orizu 0.5.20 → 0.5.21

package/dist/sandbox-provider.d.ts

@@ -0,0 +1,193 @@
+/**
+ * Orizu-owned sandbox provider seam (ALI-973 / WS-F, per ADR-003).
+ *
+ * ADR-003 requires every provider call (GitHub, Daytona) to sit behind an
+ * Orizu-owned interface so the durable contracts never couple to whichever SDK
+ * wins the quarter. This module defines `SandboxProvider` / `SandboxSession` and
+ * ships two implementations:
+ *
+ *   - `createDaytonaProvider(config)` — a THIN adapter over the Daytona
+ *     TypeScript SDK (`@daytonaio/sdk`). The SDK is imported LAZILY (dynamic
+ *     import through a non-literal specifier) so the CLI works without it
+ *     installed, the app build never resolves it, and tests never load it.
+ *   - `createLocalSimProvider(config)` — a temp-dir + real-`git`-subprocess
+ *     simulation used by the unit tests AND by the no-credentials local-sim
+ *     rehearsal. It is a genuine rehearsal (real clone/commit/push against a
+ *     local bare repo), not a mock.
+ *
+ * SDK-vs-docs note (verified against daytona.io/docs, 2026-07-02): the real
+ * `git.clone` signature is
+ *   clone(url, path, branch?, commitId?, username?, password?, insecureSkipTls?)
+ * i.e. a `commitId` argument sits BETWEEN `branch` and `username` — the ALI-973
+ * brief documented `clone(url, path, branch, username, password)`. The adapter
+ * passes `undefined` for `commitId` so the Orizu seam keeps its clean per-call
+ * credential shape. `git.push(path, username?, password?)` matches the brief.
+ */
+export interface ExecResult {
+    stdout: string;
+    exitCode: number;
+    /** Captured standard error, when the provider exposes it — surfaced so a
+     *  failed git clone/push does not vanish behind a masked exit code. */
+    stderr?: string;
+}
+export interface GitCloneParams {
+    url: string;
+    path: string;
+    branch?: string;
+    username?: string;
+    password?: string;
+}
+export interface GitPushParams {
+    path: string;
+    username?: string;
+    password?: string;
+}
+/**
+ * A single request transform a provider firewall applies to matching egress
+ * (the credential-brokering seam: inject a header at the proxy so the raw
+ * secret never enters the sandbox). Structurally identical to @vercel/sandbox's
+ * `NetworkPolicyRule` so a Vercel adapter can pass it straight through, but
+ * declared here so the Orizu seam owns the shape (ADR-003).
+ */
+export interface SandboxEgressRule {
+    transform?: {
+        headers?: Record<string, string>;
+    }[];
+}
+/**
+ * Provider-neutral egress/network policy, wired to the sandbox firewall AT
+ * CREATE. G5 (ALI-1006) owns the allowlist CONTENT; this slice only plumbs the
+ * option so the orchestrator can pass a policy (incl. a credential-injection
+ * transform for a model endpoint). `'allow-all'` is the provider default when
+ * omitted. Structurally compatible with @vercel/sandbox's `NetworkPolicy`.
+ */
+export type SandboxEgressPolicy = 'allow-all' | 'deny-all' | {
+    allow?: string[] | Record<string, SandboxEgressRule[]>;
+    subnets?: {
+        allow?: string[];
+        deny?: string[];
+    };
+};
+export interface SandboxCreateOpts {
+    language?: string;
+    snapshot?: string;
+    envVars?: Record<string, string>;
+    labels?: Record<string, string>;
+    /** Sandbox auto-terminate timeout (ms). Honored where the provider supports a
+     *  configurable session length (Vercel); ignored by providers that do not. */
+    timeoutMs?: number;
+    /** Provider runtime image hint (e.g. Vercel 'node24'); provider-specific. */
+    runtime?: string;
+    /**
+     * Provider custom-image ref (e.g. Vercel VCR 'orizu-hosted-runtime:<tag>').
+     * When set, the provider creates the sandbox from this pre-baked image instead
+     * of the base runtime (ALI-1017). Provider-specific — Vercel honors it; other
+     * providers ignore it. `runtime` is still the base-runtime hint when no image
+     * is given.
+     */
+    image?: string;
+    /** vCPUs to allocate (memory scales with vCPUs on Vercel: 2048 MiB each). */
+    vcpus?: number;
+    /** Ports to expose from the sandbox (provider-specific; up to 4 on Vercel). */
+    ports?: number[];
+    /** Egress/firewall policy applied at create (G5 supplies content). */
+    egressPolicy?: SandboxEgressPolicy;
+}
+/**
+ * A live sandbox. Per-call git credentials mirror Daytona's contract (and
+ * OpenInspect's credential-helper shape): the token is handed to each operation,
+ * never written once per session.
+ */
+export interface SandboxSession {
+    readonly id: string;
+    gitClone(params: GitCloneParams): Promise<void>;
+    gitPush(params: GitPushParams): Promise<void>;
+    exec(command: string, opts?: {
+        cwd?: string;
+    }): Promise<ExecResult>;
+    readFile(path: string): Promise<string>;
+    writeFile(path: string, content: string): Promise<void>;
+    fileExists(path: string): Promise<boolean>;
+    destroy(): Promise<void>;
+    /**
+     * Extend the sandbox's auto-terminate timeout by `durationMs` (keepalive).
+     * OPTIONAL: only providers with a configurable session length implement it
+     * (Vercel). Host-side token rotation uses it to keep a long session alive
+     * beyond the provider's default. Absent on local-sim / Daytona.
+     */
+    extendTimeout?(durationMs: number): Promise<void>;
+    /**
+     * Snapshot the running sandbox's filesystem state and RETURN the snapshot id
+     * (ALI-1017). A new sandbox can then boot from it via `createSandbox({ snapshot })`
+     * — the zero-Docker "prebaked runtime" path (parallel to a VCR `image`).
+     * OPTIONAL: only providers with a snapshot API implement it (Vercel); absent on
+     * local-sim / Daytona. NOTE: the provider may STOP this sandbox as part of the
+     * snapshot (Vercel does), so treat the session as spent afterwards.
+     *
+     * `expiration` is an optional TTL (ms) for the snapshot; 0 = never expire.
+     */
+    snapshot?(opts?: {
+        expiration?: number;
+    }): Promise<string>;
+}
+export type SandboxProviderKind = 'daytona' | 'local-sim' | 'vercel';
+export interface SandboxProvider {
+    readonly kind: SandboxProviderKind;
+    createSandbox(opts?: SandboxCreateOpts): Promise<SandboxSession>;
+}
+export interface DaytonaProviderConfig {
+    apiKey: string;
+    apiUrl?: string;
+    target?: string;
+}
+interface DaytonaSdkSandbox {
+    id: string;
+    git: {
+        clone(url: string, path: string, branch?: string, commitId?: string, username?: string, password?: string, insecureSkipTls?: boolean): Promise<void>;
+        push(path: string, username?: string, password?: string): Promise<void>;
+    };
+    process: {
+        executeCommand(command: string, cwd?: string, env?: Record<string, string>, timeout?: number): Promise<{
+            exitCode?: number;
+            result?: string;
+            stderr?: string;
+            artifacts?: {
+                stdout?: string;
+                stderr?: string;
+            };
+        }>;
+    };
+    fs: {
+        downloadFile(remotePath: string): Promise<Uint8Array | Buffer | string>;
+        uploadFile(file: Uint8Array, remotePath: string): Promise<void>;
+        getFileDetails(path: string): Promise<unknown>;
+    };
+    delete(timeout?: number): Promise<void>;
+}
+interface DaytonaSdkClient {
+    create(params?: Record<string, unknown>, options?: {
+        timeout: number;
+    }): Promise<DaytonaSdkSandbox>;
+}
+export interface DaytonaSdkModule {
+    Daytona: new (config: {
+        apiKey: string;
+        apiUrl?: string;
+        target?: string;
+    }) => DaytonaSdkClient;
+}
+/** Test seam: inject a fake SDK module so the adapter's request mapping can be
+ *  exercised without the real package installed or a network. */
+export interface DaytonaProviderDeps {
+    loadModule?: () => Promise<DaytonaSdkModule>;
+}
+export declare function createDaytonaProvider(config: DaytonaProviderConfig, deps?: DaytonaProviderDeps): SandboxProvider;
+export interface LocalSimProviderConfig {
+    /** Parent directory for per-sandbox temp dirs (defaults to the OS tmpdir). */
+    rootDir?: string;
+    /** Skip filesystem teardown on destroy() (used by --keep for debugging). */
+    keepOnDestroy?: boolean;
+}
+export declare function createLocalSimProvider(config?: LocalSimProviderConfig): SandboxProvider;
+export {};
+//# sourceMappingURL=sandbox-provider.d.ts.map

package/dist/sandbox-provider.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandbox-provider.d.ts","sourceRoot":"","sources":["../src/sandbox-provider.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AAOH,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,MAAM,CAAA;IACd,QAAQ,EAAE,MAAM,CAAA;IAChB;2EACuE;IACvE,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,cAAc;IAC7B,GAAG,EAAE,MAAM,CAAA;IACX,IAAI,EAAE,MAAM,CAAA;IACZ,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAA;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED;;;;;;GAMG;AACH,MAAM,WAAW,iBAAiB;IAChC,SAAS,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,EAAE,CAAA;CACnD;AAED;;;;;;GAMG;AACH,MAAM,MAAM,mBAAmB,GAC3B,WAAW,GACX,UAAU,GACV;IACE,KAAK,CAAC,EAAE,MAAM,EAAE,GAAG,MAAM,CAAC,MAAM,EAAE,iBAAiB,EAAE,CAAC,CAAA;IACtD,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAA;KAAE,CAAA;CAChD,CAAA;AAEL,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAChC,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B;kFAC8E;IAC9E,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,6EAA6E;IAC7E,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;;;;;;OAMG;IACH,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,6EAA6E;IAC7E,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,+EAA+E;IAC/E,KAAK,CAAC,EAAE,MAAM,EAAE,CAAA;IAChB,sEAAsE;IACtE,YAAY,CAAC,EAAE,mBAAmB,CAAA;CACnC;AAED;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAA;IACnB,QAAQ,CAAC,MAAM,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAC/C,OAAO,CAAC,MAAM,EAAE,aAAa,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAC7C,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,UAAU,CAAC,CAAA;IACnE,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;IACvC,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACvD,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAAA;IAC1C,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;IACxB;;;;;OAKG;IACH,aAAa,CAAC,CAAC,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACjD;;;;;;;;;OASG;IACH,QAAQ,CAAC,CAAC,IAAI,CAAC,EAAE;QAAE,UAAU,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAA;CAC3D;AAED,MAAM,MAAM,mBAAmB,GAAG,SAAS,GAAG,WAAW,GAAG,QAAQ,CAAA;AAEpE,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,IAAI,EAAE,mBAAmB,CAAA;IAClC,aAAa,CAAC,IAAI,CAAC,EAAE,iBAAiB,GAAG,OAAO,CAAC,cAAc,CAAC,CAAA;CACjE;AAID,MAAM,WAAW,qBAAqB;IACpC,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAID,UAAU,iBAAiB;IACzB,EAAE,EAAE,MAAM,CAAA;IACV,GAAG,EAAE;QACH,KAAK,CACH,GAAG,EAAE,MAAM,EACX,IAAI,EAAE,MAAM,EACZ,MAAM,CAAC,EAAE,MAAM,EACf,QAAQ,CAAC,EAAE,MAAM,EACjB,QAAQ,CAAC,EAAE,MAAM,EACjB,QAAQ,CAAC,EAAE,MAAM,EACjB,eAAe,CAAC,EAAE,OAAO,GACxB,OAAO,CAAC,IAAI,CAAC,CAAA;QAChB,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;KACxE,CAAA;IACD,OAAO,EAAE;QACP,cAAc,CACZ,OAAO,EAAE,MAAM,EACf,GAAG,CAAC,EAAE,MAAM,EACZ,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC5B,OAAO,CAAC,EAAE,MAAM,GACf,OAAO,CAAC;YAAE,QAAQ,CAAC,EAAE,MAAM,CAAC;YAAC,MAAM,CAAC,EAAE,MAAM,CAAC;YAAC,MAAM,CAAC,EAAE,MAAM,CAAC;YAAC,SAAS,CAAC,EAAE;gBAAE,MAAM,CAAC,EAAE,MAAM,CAAC;gBAAC,MAAM,CAAC,EAAE,MAAM,CAAA;aAAE,CAAA;SAAE,CAAC,CAAA;KACtH,CAAA;IACD,EAAE,EAAE;QACF,YAAY,CAAC,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,UAAU,GAAG,MAAM,GAAG,MAAM,CAAC,CAAA;QACvE,UAAU,CAAC,IAAI,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;QAC/D,cAAc,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAAA;KAC/C,CAAA;IACD,MAAM,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;CACxC;AAED,UAAU,gBAAgB;IACxB,MAAM,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO,CAAC,EAAE;QAAE,OAAO,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAAA;CACpG;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,KAAK,MAAM,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,KAAK,gBAAgB,CAAA;CAChG;AAED;iEACiE;AACjE,MAAM,WAAW,mBAAmB;IAClC,UAAU,CAAC,EAAE,MAAM,OAAO,CAAC,gBAAgB,CAAC,CAAA;CAC7C;AAyGD,wBAAgB,qBAAqB,CACnC,MAAM,EAAE,qBAAqB,EAC7B,IAAI,GAAE,mBAAwB,GAC7B,eAAe,CA0CjB;AAID,MAAM,WAAW,sBAAsB;IACrC,8EAA8E;IAC9E,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,4EAA4E;IAC5E,aAAa,CAAC,EAAE,OAAO,CAAA;CACxB;AAkFD,wBAAgB,sBAAsB,CAAC,MAAM,GAAE,sBAA2B,GAAG,eAAe,CAY3F"}

package/dist/scorer-draft-push.d.ts

@@ -0,0 +1,50 @@
+/**
+ * ADR-007 P5 (ALI-1074): the session-scoped, commit-first scorer push.
+ *
+ * `orizu scorers register --session <id>` creates the version through
+ * POST /api/cli/scorers/drafts: the server commits the scorer's canonical
+ * `manifest.json` to the session's `orizu/session-*` branch (one Git Data API
+ * commit) and then writes a DRAFT row pinning
+ * repo_path/content_sha/commit_sha. The draft seals — and gets its version
+ * number — when the session branch merges via `orizu session finish` +
+ * promotion-manifest apply.
+ *
+ * The git filename is ALWAYS `manifest.json`: the server read side
+ * (lib/artifact-git-read.ts SCORER_MANIFEST_FILENAMES) prefers that name.
+ *
+ * Lives outside index.ts per the CLI line ratchet (ALI-976); mirrors
+ * prompt-draft-push.ts.
+ */
+/** index.ts helpers injected by the command dispatcher (report-comments-cli
+ * precedent: command modules receive a ctx instead of importing the CLI
+ * entrypoint). */
+export interface ScorersRegisterContext {
+    getArg: (name: string) => string | null;
+    resolveProjectSlug: (projectArg: string | null) => Promise<string>;
+    readJsonFile: (pathArg: string) => Record<string, unknown>;
+    hasJsonFlag: () => boolean;
+    printJson: (value: Record<string, unknown>) => void;
+    printLine: (message?: string) => void;
+    parseJsonResponse: (response: Response, label: string) => Promise<Record<string, unknown>>;
+}
+/** The full `orizu scorers register` command (moved out of index.ts for the
+ * CLI line ratchet): --session takes the ADR-007 P5 commit-first draft path,
+ * sessionless stays on the legacy create endpoint. */
+export declare function runScorersRegister(ctx: ScorersRegisterContext): Promise<void>;
+export interface PushScorerDraftInput {
+    project: string;
+    sessionId: string;
+    name: string;
+    /** The parsed scorer manifest (the artifact's canonical bytes). */
+    manifest: Record<string, unknown>;
+    promptVersionId?: string;
+    runnerVersionId?: string;
+    /** Test seam; defaults to the authenticated CLI fetcher. */
+    fetcher?: (path: string, init?: RequestInit) => Promise<Response>;
+}
+export interface PushScorerDraftResult {
+    data: Record<string, unknown>;
+    message: string;
+}
+export declare function pushScorerDraft(input: PushScorerDraftInput): Promise<PushScorerDraftResult>;
+//# sourceMappingURL=scorer-draft-push.d.ts.map

package/dist/scorer-draft-push.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"scorer-draft-push.d.ts","sourceRoot":"","sources":["../src/scorer-draft-push.ts"],"names":[],"mappings":"AAIA;;;;;;;;;;;;;;;;GAgBG;AAEH;;kBAEkB;AAClB,MAAM,WAAW,sBAAsB;IACrC,MAAM,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,MAAM,GAAG,IAAI,CAAA;IACvC,kBAAkB,EAAE,CAAC,UAAU,EAAE,MAAM,GAAG,IAAI,KAAK,OAAO,CAAC,MAAM,CAAC,CAAA;IAClE,YAAY,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAC1D,WAAW,EAAE,MAAM,OAAO,CAAA;IAC1B,SAAS,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,IAAI,CAAA;IACnD,SAAS,EAAE,CAAC,OAAO,CAAC,EAAE,MAAM,KAAK,IAAI,CAAA;IACrC,iBAAiB,EAAE,CAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAA;CAC3F;AAED;;sDAEsD;AACtD,wBAAsB,kBAAkB,CAAC,GAAG,EAAE,sBAAsB,GAAG,OAAO,CAAC,IAAI,CAAC,CAyDnF;AAED,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,MAAM,CAAA;IACf,SAAS,EAAE,MAAM,CAAA;IACjB,IAAI,EAAE,MAAM,CAAA;IACZ,mEAAmE;IACnE,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IACjC,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,4DAA4D;IAC5D,OAAO,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAA;CAClE;AAED,MAAM,WAAW,qBAAqB;IACpC,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAC7B,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,wBAAsB,eAAe,CAAC,KAAK,EAAE,oBAAoB,GAAG,OAAO,CAAC,qBAAqB,CAAC,CAsCjG"}

package/dist/scorers-cli.d.ts

@@ -0,0 +1,20 @@
+/**
+ * The `orizu scorers` subcommand router (ALI-1554).
+ *
+ * The five historical branches lived inline in index.ts; folding them here
+ * reclaims the headroom the CLI line ratchet (ALI-976) requires before a new
+ * subcommand may be added. Behaviour is unchanged: each handler is the same
+ * function index.ts already called, and an unrecognised subcommand returns
+ * null so main() still falls through to printUsage().
+ */
+export interface ScorersCommandIo {
+    list: () => Promise<void>;
+    register: () => Promise<void>;
+    detail: () => Promise<void>;
+    labelsSet: () => Promise<void>;
+    exec: () => Promise<void>;
+    verifyParity: (args: string[]) => Promise<number>;
+}
+/** Returns the process exit code, or null when the subcommand is not ours. */
+export declare function scorersCommand(args: string[], io: ScorersCommandIo): Promise<number | null>;
+//# sourceMappingURL=scorers-cli.d.ts.map

package/dist/scorers-cli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"scorers-cli.d.ts","sourceRoot":"","sources":["../src/scorers-cli.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;IACzB,QAAQ,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;IAC7B,MAAM,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;IAC3B,SAAS,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;IAC9B,IAAI,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;IACzB,YAAY,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,OAAO,CAAC,MAAM,CAAC,CAAA;CAClD;AAED,8EAA8E;AAC9E,wBAAsB,cAAc,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,gBAAgB,GAAG,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAoBjG"}

package/dist/scorers-cli.js

@@ -0,0 +1,22 @@
+/** Returns the process exit code, or null when the subcommand is not ours. */
+export async function scorersCommand(args, io) {
+    const subcommand = args[1];
+    // A Map, not an object literal: `scorers constructor` must not resolve to
+    // an Object.prototype member and be treated as a known subcommand.
+    const handlers = new Map([
+        ['list', io.list], ['register', io.register], ['detail', io.detail], ['exec', io.exec],
+    ]);
+    const handler = subcommand === undefined ? undefined : handlers.get(subcommand);
+    if (handler) {
+        await handler();
+        return 0;
+    }
+    if (subcommand === 'labels' && args[2] === 'set') {
+        await io.labelsSet();
+        return 0;
+    }
+    if (subcommand === 'verify-parity') {
+        return io.verifyParity(args.slice(2));
+    }
+    return null;
+}

package/dist/scorers-verify-parity-bridge.d.ts

@@ -0,0 +1,16 @@
+/**
+ * The customer-side half of `orizu scorers verify-parity` (ALI-1554).
+ *
+ * Embedded as a string constant and written to a temp file at run time on
+ * purpose: no packaging (`files`/`vendor`) change, so there is nothing for the
+ * published tarball to lose. It runs in the CUSTOMER's interpreter and the
+ * customer's unscrubbed environment — it is their metric in their process.
+ *
+ * Contract: reads `{rows: [{row_id, row, model_output}]}` from
+ * ORIZU_PARITY_INPUT_PATH, calls `<module>:<function>(row, model_output)` per
+ * row, and writes `{scores: [{row_id, score} | {row_id, error}]}` to
+ * ORIZU_PARITY_OUTPUT_PATH. Exit 2 (with the reason on stderr) means the check
+ * could not run at all; a per-row exception is that row's error, not an abort.
+ */
+export declare const PARITY_BRIDGE_SOURCE = "import json, math, os, sys, traceback\n\n\ndef fail(message):\n    sys.stderr.write(message + \"\\n\")\n    sys.exit(2)\n\n\ndef last_line(text):\n    lines = [line for line in text.strip().splitlines() if line.strip()]\n    return lines[-1] if lines else \"unknown error\"\n\n\ndef extract(value):\n    # Mirrors orizu_gepa.optimizer._score_from_scorer, minus the clamp: the\n    # CLI clamps both sides so the compared number is what GEPA optimizes.\n    # bool is deliberately NOT rejected: measured, _score_from_scorer(extra=\n    # {'score': True}) -> (1.0, None) and False -> (0.0, None), because bool is an\n    # int subclass and Python's numeric branch accepts it. An exact-match DSPy\n    # metric returning True must compare, not error.\n    if isinstance(value, (int, float)):\n        raw = value\n    elif isinstance(value, dict):\n        raw = value.get(\"score\")\n        if raw is None and isinstance(value.get(\"model_response\"), dict):\n            raw = value[\"model_response\"].get(\"score\")\n    elif hasattr(value, \"score\"):\n        # official GEPA's EvaluationResult NamedTuple and friends;\n        # objective_scores is deliberately ignored.\n        raw = value.score\n    else:\n        raise TypeError(\"scorer result must be a number, a dict with a numeric 'score', or an object with a numeric .score attribute\")\n    if isinstance(raw, str):\n        raw = float(raw)\n    if not isinstance(raw, (int, float)):\n        raise TypeError(\"scorer result must include a numeric score\")\n    raw = float(raw)\n    if not math.isfinite(raw):\n        raise ValueError(\"scorer result score must be finite\")\n    return raw\n\n\ntarget = os.environ.get(\"ORIZU_PARITY_ORIGINAL\", \"\")\nmodule_name, separator, function_name = target.partition(\":\")\nif not separator or not module_name or not function_name:\n    fail(\"--original must be <module>:<function> (got %r)\" % target)\n\n# The customer's metric lives beside the command's cwd, not beside this\n# generated bridge file, so cwd must win over the script directory.\nsys.path.insert(0, os.getcwd())\ntry:\n    import importlib\n\n    module = importlib.import_module(module_name)\nexcept Exception:\n    fail(\"Failed to import original metric module '%s': %s\" % (module_name, last_line(traceback.format_exc())))\n\nif not hasattr(module, function_name):\n    fail(\"Original metric module '%s' has no attribute '%s'\" % (module_name, function_name))\nmetric = getattr(module, function_name)\nif not callable(metric):\n    fail(\"Original metric '%s:%s' is not callable\" % (module_name, function_name))\n\nwith open(os.environ[\"ORIZU_PARITY_INPUT_PATH\"], encoding=\"utf-8\") as handle:\n    payload = json.load(handle)\n\nscores = []\nfor item in payload[\"rows\"]:\n    try:\n        scores.append({\"row_id\": item[\"row_id\"], \"score\": extract(metric(item[\"row\"], item[\"model_output\"]))})\n    except Exception:\n        scores.append({\"row_id\": item[\"row_id\"], \"error\": last_line(traceback.format_exc())})\n\nwith open(os.environ[\"ORIZU_PARITY_OUTPUT_PATH\"], \"w\", encoding=\"utf-8\") as handle:\n    json.dump({\"scores\": scores}, handle)\n";
+//# sourceMappingURL=scorers-verify-parity-bridge.d.ts.map

package/dist/scorers-verify-parity-bridge.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"scorers-verify-parity-bridge.d.ts","sourceRoot":"","sources":["../src/scorers-verify-parity-bridge.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AACH,eAAO,MAAM,oBAAoB,8lGA2EhC,CAAA"}

package/dist/scorers-verify-parity-bridge.js

@@ -0,0 +1,90 @@
+/**
+ * The customer-side half of `orizu scorers verify-parity` (ALI-1554).
+ *
+ * Embedded as a string constant and written to a temp file at run time on
+ * purpose: no packaging (`files`/`vendor`) change, so there is nothing for the
+ * published tarball to lose. It runs in the CUSTOMER's interpreter and the
+ * customer's unscrubbed environment — it is their metric in their process.
+ *
+ * Contract: reads `{rows: [{row_id, row, model_output}]}` from
+ * ORIZU_PARITY_INPUT_PATH, calls `<module>:<function>(row, model_output)` per
+ * row, and writes `{scores: [{row_id, score} | {row_id, error}]}` to
+ * ORIZU_PARITY_OUTPUT_PATH. Exit 2 (with the reason on stderr) means the check
+ * could not run at all; a per-row exception is that row's error, not an abort.
+ */
+export const PARITY_BRIDGE_SOURCE = `import json, math, os, sys, traceback
+
+
+def fail(message):
+    sys.stderr.write(message + "\\n")
+    sys.exit(2)
+
+
+def last_line(text):
+    lines = [line for line in text.strip().splitlines() if line.strip()]
+    return lines[-1] if lines else "unknown error"
+
+
+def extract(value):
+    # Mirrors orizu_gepa.optimizer._score_from_scorer, minus the clamp: the
+    # CLI clamps both sides so the compared number is what GEPA optimizes.
+    # bool is deliberately NOT rejected: measured, _score_from_scorer(extra=
+    # {'score': True}) -> (1.0, None) and False -> (0.0, None), because bool is an
+    # int subclass and Python's numeric branch accepts it. An exact-match DSPy
+    # metric returning True must compare, not error.
+    if isinstance(value, (int, float)):
+        raw = value
+    elif isinstance(value, dict):
+        raw = value.get("score")
+        if raw is None and isinstance(value.get("model_response"), dict):
+            raw = value["model_response"].get("score")
+    elif hasattr(value, "score"):
+        # official GEPA's EvaluationResult NamedTuple and friends;
+        # objective_scores is deliberately ignored.
+        raw = value.score
+    else:
+        raise TypeError("scorer result must be a number, a dict with a numeric 'score', or an object with a numeric .score attribute")
+    if isinstance(raw, str):
+        raw = float(raw)
+    if not isinstance(raw, (int, float)):
+        raise TypeError("scorer result must include a numeric score")
+    raw = float(raw)
+    if not math.isfinite(raw):
+        raise ValueError("scorer result score must be finite")
+    return raw
+
+
+target = os.environ.get("ORIZU_PARITY_ORIGINAL", "")
+module_name, separator, function_name = target.partition(":")
+if not separator or not module_name or not function_name:
+    fail("--original must be <module>:<function> (got %r)" % target)
+
+# The customer's metric lives beside the command's cwd, not beside this
+# generated bridge file, so cwd must win over the script directory.
+sys.path.insert(0, os.getcwd())
+try:
+    import importlib
+
+    module = importlib.import_module(module_name)
+except Exception:
+    fail("Failed to import original metric module '%s': %s" % (module_name, last_line(traceback.format_exc())))
+
+if not hasattr(module, function_name):
+    fail("Original metric module '%s' has no attribute '%s'" % (module_name, function_name))
+metric = getattr(module, function_name)
+if not callable(metric):
+    fail("Original metric '%s:%s' is not callable" % (module_name, function_name))
+
+with open(os.environ["ORIZU_PARITY_INPUT_PATH"], encoding="utf-8") as handle:
+    payload = json.load(handle)
+
+scores = []
+for item in payload["rows"]:
+    try:
+        scores.append({"row_id": item["row_id"], "score": extract(metric(item["row"], item["model_output"]))})
+    except Exception:
+        scores.append({"row_id": item["row_id"], "error": last_line(traceback.format_exc())})
+
+with open(os.environ["ORIZU_PARITY_OUTPUT_PATH"], "w", encoding="utf-8") as handle:
+    json.dump({"scores": scores}, handle)
+`;

package/dist/scorers-verify-parity.d.ts

@@ -0,0 +1,81 @@
+/**
+ * `orizu scorers verify-parity` (ALI-1554): prove a migrated Orizu scorer
+ * runner scores the same rows the same way the customer's original Python
+ * metric does, UNDER THE PAYLOAD `run-gepa` sends the runner, before the first
+ * optimization spends tokens on a scorer that silently disagrees.
+ *
+ * Honesty note: `run-gepa` executes runners through the vendored Python
+ * `run_file_contract_runner`, not this TypeScript path. What this proves is
+ * "the runner, under the run-gepa flat_row payload, via the CLI's file
+ * contract". The payload is asserted key-for-key by
+ * test/scorers-verify-parity.test.ts (H9) against a literal hand-derived from
+ * `make_scorer_runner`'s flat_row branch — NOT against a fixture shared with
+ * `test_flat_row_payload_shape_mirrors_score_run_exec`, so a rename on the
+ * Python side turns only the Python test red until the shared fixture lands
+ * (ALI-1556). The env allowlist is the TypeScript one.
+ */
+export interface VerifyParityIo {
+    json: boolean;
+    print: (line: string) => void;
+    printErr: (line: string) => void;
+    materializeRunnerVersion: (runnerVersionId: string) => Promise<{
+        runnerDir: string;
+        cleanup: () => void;
+    }>;
+    readRunnerManifest: (runnerDir: string) => {
+        command: string[];
+        supports_body_kind?: string[];
+    };
+    runnerSubprocessEnv: (inputPath: string, outputPath: string) => NodeJS.ProcessEnv;
+    timeoutMs: number;
+    maxOutputBytes: number;
+    fetcher?: (path: string, init?: RequestInit) => Promise<Response>;
+    /** Test seam for the pairing tests (H8); production always uses the embedded bridge. */
+    bridgeSource?: string;
+}
+interface ExtractedScore {
+    raw: number;
+    score: number;
+}
+/**
+ * Python's `float()` grammar, not JavaScript's `Number()`. Measured 2026-08-23:
+ * `float` REJECTS '0x10', '0b1', '0o7', '' and '1,5' (all of which `Number`
+ * happily converts or turns into 0), and ACCEPTS '  0.5  ', '1e3', '1_0', '.5',
+ * '5.', '+3', 'inf', 'Infinity' and 'nan'. Anything `_score_from_scorer` would
+ * refuse during the real optimization must be refused here too, or this gate
+ * certifies a scorer run-gepa cannot use. `inf`/`nan` parse and are then
+ * rejected by the finite check, exactly as Python does.
+ */
+export declare function pythonFloat(text: string): number | null;
+export declare function extractParityScore(value: unknown, label: string): ExtractedScore;
+/**
+ * The Orizu side spawns the runner ONCE PER ROW, so each row gets the full
+ * per-row budget. The original side scores the whole sample in ONE bridge
+ * process, so handing it the per-row budget gives an N-row split 1/N of the
+ * time the Orizu side had: a 40-row split with an LLM-judge metric is SIGTERMed
+ * before it can write output.json and every row's work is lost (round-2 review).
+ * Scale by the sample length, bounded so a huge split cannot ask for forever.
+ */
+export declare function bridgeBudget(perRowTimeoutMs: number, perRowMaxOutputBytes: number, sampleLength: number): {
+    timeoutMs: number;
+    maxOutputBytes: number;
+};
+/**
+ * Resolve the scorer input contract and candidate-output row field with the
+ * SAME precedence `run-gepa` uses (`orizu_gepa.runner.resolve_scorer_input_contract`):
+ * explicit flag > runner manifest (`scorer_input_contract`, `candidate_output_field`)
+ * > defaults (`gepa`, `model_output`). Resolution is by PRESENCE, not truthiness,
+ * so a blank manifest value fails validation instead of silently defaulting.
+ *
+ * Parity may only be claimed for the payload run-gepa will actually send, so a
+ * contract that resolves to anything but `flat_row` is refused here rather than
+ * assumed: proving parity on a shape the optimization never uses is the exact
+ * silent-agreement failure this command exists to prevent.
+ */
+export declare function resolveScorerInputContract(manifest: Record<string, unknown>, flagContract: string | null, flagField: string | null): {
+    contract: string;
+    field: string;
+};
+export declare function verifyScorerParityCommand(args: string[], io: VerifyParityIo): Promise<number>;
+export {};
+//# sourceMappingURL=scorers-verify-parity.d.ts.map

package/dist/scorers-verify-parity.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"scorers-verify-parity.d.ts","sourceRoot":"","sources":["../src/scorers-verify-parity.ts"],"names":[],"mappings":"AAaA;;;;;;;;;;;;;;;GAeG;AACH,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,OAAO,CAAA;IACb,KAAK,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAA;IAC7B,QAAQ,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAA;IAChC,wBAAwB,EAAE,CAAC,eAAe,EAAE,MAAM,KAAK,OAAO,CAAC;QAAE,SAAS,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,IAAI,CAAA;KAAE,CAAC,CAAA;IAC1G,kBAAkB,EAAE,CAAC,SAAS,EAAE,MAAM,KAAK;QAAE,OAAO,EAAE,MAAM,EAAE,CAAC;QAAC,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAA;KAAE,CAAA;IAC/F,mBAAmB,EAAE,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,KAAK,MAAM,CAAC,UAAU,CAAA;IACjF,SAAS,EAAE,MAAM,CAAA;IACjB,cAAc,EAAE,MAAM,CAAA;IACtB,OAAO,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAA;IACjE,wFAAwF;IACxF,YAAY,CAAC,EAAE,MAAM,CAAA;CACtB;AAKD,UAAU,cAAc;IAAG,GAAG,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE;AAmCvD;;;;;;;;GAQG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAMvD;AAED,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,GAAG,cAAc,CAiBhF;AAsDD;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAC1B,eAAe,EAAE,MAAM,EACvB,oBAAoB,EAAE,MAAM,EAC5B,YAAY,EAAE,MAAM,GACnB;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,CAO/C;AAKD;;;;;;;;;;;GAWG;AACH,wBAAgB,0BAA0B,CACxC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACjC,YAAY,EAAE,MAAM,GAAG,IAAI,EAC3B,SAAS,EAAE,MAAM,GAAG,IAAI,GACvB;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CA0BrC;AA8BD,wBAAsB,yBAAyB,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CA0SnG"}