@pome-sh/cli 0.1.0 → 0.1.1

package/dist/build-info.json

@@ -1,6 +1,6 @@
 {
   "package": "pome-sh",
-  "version": "0.1.0",
-  "git_sha": "aa4ca816d3a90491953220fa5e2a3f67ab968a17",
-  "build_time": "2026-07-10T13:24:30.193Z"
+  "version": "0.1.1",
+  "git_sha": "304c1dd67a202ab5c6142553736fd9e85b4d8061",
+  "build_time": "2026-07-10T18:51:42.983Z"
 }

package/dist/src/doctor/checks.js

@@ -97,8 +97,8 @@ async function checkTwinReachable(_configDir) {
     });
     let harness;
     try {
-        // Dynamic import: the twin harness pulls in better-sqlite3, which the
-        // hosted-mode gate must never load better-sqlite3 unnecessarily.
+        // Dynamic import: the twin harness pulls in the twin packages and their
+        // node:sqlite-backed state, which the hosted-mode gate must never load.
         const { bootTwin } = await import("../twin/twinHarness.js");
         harness = await bootTwin({
             twin: "github",

package/node_modules/@pome-sh/sdk/dist/db.d.ts

@@ -1,7 +1,8 @@
 /**
  * Mutation outcome — the shape both better-sqlite3's `RunResult` and
- * node:sqlite's `StatementResultingChanges` return, so the M2 swap (F-703)
- * keeps this interface as-is.
+ * node:sqlite's `StatementResultingChanges` return; `changes` is coerced to
+ * number, `lastInsertRowid` may be a bigint for rowids beyond safe-integer
+ * range.
  */
 export interface TwinRunResult {
     changes: number;
@@ -15,8 +16,7 @@ export interface TwinStatement {
 /**
  * A wrapped transaction function (better-sqlite3's shape). The `.immediate`
  * variant takes the write lock up front (BEGIN IMMEDIATE) — github's domain
- * runs its mutations under it. The M2 node:sqlite implementation backs both
- * call forms with explicit BEGIN [IMMEDIATE]/COMMIT/ROLLBACK.
+ * runs its mutations under it.
  */
 export type TwinTransaction<F extends (...args: never[]) => unknown> = F & {
     immediate: F;

package/node_modules/@pome-sh/sdk/dist/db.js

@@ -1,28 +1,97 @@
 // SPDX-License-Identifier: Apache-2.0
 //
-// Engine SQLite driver wrapper (F-681). Twins never import a sqlite driver
-// directly — they call `openTwinDatabase()` and consume the `TwinDatabase`
-// interface below. The interface is a same-shape subset of better-sqlite3
-// (prepare/exec/pragma/transaction/close), which is exactly the surface the
-// M2 node:sqlite swap (F-703) re-implements: swapping the driver is a change
-// to THIS file only. `transaction(fn)` keeps better-sqlite3's shape (returns
-// a callable running fn atomically) because twin domain code has dozens of
-// call sites relying on it and node:sqlite ships no equivalent — the M2
-// implementation backs it with BEGIN IMMEDIATE/COMMIT/ROLLBACK.
-//
-// The driver is loaded lazily via createRequire so @pome-sh/sdk can declare
-// better-sqlite3 as an optional peer dependency: twins that bring their own
-// database never pay for (or compile) the native module.
+// Engine SQLite driver wrapper (F-681), backed by node:sqlite since F-703 —
+// zero native deps, no compiler toolchain at install. Twins never import a
+// sqlite driver directly — they call `openTwinDatabase()` and consume the
+// `TwinDatabase` interface below. The interface keeps better-sqlite3's shape
+// (prepare/exec/pragma/transaction/close) because twin domain code has dozens
+// of call sites relying on it; node:sqlite ships no pragma()/transaction()
+// equivalents, so this file emulates them: `pragma()` prepares a `PRAGMA …`
+// statement, `transaction(fn)` (+ `.immediate`) wraps fn in explicit
+// BEGIN [IMMEDIATE]/COMMIT/ROLLBACK at the outermost level and in a SAVEPOINT
+// when called inside an open transaction — better-sqlite3's nesting semantics,
+// which twin-github's seed() relies on (it wraps createIssue et al., each of
+// which opens its own transaction).
 import { mkdirSync } from "node:fs";
-import { createRequire } from "node:module";
 import { dirname } from "node:path";
-let driver;
-function loadDriver() {
-    if (!driver) {
-        const require = createRequire(import.meta.url);
-        driver = require("better-sqlite3");
+import { DatabaseSync } from "node:sqlite";
+class NodeSqliteTwinDatabase {
+    #db;
+    #savepointId = 0;
+    constructor(path) {
+        this.#db = new DatabaseSync(path);
+    }
+    prepare(sql) {
+        const statement = this.#db.prepare(sql);
+        return {
+            run: (...params) => {
+                const result = statement.run(...params);
+                return { changes: Number(result.changes), lastInsertRowid: result.lastInsertRowid };
+            },
+            get: (...params) => statement.get(...params),
+            all: (...params) => statement.all(...params),
+        };
+    }
+    exec(sql) {
+        this.#db.exec(sql);
+    }
+    pragma(statement, options = {}) {
+        const rows = this.#db.prepare(`PRAGMA ${statement}`).all();
+        if (!options.simple)
+            return rows;
+        const row = rows[0];
+        if (row === undefined)
+            return undefined;
+        return row[Object.keys(row)[0]];
+    }
+    transaction(fn) {
+        const wrap = (begin) => ((...args) => {
+            if (this.#db.isTransaction)
+                return this.#nested(fn, args);
+            this.#db.exec(begin);
+            try {
+                const out = fn(...args);
+                this.#db.exec("COMMIT");
+                return out;
+            }
+            catch (error) {
+                // Certain failures (e.g. ON CONFLICT ROLLBACK) end the transaction
+                // themselves — only roll back when one is still open.
+                if (this.#db.isTransaction)
+                    this.#db.exec("ROLLBACK");
+                throw error;
+            }
+        });
+        const tx = wrap("BEGIN");
+        tx.immediate = wrap("BEGIN IMMEDIATE");
+        return tx;
+    }
+    /**
+     * A transaction function called inside an open transaction joins it under a
+     * SAVEPOINT (better-sqlite3's semantics): success releases the savepoint,
+     * failure rolls back only the inner work. The BEGIN variant is irrelevant
+     * here — the outermost transaction already holds the lock mode.
+     */
+    #nested(fn, args) {
+        const name = `_pome_tx_${this.#savepointId++}`;
+        this.#db.exec(`SAVEPOINT ${name}`);
+        try {
+            const out = fn(...args);
+            this.#db.exec(`RELEASE ${name}`);
+            return out;
+        }
+        catch (error) {
+            if (this.#db.isTransaction) {
+                this.#db.exec(`ROLLBACK TO ${name}`);
+                this.#db.exec(`RELEASE ${name}`);
+            }
+            throw error;
+        }
+    }
+    close() {
+        if (this.#db.isOpen)
+            this.#db.close();
     }
-    return driver;
 }
 /**
  * Open a twin database with the pome pragma set applied:
@@ -34,11 +103,10 @@ export function openTwinDatabase(path = ":memory:", options = {}) {
     if (path !== ":memory:") {
         mkdirSync(dirname(path), { recursive: true });
     }
-    const Database = loadDriver();
-    const db = new Database(path);
-    db.pragma("busy_timeout = 5000");
-    db.pragma("journal_mode = WAL");
-    db.pragma("foreign_keys = ON");
+    const db = new NodeSqliteTwinDatabase(path);
+    db.exec("PRAGMA busy_timeout = 5000");
+    db.exec("PRAGMA journal_mode = WAL");
+    db.exec("PRAGMA foreign_keys = ON");
     options.migrate?.(db);
     return db;
 }

package/node_modules/@pome-sh/sdk/dist/db.js.map

@@ -1 +1 @@
-{"version":3,"file":"db.js","sourceRoot":"","sources":["../src/db.ts"],"names":[],"mappings":"AAAA,sCAAsC;AACtC,EAAE;AACF,2EAA2E;AAC3E,2EAA2E;AAC3E,0EAA0E;AAC1E,4EAA4E;AAC5E,6EAA6E;AAC7E,6EAA6E;AAC7E,2EAA2E;AAC3E,wEAAwE;AACxE,gEAAgE;AAChE,EAAE;AACF,4EAA4E;AAC5E,4EAA4E;AAC5E,yDAAyD;AAEzD,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AA0CpC,IAAI,MAAqC,CAAC;AAE1C,SAAS,UAAU;IACjB,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC/C,MAAM,GAAG,OAAO,CAAC,gBAAgB,CAAsB,CAAC;IAC1D,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAC9B,IAAI,GAAG,UAAU,EACjB,UAAmC,EAAE;IAErC,IAAI,IAAI,KAAK,UAAU,EAAE,CAAC;QACxB,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAChD,CAAC;IACD,MAAM,QAAQ,GAAG,UAAU,EAAE,CAAC;IAC9B,MAAM,EAAE,GAAG,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC;IAC9B,EAAE,CAAC,MAAM,CAAC,qBAAqB,CAAC,CAAC;IACjC,EAAE,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAChC,EAAE,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC;IAC/B,OAAO,CAAC,OAAO,EAAE,CAAC,EAAE,CAAC,CAAC;IACtB,OAAO,EAAE,CAAC;AACZ,CAAC"}
+{"version":3,"file":"db.js","sourceRoot":"","sources":["../src/db.ts"],"names":[],"mappings":"AAAA,sCAAsC;AACtC,EAAE;AACF,4EAA4E;AAC5E,2EAA2E;AAC3E,0EAA0E;AAC1E,6EAA6E;AAC7E,8EAA8E;AAC9E,2EAA2E;AAC3E,4EAA4E;AAC5E,qEAAqE;AACrE,8EAA8E;AAC9E,+EAA+E;AAC/E,6EAA6E;AAC7E,oCAAoC;AAEpC,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAwC3C,MAAM,sBAAsB;IACjB,GAAG,CAAe;IAC3B,YAAY,GAAG,CAAC,CAAC;IAEjB,YAAY,IAAY;QACtB,IAAI,CAAC,GAAG,GAAG,IAAI,YAAY,CAAC,IAAI,CAAC,CAAC;IACpC,CAAC;IAED,OAAO,CAAC,GAAW;QACjB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QAExC,OAAO;YACL,GAAG,EAAE,CAAC,GAAG,MAAiB,EAAiB,EAAE;gBAC3C,MAAM,MAAM,GAAG,SAAS,CAAC,GAAG,CAAC,GAAI,MAAiB,CAAC,CAAC;gBACpD,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,eAAe,EAAE,MAAM,CAAC,eAAe,EAAE,CAAC;YACtF,CAAC;YACD,GAAG,EAAE,CAAC,GAAG,MAAiB,EAAW,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,GAAI,MAAiB,CAAC;YAC5E,GAAG,EAAE,CAAC,GAAG,MAAiB,EAAa,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,GAAI,MAAiB,CAAC;SAC/E,CAAC;IACJ,CAAC;IAED,IAAI,CAAC,GAAW;QACd,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACrB,CAAC;IAED,MAAM,CAAC,SAAiB,EAAE,UAAgC,EAAE;QAC1D,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,UAAU,SAAS,EAAE,CAAC,CAAC,GAAG,EAAoC,CAAC;QAC7F,IAAI,CAAC,OAAO,CAAC,MAAM;YAAE,OAAO,IAAI,CAAC;QACjC,MAAM,GAAG,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACpB,IAAI,GAAG,KAAK,SAAS;YAAE,OAAO,SAAS,CAAC;QACxC,OAAO,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAW,CAAC,CAAC;IAC5C,CAAC;IAED,WAAW,CAA0C,EAAK;QACxD,MAAM,IAAI,GAAG,CAAC,KAAa,EAAK,EAAE,CAChC,CAAC,CAAC,GAAG,IAAa,EAAE,EAAE;YACpB,IAAI,IAAI,CAAC,GAAG,CAAC,aAAa;gBAAE,OAAO,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACrB,IAAI,CAAC;gBACH,MAAM,GAAG,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;gBACxB,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;gBACxB,OAAO,GAAG,CAAC;YACb,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,mEAAmE;gBACnE,sDAAsD;gBACtD,IAAI,IAAI,CAAC,GAAG,CAAC,aAAa;oBAAE,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;gBACtD,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC,CAAM,CAAC;QACV,MAAM,EAAE,GAAG,IAAI,CAAC,OAAO,CAAuB,CAAC;QAC/C,EAAE,CAAC,SAAS,GAAG,IAAI,CAAC,iBAAiB,CAAC,CAAC;QACvC,OAAO,EAAE,CAAC;IACZ,CAAC;IAED;;;;;OAKG;IACH,OAAO,CAA0C,EAAK,EAAE,IAAa;QACnE,MAAM,IAAI,GAAG,YAAY,IAAI,CAAC,YAAY,EAAE,EAAE,CAAC;QAC/C,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,IAAI,EAAE,CAAC,CAAC;QACnC,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;YACxB,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;YACjC,OAAO,GAAG,CAAC;QACb,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,IAAI,CAAC,GAAG,CAAC,aAAa,EAAE,CAAC;gBAC3B,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,eAAe,IAAI,EAAE,CAAC,CAAC;gBACrC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;YACnC,CAAC;YACD,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED,KAAK;QACH,IAAI,IAAI,CAAC,GAAG,CAAC,MAAM;YAAE,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC;IACxC,CAAC;CACF;AAED;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAC9B,IAAI,GAAG,UAAU,EACjB,UAAmC,EAAE;IAErC,IAAI,IAAI,KAAK,UAAU,EAAE,CAAC;QACxB,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAChD,CAAC;IACD,MAAM,EAAE,GAAG,IAAI,sBAAsB,CAAC,IAAI,CAAC,CAAC;IAC5C,EAAE,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACtC,EAAE,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;IACrC,EAAE,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACpC,OAAO,CAAC,OAAO,EAAE,CAAC,EAAE,CAAC,CAAC;IACtB,OAAO,EAAE,CAAC;AACZ,CAAC"}

package/node_modules/@pome-sh/sdk/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/sdk",
-  "version": "0.3.0",
+  "version": "0.3.1",
   "description": "Twin engine for Pome digital twins — defineTwin() + serve(): HTTP mounting, bearer auth, trace recorder, MCP dispatch, SQLite-backed state.",
   "private": false,
   "type": "module",
@@ -37,15 +37,11 @@
     "zod": "^4.1.13"
   },
   "peerDependencies": {
-    "@hono/node-server": "^2.0.8",
-    "better-sqlite3": "^12.5.0"
+    "@hono/node-server": "^2.0.8"
   },
   "peerDependenciesMeta": {
     "@hono/node-server": {
       "optional": true
-    },
-    "better-sqlite3": {
-      "optional": true
     }
   },
   "engines": {

package/node_modules/@pome-sh/twin-github/README.md

@@ -105,7 +105,8 @@ that pins and verifies the new signed digest.
   build regenerates one on each rebuild)
 - `npm run build` exits 0 and emits `dist/src/server.js`
 - Built output is loadable under Node 24 — the snapshot runs `runtime: "node24"`.
-  Uses Node-native `better-sqlite3` (N-API) in the Node 24 runtime image.
+  SQLite is the built-in `node:sqlite` (via the sdk's `openTwinDatabase()`,
+  F-703) — no native modules, no compiler toolchain.
 
 ### Runtime
 

package/node_modules/@pome-sh/twin-github/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/twin-github",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "Deterministic GitHub-shaped twin for agent testing — REST + MCP, SQLite-backed state.",
   "private": false,
   "type": "module",
@@ -33,9 +33,8 @@
   },
   "dependencies": {
     "@hono/node-server": "^2.0.8",
-    "@pome-sh/sdk": "0.3.0",
+    "@pome-sh/sdk": "0.3.1",
     "@pome-sh/shared-types": "0.6.0",
-    "better-sqlite3": "^12.5.0",
     "hono": "^4.12.27",
     "zod": "^4.1.13"
   },

package/node_modules/@pome-sh/twin-slack/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/twin-slack",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "Deterministic Slack Web API twin for agent testing — REST + MCP, SQLite-backed state.",
   "private": false,
   "type": "module",
@@ -33,9 +33,8 @@
   },
   "dependencies": {
     "@hono/node-server": "^2.0.8",
-    "@pome-sh/sdk": "0.3.0",
+    "@pome-sh/sdk": "0.3.1",
     "@pome-sh/shared-types": "0.6.0",
-    "better-sqlite3": "^12.5.0",
     "hono": "^4.12.27",
     "zod": "^4.1.13"
   },

package/node_modules/@pome-sh/twin-stripe/README.md

@@ -182,8 +182,8 @@ that pins and verifies the new signed digest.
   required, the snapshot build regenerates one on each rebuild).
 - `npm run build` exits 0 and emits `dist/src/server.js`.
 - Built output is loadable under Node 24 — the snapshot runs
-  `runtime: "node24"`. Uses Node-native `better-sqlite3` (
-  fine; we install python3 + make + g++ in the Dockerfile build stage).
+  `runtime: "node24"`. SQLite is the built-in `node:sqlite` (via the sdk's
+  `openTwinDatabase()`, F-703) — no native modules, no compiler toolchain.
 
 ### Runtime
 

package/node_modules/@pome-sh/twin-stripe/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/twin-stripe",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "description": "Deterministic test double for Stripe x402 machine payments — stateful clone agents can drive without burning sandbox quota or real chain gas.",
   "private": false,
   "type": "module",
@@ -52,9 +52,8 @@
   "license": "Apache-2.0",
   "dependencies": {
     "@hono/node-server": "^2.0.8",
-    "@pome-sh/sdk": "0.3.0",
+    "@pome-sh/sdk": "0.3.1",
     "@pome-sh/shared-types": "0.6.0",
-    "better-sqlite3": "^12.5.0",
     "hono": "^4.12.27",
     "zod": "^4.1.13"
   },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/cli",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "description": "Digital-twin testing for AI agents — run scenarios against resettable local or hosted twins, record tool calls, and score the result.",
   "keywords": [
     "ai",
@@ -52,6 +52,7 @@
     "test": "vitest run",
     "test:e2e": "vitest run test/e2e",
     "gate:no-eval": "node ../scripts/no-eval-in-oss.mjs",
+    "gate:no-native": "node ../scripts/no-native-modules.mjs --root .",
     "gate:recorder-overhead": "tsx scripts/recorder-overhead-gate.ts",
     "changeset": "changeset",
     "changeset:version": "changeset version",
@@ -65,13 +66,12 @@
     "@anthropic-ai/sdk": "^0.110.0",
     "@hono/node-server": "^2.0.6",
     "@openrouter/ai-sdk-provider": "^3.0.0",
-    "@pome-sh/sdk": "0.3.0",
+    "@pome-sh/sdk": "0.3.1",
     "@pome-sh/shared-types": "0.6.0",
-    "@pome-sh/twin-github": "0.1.1",
-    "@pome-sh/twin-slack": "0.1.1",
-    "@pome-sh/twin-stripe": "0.2.1",
+    "@pome-sh/twin-github": "0.1.2",
+    "@pome-sh/twin-slack": "0.1.2",
+    "@pome-sh/twin-stripe": "0.2.2",
     "ai": "^7.0.18",
-    "better-sqlite3": "^12.5.0",
     "commander": "^15.0.0",
     "hono": "^4.10.7",
     "yaml": "^2.8.2",
@@ -80,7 +80,6 @@
   "devDependencies": {
     "@changesets/changelog-github": "^0.5.1",
     "@changesets/cli": "^2.31.0",
-    "@types/better-sqlite3": "^7.6.13",
     "@types/node": "^26.0.1",
     "tsx": "^4.21.0",
     "typescript": "^5.9.3",

package/dist/src/recorder/verdictArtifact.d.ts

@@ -1,71 +0,0 @@
-import type { CriterionResult } from "../types/shared.js";
-export declare const VERDICT_ARTIFACT_VERSION = 1;
-export declare const VERDICT_FILENAME = "verdict.json";
-export interface VerdictArtifact {
-    version: number;
-    /** Provenance: the only writer is the /finalize response path. */
-    source: "cloud-finalize";
-    task_name: string;
-    /** The scenario path as the run invocation saw it (may move later —
-     *  readers must tolerate a dangling path). */
-    scenario_path: string;
-    /** Shared trial-group id (`grp_` + nanoid21); null on single runs. */
-    group_id: string | null;
-    session_id: string;
-    cloud_run_id: string;
-    cloud_dashboard_url: string;
-    judge_model: string | null;
-    /** Cloud-authoritative satisfaction score, 0-100. */
-    score: number;
-    pass_threshold: number;
-    passed: boolean;
-    criteria_results: CriterionResult[];
-    duration_ms: number;
-    finalized_at: string;
-}
-export interface TrialVerdict {
-    runDir: string;
-    verdict: VerdictArtifact;
-}
-export declare function writeVerdictArtifact(runDir: string, verdict: VerdictArtifact): Promise<void>;
-export declare function readVerdictArtifact(runDir: string): Promise<TrialVerdict | null>;
-/** Scan an artifacts root for finalized trials — exactly the two-level
- *  layout `<root>/<task-slug>/<session-id>/verdict.json`. Anything
- *  unreadable is skipped. */
-export declare function scanVerdictArtifacts(artifactsRoot: string): Promise<TrialVerdict[]>;
-export interface RunSet {
-    /** null = a single run that never had a group. */
-    groupId: string | null;
-    taskName: string;
-    /** The scenario path recorded at run time (first trial's). */
-    scenarioPath: string;
-    /** Trials sorted by finalized_at ascending. */
-    trials: TrialVerdict[];
-    latestFinalizedAt: string;
-    anyFailed: boolean;
-}
-/** Group trials into run sets: trials sharing a group_id form one set; a
- *  null group_id is its own single-run set. */
-export declare function groupRunSets(trials: TrialVerdict[]): RunSet[];
-/** The newest run set with at least one failed (completed) trial. */
-export declare function latestFailedRunSet(sets: RunSet[]): RunSet | null;
-export interface RunSetDiscovery {
-    kind: "trial-dir" | "root";
-    /** The set to build a fix prompt from (per `kind` semantics: a trial dir
-     *  targets ITS set regardless of outcome; a root targets the latest
-     *  FAILED set). Null when nothing matches. */
-    set: RunSet | null;
-    /** Total finalized run sets seen — lets the caller distinguish "no runs
-     *  at all" from "runs exist but none failed". */
-    totalSets: number;
-}
-/**
- * Resolve what `pome fix-prompt [target]` should read.
- * - `target` = a trial run dir (has verdict.json): that trial's whole set —
- *   the user pointed at it, outcome doesn't matter.
- * - `target` = an artifacts root: the latest failed set under it.
- */
-export declare function discoverRunSet(target: string): Promise<RunSetDiscovery>;
-/** Load a trial's captured events.jsonl (raw trace) for prompt assembly.
- *  Missing/corrupt lines are skipped — the prompt degrades, never throws. */
-export declare function loadTrialEvents(runDir: string): Promise<unknown[]>;

package/dist/src/recorder/verdictArtifact.js

@@ -1,207 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// FDRS-644 — the per-trial CLOUD verdict artifact.
-//
-// Hosted `pome run` persists each finalized trial's cloud verdict payload
-// (what /finalize returned and the terminal rendered) as `verdict.json`
-// next to the trial's raw trace. This is a provenance-labeled CACHE of the
-// cloud judge's output — NOT a local score: the OSS CLI still has no
-// scoring engine (FDRS-657 / no-eval-in-oss), score.json is still never
-// written, and deleting verdict.json loses nothing the dashboard doesn't
-// hold. `pome fix-prompt` reads these to hand grouped failure signatures
-// to the user's coding agent without a credentialed cloud round-trip —
-// the F0-3 / L5 intent recorded in uploadAndFinalize.ts ("so the CLI can
-// render per-criterion verdicts … without a follow-up cloud round-trip").
-//
-// Artifact-layout knowledge lives HERE and only here:
-//   <artifacts-root>/<task-slug>/<session-id>/verdict.json
-// (runDir shape from recorder/artifacts.ts `writeRunArtifactsCore`).
-import { existsSync } from "node:fs";
-import { readdir, readFile, writeFile } from "node:fs/promises";
-import { join } from "node:path";
-export const VERDICT_ARTIFACT_VERSION = 1;
-export const VERDICT_FILENAME = "verdict.json";
-export async function writeVerdictArtifact(runDir, verdict) {
-    await writeFile(join(runDir, VERDICT_FILENAME), `${JSON.stringify(verdict, null, 2)}\n`, "utf8");
-}
-/** Read one trial's verdict.json. Returns null when absent or when the file
- *  isn't a recognizable verdict artifact (foreign/corrupt files are skipped,
- *  never thrown on — fix-prompt discovery must survive a messy runs/). */
-/** Every field the fix-prompt pipeline dereferences must hold its declared
- *  shape, or the FILE is rejected — discovery treats a half-recognizable
- *  verdict.json as foreign rather than crashing downstream on it. */
-function isVerdictArtifact(parsed) {
-    if (typeof parsed !== "object" || parsed === null)
-        return false;
-    const v = parsed;
-    if (v.source !== "cloud-finalize")
-        return false;
-    if (typeof v.session_id !== "string")
-        return false;
-    if (typeof v.task_name !== "string")
-        return false;
-    if (typeof v.scenario_path !== "string")
-        return false;
-    if (v.group_id !== null && typeof v.group_id !== "string")
-        return false;
-    if (typeof v.finalized_at !== "string")
-        return false;
-    if (typeof v.passed !== "boolean")
-        return false;
-    if (typeof v.score !== "number")
-        return false;
-    if (!Array.isArray(v.criteria_results))
-        return false;
-    return v.criteria_results.every((r) => {
-        if (typeof r !== "object" || r === null)
-            return false;
-        const result = r;
-        const criterion = result.criterion;
-        return (typeof criterion === "object" &&
-            criterion !== null &&
-            typeof criterion.text === "string" &&
-            typeof result.reason === "string" &&
-            typeof result.passed === "boolean" &&
-            typeof result.skipped === "boolean");
-    });
-}
-export async function readVerdictArtifact(runDir) {
-    const path = join(runDir, VERDICT_FILENAME);
-    let raw;
-    try {
-        raw = await readFile(path, "utf8");
-    }
-    catch {
-        return null;
-    }
-    try {
-        const parsed = JSON.parse(raw);
-        if (!isVerdictArtifact(parsed))
-            return null;
-        return { runDir, verdict: parsed };
-    }
-    catch {
-        return null;
-    }
-}
-/** Scan an artifacts root for finalized trials — exactly the two-level
- *  layout `<root>/<task-slug>/<session-id>/verdict.json`. Anything
- *  unreadable is skipped. */
-export async function scanVerdictArtifacts(artifactsRoot) {
-    const found = [];
-    let slugs;
-    try {
-        slugs = await readdir(artifactsRoot);
-    }
-    catch {
-        return found;
-    }
-    for (const slug of slugs) {
-        const slugDir = join(artifactsRoot, slug);
-        let runIds;
-        try {
-            runIds = await readdir(slugDir);
-        }
-        catch {
-            continue;
-        }
-        for (const runId of runIds) {
-            const trial = await readVerdictArtifact(join(slugDir, runId));
-            if (trial)
-                found.push(trial);
-        }
-    }
-    return found;
-}
-/** Group trials into run sets: trials sharing a group_id form one set; a
- *  null group_id is its own single-run set. */
-export function groupRunSets(trials) {
-    const byKey = new Map();
-    for (const trial of trials) {
-        const key = trial.verdict.group_id ?? `solo:${trial.verdict.session_id}`;
-        const bucket = byKey.get(key);
-        if (bucket)
-            bucket.push(trial);
-        else
-            byKey.set(key, [trial]);
-    }
-    const sets = [];
-    for (const bucket of byKey.values()) {
-        bucket.sort((a, b) => a.verdict.finalized_at.localeCompare(b.verdict.finalized_at));
-        const last = bucket[bucket.length - 1];
-        sets.push({
-            groupId: bucket[0].verdict.group_id,
-            taskName: bucket[0].verdict.task_name,
-            scenarioPath: bucket[0].verdict.scenario_path,
-            trials: bucket,
-            latestFinalizedAt: last.verdict.finalized_at,
-            anyFailed: bucket.some((t) => !t.verdict.passed),
-        });
-    }
-    sets.sort((a, b) => a.latestFinalizedAt.localeCompare(b.latestFinalizedAt));
-    return sets;
-}
-/** The newest run set with at least one failed (completed) trial. */
-export function latestFailedRunSet(sets) {
-    for (let i = sets.length - 1; i >= 0; i -= 1) {
-        if (sets[i].anyFailed)
-            return sets[i];
-    }
-    return null;
-}
-/**
- * Resolve what `pome fix-prompt [target]` should read.
- * - `target` = a trial run dir (has verdict.json): that trial's whole set —
- *   the user pointed at it, outcome doesn't matter.
- * - `target` = an artifacts root: the latest failed set under it.
- */
-export async function discoverRunSet(target) {
-    const anchor = await readVerdictArtifact(target);
-    if (anchor) {
-        // Trial dir → its set. Layout is <root>/<slug>/<runId>, so the root is
-        // two levels up; a moved/isolated dir degrades to a set of one.
-        const root = join(target, "..", "..");
-        const sets = groupRunSets(await scanVerdictArtifacts(root));
-        const own = sets.find((s) => (anchor.verdict.group_id !== null &&
-            s.groupId === anchor.verdict.group_id) ||
-            (anchor.verdict.group_id === null &&
-                s.trials.length === 1 &&
-                s.trials[0].verdict.session_id === anchor.verdict.session_id)) ?? groupRunSets([anchor])[0];
-        return { kind: "trial-dir", set: own, totalSets: Math.max(sets.length, 1) };
-    }
-    if (!existsSync(target))
-        return { kind: "root", set: null, totalSets: 0 };
-    const sets = groupRunSets(await scanVerdictArtifacts(target));
-    return {
-        kind: "root",
-        set: latestFailedRunSet(sets),
-        totalSets: sets.length,
-    };
-}
-/** Load a trial's captured events.jsonl (raw trace) for prompt assembly.
- *  Missing/corrupt lines are skipped — the prompt degrades, never throws. */
-export async function loadTrialEvents(runDir) {
-    let raw;
-    try {
-        raw = await readFile(join(runDir, "events.jsonl"), "utf8");
-    }
-    catch {
-        return [];
-    }
-    const events = [];
-    for (const line of raw.split("\n")) {
-        const trimmed = line.trim();
-        if (!trimmed)
-            continue;
-        try {
-            const parsed = JSON.parse(trimmed);
-            // Valid JSON but not an event object (`null`, `3`, `"x"`) is corrupt
-            // for our purposes — renderEvent dereferences fields on it.
-            if (typeof parsed === "object" && parsed !== null)
-                events.push(parsed);
-        }
-        catch {
-            // skip corrupt row
-        }
-    }
-    return events;
-}

package/dist/src/score/view.d.ts

@@ -1,38 +0,0 @@
-import type { z } from "zod";
-import type { criterionSchema } from "../types/shared.js";
-type WireCriterion = z.infer<typeof criterionSchema>;
-export type CriterionOutcome = "passed" | "failed" | "skipped" | "errored";
-export type CriterionResult = {
-    criterion: WireCriterion;
-    outcome?: CriterionOutcome;
-    passed: boolean;
-    skipped: boolean;
-    reason: string;
-    confidence?: number;
-    judge_model?: string;
-    judge_tokens_in?: number;
-    judge_tokens_out?: number;
-    judge_has_usage?: boolean;
-};
-export type Score = {
-    satisfaction: number;
-    passed: number;
-    failed: number;
-    skipped: number;
-    errored: number;
-    total_required: number;
-    evaluated: boolean;
-    can_pass: boolean;
-    results: CriterionResult[];
-    judge_model: string | null;
-    judge_tokens_in: number | null;
-    judge_tokens_out: number | null;
-};
-export declare function outcomeOf(result: CriterionResult): CriterionOutcome;
-export type ScoreStatus = "pass" | "fail" | "unevaluated";
-export declare function scoreStatus(score: Score, passThreshold: number): ScoreStatus;
-export declare function scenarioPassed(score: Score, passThreshold: number): boolean;
-export declare function markerFor(outcome: CriterionOutcome): string;
-export declare function scoreCountsSummary(score: Score): string;
-export declare function runScoreLine(score: Score, passThreshold: number, unevaluatedNumericLabel: string): string;
-export {};