mneme-ai 1.4.0 → 1.6.0

package/bin/git-mneme.js

@@ -0,0 +1,63 @@
+#!/usr/bin/env node
+/**
+ * git-mneme — the git subcommand wrapper.
+ *
+ * Allows users to type `git mneme <subcommand>` instead of `mneme <subcommand>`.
+ * git treats any binary in PATH named `git-<name>` as a subcommand, so once
+ * Mneme is installed globally + this script is on PATH, `git mneme why src/auth.ts`
+ * works identically to `mneme why src/auth.ts`.
+ *
+ * This is the foundation of the v1.5.0 "stand beside git" positioning:
+ * Mneme is no longer just an MCP plugin for AI tools — it's a git extension
+ * that any developer using git, on any platform (GitHub / GitLab / Bitbucket
+ * / Gitea / self-hosted), can install and benefit from.
+ *
+ * Implementation: pure pass-through to the same dist/index.js entrypoint
+ * with argv shifted by one (drops the leading 'mneme' that git would
+ * usually pass via argv[2]). All flags, all commands, all behaviors
+ * identical to invoking `mneme` directly.
+ */
+
+// Same SQLite-warning suppression as mneme.js — keep behavior identical.
+const originalEmit = process.emit;
+process.emit = function (name, data, ...rest) {
+  if (
+    name === "warning" &&
+    typeof data === "object" &&
+    data !== null &&
+    data.name === "ExperimentalWarning" &&
+    typeof data.message === "string" &&
+    data.message.includes("SQLite is an experimental feature")
+  ) {
+    return false;
+  }
+  return originalEmit.call(this, name, data, ...rest);
+};
+
+// argv layout when invoked as a git subcommand:
+//   process.argv[0] = node
+//   process.argv[1] = git-mneme.js
+//   process.argv[2..] = the subcommand + flags ('why', 'audit', etc)
+// The dist/index.js parser expects the same shape as `mneme`, so we just
+// pass argv through unchanged after re-aliasing argv[1] for nicer error
+// messages ("mneme" not "git-mneme.js").
+const passthrough = ["mneme", ...process.argv.slice(2)];
+
+// --version fast path — same as mneme.js
+const arg = process.argv[2];
+if (process.argv.length === 3 && (arg === "--version" || arg === "-V")) {
+  import("node:fs").then((fs) =>
+    import("node:url").then((urlMod) =>
+      import("node:path").then((path) => {
+        const here = path.dirname(urlMod.fileURLToPath(import.meta.url));
+        const pkg = JSON.parse(
+          fs.readFileSync(path.join(here, "..", "package.json"), "utf8"),
+        );
+        process.stdout.write(pkg.version + "\n");
+        process.exit(0);
+      }),
+    ),
+  );
+} else {
+  import("../dist/index.js").then((m) => m.run(passthrough));
+}

package/dist/commands/benchmark.d.ts

@@ -0,0 +1,83 @@
+/**
+ * `mneme benchmark` — the AI Memory Benchmark.
+ *
+ * Strategic positioning: the Lighthouse-of-AI-memory. Mneme grades AI
+ * memory implementations (Mneme itself, Claude's native context, GPT's,
+ * Cursor's bundled, etc.) using a standardized probe set + the same
+ * Super Sonic Engine grader that already ships in Mneme.
+ *
+ * Why this is uniquely Mneme's seat: when every AI vendor ships native
+ * "repo memory", who grades them? Anthropic can't grade Anthropic;
+ * OpenAI can't grade OpenAI. Mneme is the only vendor-neutral auditor
+ * — and the only one that can publish a fair public leaderboard.
+ *
+ * v0.1 ships:
+ *   • Probe set (24 memory questions across 6 categories)
+ *   • Local-mode benchmarking (grades Mneme itself end-to-end)
+ *   • Markdown leaderboard generation
+ *   • Stub for future "external mode" that grades third-party memory
+ *     implementations via their public APIs
+ */
+export interface BenchmarkOptions {
+    cwd: string;
+    /** Compare against named external implementations.
+     *  v0.1 supports only "mneme-self". v1.0 will add claude / gpt / cursor. */
+    targets?: string[];
+    /** Output markdown leaderboard to this path */
+    out?: string;
+    /** Machine-readable JSON */
+    json?: boolean;
+    /** Number of probes to run (default: all 24) */
+    probes?: number;
+}
+interface Probe {
+    id: string;
+    category: "factual-recall" | "causal-explanation" | "lineage-trace" | "regression-prediction" | "cited-rationale" | "uncertainty-honesty";
+    question: string;
+    /** Expected qualities the answer must satisfy. Each is a binary check. */
+    rubric: ProbeRubric[];
+}
+interface ProbeRubric {
+    id: string;
+    /** Pass = answer contains AT LEAST ONE of these regex patterns */
+    must_match?: RegExp[];
+    /** Pass = answer contains NONE of these regex patterns */
+    must_not_match?: RegExp[];
+    /** Description shown in the report */
+    description: string;
+}
+interface ProbeResult {
+    probeId: string;
+    category: Probe["category"];
+    question: string;
+    answer: string;
+    passed: number;
+    total: number;
+    failedRubrics: Array<{
+        id: string;
+        description: string;
+    }>;
+    durationMs: number;
+}
+interface BenchmarkResult {
+    target: string;
+    ranAt: string;
+    totalProbes: number;
+    totalRubricChecks: number;
+    passedRubricChecks: number;
+    scoreByCategory: Record<string, {
+        passed: number;
+        total: number;
+        pct: number;
+    }>;
+    overallScore: number;
+    probeResults: ProbeResult[];
+}
+declare function checkRubric(rubric: ProbeRubric, answer: string): boolean;
+declare function renderMarkdownLeaderboard(results: BenchmarkResult[]): string;
+export declare function benchmarkCommand(opts: BenchmarkOptions): Promise<number>;
+export declare const _PROBES_FOR_TESTS: Probe[];
+export declare const _checkRubricForTests: typeof checkRubric;
+export declare const _renderMarkdownLeaderboardForTests: typeof renderMarkdownLeaderboard;
+export {};
+//# sourceMappingURL=benchmark.d.ts.map

package/dist/commands/benchmark.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"benchmark.d.ts","sourceRoot":"","sources":["../../src/commands/benchmark.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AASH,MAAM,WAAW,gBAAgB;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ;gFAC4E;IAC5E,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,+CAA+C;IAC/C,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,4BAA4B;IAC5B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,gDAAgD;IAChD,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,UAAU,KAAK;IACb,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EACJ,gBAAgB,GAChB,oBAAoB,GACpB,eAAe,GACf,uBAAuB,GACvB,iBAAiB,GACjB,qBAAqB,CAAC;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,0EAA0E;IAC1E,MAAM,EAAE,WAAW,EAAE,CAAC;CACvB;AAED,UAAU,WAAW;IACnB,EAAE,EAAE,MAAM,CAAC;IACX,kEAAkE;IAClE,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,0DAA0D;IAC1D,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,sCAAsC;IACtC,WAAW,EAAE,MAAM,CAAC;CACrB;AAyUD,UAAU,WAAW;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,KAAK,CAAC,UAAU,CAAC,CAAC;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC1D,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,UAAU,eAAe;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,eAAe,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,GAAG,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAChF,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,WAAW,EAAE,CAAC;CAC7B;AAED,iBAAS,WAAW,CAAC,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAUjE;AAiFD,iBAAS,yBAAyB,CAAC,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,CAuErE;AAED,wBAAsB,gBAAgB,CAAC,IAAI,EAAE,gBAAgB,GAAG,OAAO,CAAC,MAAM,CAAC,CAwD9E;AAED,eAAO,MAAM,iBAAiB,SAAS,CAAC;AACxC,eAAO,MAAM,oBAAoB,oBAAc,CAAC;AAChD,eAAO,MAAM,kCAAkC,kCAA4B,CAAC"}