig, embedTexts, embeddingInput, loadConfig, publishSourcePack, readSourcePack, resolveConfig, writeSourcePack };
+//# sourceMappingURL=index.mjs.map

package/dist/index.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.mjs","names":[],"sources":["../src/fn/defineConfig.ts","../src/pack/decodeEmbeddings.ts","../src/pack/readSourcePack.ts"],"sourcesContent":["import type { UserConfig } from \"../config/interfaces/UserConfig\";\n\nexport function defineConfig(config: UserConfig): UserConfig {\n  return config;\n}\n","const LITTLE_ENDIAN_HOST = new Uint8Array(new Uint16Array([1]).buffer)[0] === 1;\n\n/**\n * Inverse of `encodeEmbeddings`. On a little-endian host the result is a view on the bytes when they are 4-byte aligned\n * (one copy otherwise); a big-endian host decodes every value.\n */\nexport function decodeEmbeddings(buffer: Uint8Array, littleEndianHost = LITTLE_ENDIAN_HOST): Float32Array {\n  if (buffer.byteLength % 4 !== 0) {\n    throw new Error(`Invalid embeddings file: ${buffer.byteLength} bytes is not a multiple of 4`);\n  }\n\n  if (littleEndianHost) {\n    const aligned = buffer.byteOffset % 4 === 0 ? buffer : new Uint8Array(buffer);\n\n    return new Float32Array(aligned.buffer, aligned.byteOffset, aligned.byteLength / 4);\n  }\n\n  const view = new DataView(buffer.buffer, buffer.byteOffset, buffer.byteLength);\n  const vectors = new Float32Array(buffer.byteLength / 4);\n\n  for (let index = 0; index < vectors.length; index += 1) {\n    vectors[index] = view.getFloat32(index * 4, true);\n  }\n\n  return vectors;\n}\n","import { promisify } from \"node:util\";\nimport { gunzip } from \"node:zlib\";\nimport { MANIFEST_FILE, SOURCE_PACK_FORMAT_VERSION } from \"../constants\";\nimport type { SourceChunk } from \"../interfaces/SourceChunk\";\nimport type { SourcePack } from \"../interfaces/SourcePack\";\nimport type { SourcePackManifest } from \"../interfaces/SourcePackManifest\";\nimport { decodeEmbeddings } from \"./decodeEmbeddings\";\nimport { readArchiveEntries } from \"./readArchiveEntries\";\n\nconst gunzipAsync = promisify(gunzip);\n\nfunction required(files: Map<string, Buffer>, name: string): Buffer {\n  const file = files.get(name);\n\n  if (!file) {\n    throw new Error(`Invalid source pack: ${name} is missing`);\n  }\n\n  return file;\n}\n\nfunction readEmbeddings(files: Map<string, Buffer>, manifest: SourcePackManifest): Float32Array | null {\n  const { embedding, chunks } = manifest;\n\n  if (!embedding) {\n    return null;\n  }\n\n  const embeddings = decodeEmbeddings(required(files, embedding.file));\n\n  if (embeddings.length !== chunks.count * embedding.dimensions) {\n    throw new Error(`Invalid source pack: expected ${chunks.count} x ${embedding.dimensions} floats, found ${embeddings.length}`);\n  }\n\n  return embeddings;\n}\n\n/**\n * Reads a `source-pack-<version>.tar.gz` (file path or bytes) and checks its consistency\n * (format version, chunk count, embedding size).\n */\nexport async function readSourcePack(source: string | Uint8Array): Promise<SourcePack> {\n  const files = await readArchiveEntries(source);\n  const manifest = JSON.parse(required(files, MANIFEST_FILE).toString(\"utf8\")) as SourcePackManifest;\n\n  if (manifest.formatVersion !== SOURCE_PACK_FORMAT_VERSION) {\n    throw new Error(`Unsupported source pack format ${manifest.formatVersion} (this reader supports ${SOURCE_PACK_FORMAT_VERSION})`);\n  }\n\n  const chunks = (await gunzipAsync(required(files, manifest.chunks.file)))\n    .toString(\"utf8\")\n    .split(\"\\n\")\n    .filter(Boolean)\n    .map((line) => JSON.parse(line) as SourceChunk);\n\n  if (chunks.length !== manifest.chunks.count) {\n    throw new Error(`Invalid source pack: manifest declares ${manifest.chunks.count} chunks, found ${chunks.length}`);\n  }\n\n  const embeddings = readEmbeddings(files, manifest);\n  const dimensions = manifest.embedding?.dimensions ?? 0;\n  const attachments = new Map(manifest.attachments.map((attachment) => [attachment.path, required(files, attachment.file)]));\n\n  return {\n    manifest,\n    chunks,\n    embeddings,\n    attachments,\n    getEmbedding(index) {\n      if (!embeddings || index < 0 || index >= chunks.length) {\n        return null;\n      }\n\n      return embeddings.subarray(index * dimensions, (index + 1) * dimensions);\n    }\n  };\n}\n"],"mappings":";;;;;;;;AAEA,SAAgB,aAAa,QAAgC;CAC3D,OAAO;AACT;;;;ACJA,MAAM,qBAAqB,IAAI,WAAW,IAAI,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,OAAO;;;;;AAM9E,SAAgB,iBAAiB,QAAoB,mBAAmB,oBAAkC;CACxG,IAAI,OAAO,aAAa,MAAM,GAC5B,MAAM,IAAI,MAAM,4BAA4B,OAAO,WAAW,8BAA8B;CAG9F,IAAI,kBAAkB;EACpB,MAAM,UAAU,OAAO,aAAa,MAAM,IAAI,SAAS,IAAI,WAAW,MAAM;EAE5E,OAAO,IAAI,aAAa,QAAQ,QAAQ,QAAQ,YAAY,QAAQ,aAAa,CAAC;CACpF;CAEA,MAAM,OAAO,IAAI,SAAS,OAAO,QAAQ,OAAO,YAAY,OAAO,UAAU;CAC7E,MAAM,UAAU,IAAI,aAAa,OAAO,aAAa,CAAC;CAEtD,KAAK,IAAI,QAAQ,GAAG,QAAQ,QAAQ,QAAQ,SAAS,GACnD,QAAQ,SAAS,KAAK,WAAW,QAAQ,GAAG,IAAI;CAGlD,OAAO;AACT;;;;AChBA,MAAM,cAAc,UAAU,MAAM;AAEpC,SAAS,SAAS,OAA4B,MAAsB;CAClE,MAAM,OAAO,MAAM,IAAI,IAAI;CAE3B,IAAI,CAAC,MACH,MAAM,IAAI,MAAM,wBAAwB,KAAK,YAAY;CAG3D,OAAO;AACT;AAEA,SAAS,eAAe,OAA4B,UAAmD;CACrG,MAAM,EAAE,WAAW,WAAW;CAE9B,IAAI,CAAC,WACH,OAAO;CAGT,MAAM,aAAa,iBAAiB,SAAS,OAAO,UAAU,IAAI,CAAC;CAEnE,IAAI,WAAW,WAAW,OAAO,QAAQ,UAAU,YACjD,MAAM,IAAI,MAAM,iCAAiC,OAAO,MAAM,KAAK,UAAU,WAAW,iBAAiB,WAAW,QAAQ;CAG9H,OAAO;AACT;;;;;AAMA,eAAsB,eAAe,QAAkD;CACrF,MAAM,QAAQ,MAAM,mBAAmB,MAAM;CAC7C,MAAM,WAAW,KAAK,MAAM,SAAS,OAAO,aAAa,CAAC,CAAC,SAAS,MAAM,CAAC;CAE3E,IAAI,SAAS,qBACX,MAAM,IAAI,MAAM,kCAAkC,SAAS,cAAc,2BAAoD,EAAE;CAGjI,MAAM,UAAU,MAAM,YAAY,SAAS,OAAO,SAAS,OAAO,IAAI,CAAC,EAAC,CACrE,SAAS,MAAM,CAAC,CAChB,MAAM,IAAI,CAAC,CACX,OAAO,OAAO,CAAC,CACf,KAAK,SAAS,KAAK,MAAM,IAAI,CAAgB;CAEhD,IAAI,OAAO,WAAW,SAAS,OAAO,OACpC,MAAM,IAAI,MAAM,0CAA0C,SAAS,OAAO,MAAM,iBAAiB,OAAO,QAAQ;CAGlH,MAAM,aAAa,eAAe,OAAO,QAAQ;CACjD,MAAM,aAAa,SAAS,WAAW,cAAc;CAGrD,OAAO;EACL;EACA;EACA;EACA,iBANsB,IAAI,SAAS,YAAY,KAAK,eAAe,CAAC,WAAW,MAAM,SAAS,OAAO,WAAW,IAAI,CAAC,CAAC,CAM5G;EACV,aAAa,OAAO;GAClB,IAAI,CAAC,cAAc,QAAQ,KAAK,SAAS,OAAO,QAC9C,OAAO;GAGT,OAAO,WAAW,SAAS,QAAQ,aAAa,QAAQ,KAAK,UAAU;EACzE;CACF;AACF"}