@agent-nexus/driftbench 0.0.0-stage → 0.1.0

package/dist/index.d.ts

@@ -0,0 +1,37 @@
+/** Where a command writes, and the folder relative paths are shown from. Tests pass their own. */
+type Io = {
+    out: (text: string) => void;
+    err: (text: string) => void;
+    readonly cwd: string;
+};
+declare const processIo: Io;
+
+type CoverageOptions = {
+    /** Path to the behaviour inventory (YAML or JSON). */
+    inventory: string;
+    /** Limit the matrix to one channel. */
+    channel?: string;
+    json?: boolean;
+};
+/**
+ * `driftbench spec coverage <paths...> --inventory <file>`: the behaviour x spec matrix and the
+ * flow x deviation matrix. Exit 0 when every applicable behaviour and every required deviation is
+ * covered, 1 when one is missing, 2 on a usage or input error.
+ */
+declare function coverageCommand(paths: string[], options: CoverageOptions, io: Io): Promise<number>;
+
+declare function validateCommand(paths: string[], options: {
+    json?: boolean;
+}, io: Io): Promise<number>;
+declare function lintCommand(paths: string[], options: {
+    json?: boolean;
+    strict?: boolean;
+}, io: Io): Promise<number>;
+
+/** Runs the CLI with arguments (without `node` and the script) and returns the exit code. */
+declare function main(argv: string[], io?: Io): Promise<number>;
+
+/** Kept equal to package.json by a test. */
+declare const VERSION = "0.1.0";
+
+export { type CoverageOptions, type Io, VERSION, coverageCommand, lintCommand, main, processIo, validateCommand };

package/dist/index.js

@@ -0,0 +1,16 @@
+import {
+  VERSION,
+  coverageCommand,
+  lintCommand,
+  main,
+  processIo,
+  validateCommand
+} from "./chunk-5DPBYSU3.js";
+export {
+  VERSION,
+  coverageCommand,
+  lintCommand,
+  main,
+  processIo,
+  validateCommand
+};

package/dist/random-agent-E43SQJDO.js

@@ -0,0 +1,66 @@
+import {
+  createAcmeAgent,
+  createStore
+} from "./chunk-57ZRX7PY.js";
+
+// ../../examples/acme/pilot/random-agent/index.ts
+import { createHash } from "crypto";
+import { PILOT_CONTROL_MARKER } from "@agent-nexus/driftbench-generate";
+var RANDOM_AGENT_DEFAULTS = {
+  seed: 7,
+  fragile: 0.5,
+  pFragile: 0.5,
+  pSturdy: 0.05
+};
+function uniform(...parts) {
+  const digest = createHash("sha256").update(parts.join("\0")).digest();
+  return digest.readUInt32BE(0) / 2 ** 32;
+}
+function createRandomAgent(options = {}) {
+  const o = { ...RANDOM_AGENT_DEFAULTS, ...options };
+  for (const key of ["fragile", "pFragile", "pSturdy"]) {
+    if (!(o[key] >= 0 && o[key] <= 1)) throw new RangeError(`createRandomAgent: ${key} must lie in [0, 1]`);
+  }
+  const opened = /* @__PURE__ */ new Map();
+  return {
+    options: o,
+    create(given) {
+      const k = opened.get(given.hash) ?? 0;
+      opened.set(given.hash, k + 1);
+      let draw = 0;
+      const store = createStore();
+      return {
+        agent: createAcmeAgent(store),
+        store,
+        given: given.hash,
+        next: () => {
+          draw += 1;
+          return uniform("run", o.seed, given.hash, k, draw);
+        }
+      };
+    },
+    async respond(messages, world, session) {
+      const observed = await session.agent.respond(messages, world);
+      const texts = messages.filter((message) => message.role === "user").map((message) => message.content);
+      const fragile = uniform("wording", o.seed, session.given, ...texts) < o.fragile;
+      const pFail = fragile ? o.pFragile : o.pSturdy;
+      const failed = session.next() < pFail;
+      const text = failed ? `${observed.text} ${PILOT_CONTROL_MARKER}` : observed.text;
+      const segments = observed.segments.map(
+        (segment, i) => failed && segment.type === "text" && i === observed.segments.length - 1 ? { ...segment, text } : segment
+      );
+      const raw = observed.raw !== null && typeof observed.raw === "object" ? observed.raw : {};
+      return {
+        ...observed,
+        segments,
+        text,
+        raw: { ...raw, pilotControl: { pFail, fragile, failed, model: { ...o } } }
+      };
+    }
+  };
+}
+export {
+  RANDOM_AGENT_DEFAULTS,
+  createRandomAgent,
+  uniform
+};

package/dist/src-5XXPWAL2.js

@@ -0,0 +1,26 @@
+import {
+  ACME,
+  SIDE_EFFECT_TOOLS,
+  STAGES,
+  TOOL_LABELS,
+  callTool,
+  createAcmeAgent,
+  createStore,
+  initialState,
+  isToolLabel,
+  takeTurn,
+  withSeed
+} from "./chunk-57ZRX7PY.js";
+export {
+  ACME,
+  SIDE_EFFECT_TOOLS,
+  STAGES,
+  TOOL_LABELS,
+  callTool,
+  createAcmeAgent,
+  createStore,
+  initialState,
+  isToolLabel,
+  takeTurn,
+  withSeed
+};

package/package.json

@@ -1,6 +1,68 @@
 {
   "name": "@agent-nexus/driftbench",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
+  "version": "0.1.0",
+  "description": "The driftbench command line: validate and run specs, campaigns in teacher-forced, free-running and perturbed modes, reports, generation and validators.",
+  "keywords": [
+    "driftbench",
+    "ai-agents",
+    "evaluation",
+    "llm",
+    "testing",
+    "cli",
+    "command-line",
+    "continuous-integration"
+  ],
+  "homepage": "https://github.com/achammah/driftbench#readme",
+  "bugs": {
+    "url": "https://github.com/achammah/driftbench/issues"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/achammah/driftbench.git",
+    "directory": "packages/cli"
+  },
+  "license": "Apache-2.0",
+  "type": "module",
+  "engines": {
+    "node": ">=22.13"
+  },
+  "bin": {
+    "driftbench": "./dist/bin.js"
+  },
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    }
+  },
+  "files": [
+    "dist"
+  ],
+  "dependencies": {
+    "commander": "^15.0.0",
+    "yaml": "^2.9.1",
+    "@agent-nexus/driftbench-adapters": "0.1.0",
+    "@agent-nexus/driftbench-checks": "0.1.0",
+    "@agent-nexus/driftbench-core": "0.1.0",
+    "@agent-nexus/driftbench-artifacts": "0.1.0",
+    "@agent-nexus/driftbench-generate": "0.1.0",
+    "@agent-nexus/driftbench-estate": "0.1.0",
+    "@agent-nexus/driftbench-judges": "0.1.0",
+    "@agent-nexus/driftbench-report": "0.1.0",
+    "@agent-nexus/driftbench-space": "0.1.0",
+    "@agent-nexus/driftbench-stats": "0.1.0",
+    "@agent-nexus/driftbench-users": "0.1.0",
+    "@agent-nexus/driftbench-store": "0.1.0",
+    "@agent-nexus/driftbench-validators": "0.1.0",
+    "@agent-nexus/driftbench-sandbox": "0.1.0"
+  },
+  "publishConfig": {
+    "access": "public",
+    "provenance": true
+  },
+  "scripts": {
+    "build": "tsup src/index.ts src/bin.ts --format esm --dts --clean --target node22"
+  }
 }