@nodetool-ai/runtime 0.7.0-rc.23 → 0.8.1

package/dist/providers/contract/probe-runner.js

@@ -0,0 +1,180 @@
+/**
+ * Runs the live half of the provider contract probes.
+ *
+ * One request per manifest entry, decoded by the production decoder the entry
+ * names. The two ways a probe fails are reported apart: a **network failure**
+ * means we never got a body to judge (DNS, timeout, 5xx, a non-JSON gateway
+ * page), and a **schema failure** means the provider answered and the response
+ * no longer decodes into what NodeTool needs. Only the second is a contract
+ * break; treating them alike would turn every flaky night into a false alarm.
+ *
+ * Nothing here writes a cassette, and nothing retains a response body: what a
+ * run keeps is the shape (see `summarizeShape`).
+ */
+import { redactText, summarizeShape } from "./redact.js";
+import { PROBE_MANIFEST } from "./probe-manifest.js";
+/** A body that never reached the decoder: the run has nothing to judge. */
+class NetworkProbeError extends Error {
+}
+async function readJsonBody(response, entry) {
+    const acceptsHttpError = entry.live?.acceptsHttpError ?? false;
+    const text = await response.text();
+    if (!response.ok && !acceptsHttpError) {
+        throw new NetworkProbeError(`HTTP ${response.status}: ${redactText(text, 200)}`);
+    }
+    try {
+        return JSON.parse(text);
+    }
+    catch {
+        // A gateway error page is a network-layer problem, not a contract change.
+        throw new NetworkProbeError(`response body is not JSON (HTTP ${response.status}): ${redactText(text, 200)}`);
+    }
+}
+export async function runProbes(options = {}) {
+    const env = options.env ?? {};
+    const fetchImpl = options.fetchImpl ?? globalThis.fetch.bind(globalThis);
+    const entries = options.entries ?? PROBE_MANIFEST;
+    const timeoutMs = options.timeoutMs ?? 60_000;
+    const now = options.now ?? (() => Date.now());
+    const results = [];
+    const perProvider = {};
+    for (const entry of entries) {
+        const spent = (perProvider[entry.provider] ??= { requests: 0, costUsd: 0 });
+        const base = {
+            id: entry.id,
+            provider: entry.provider,
+            target: entry.target,
+            decoder: entry.decoder
+        };
+        const live = entry.live;
+        if (!live) {
+            results.push({
+                ...base,
+                status: "skipped",
+                reason: `fixture only: ${entry.liveGap ?? "no live probe declared"}`,
+                durationMs: 0,
+                requests: 0,
+                costUsd: 0
+            });
+            continue;
+        }
+        const credential = live.credential ? (env[live.credential] ?? "") : "";
+        if (live.credential && !credential) {
+            results.push({
+                ...base,
+                status: "skipped",
+                reason: `${live.credential} is not set`,
+                durationMs: 0,
+                requests: 0,
+                costUsd: 0
+            });
+            continue;
+        }
+        if (spent.requests + 1 > live.maxRequests) {
+            results.push({
+                ...base,
+                status: "skipped",
+                reason: `request budget for ${entry.provider} is spent (${live.maxRequests})`,
+                durationMs: 0,
+                requests: 0,
+                costUsd: 0
+            });
+            continue;
+        }
+        if (spent.costUsd + live.estimatedCostUsd > live.maxCostUsd) {
+            results.push({
+                ...base,
+                status: "skipped",
+                reason: `cost budget for ${entry.provider} is spent (USD ${live.maxCostUsd})`,
+                durationMs: 0,
+                requests: 0,
+                costUsd: 0
+            });
+            continue;
+        }
+        const startedAt = now();
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), timeoutMs);
+        let status = "passed";
+        let reason;
+        let shape;
+        try {
+            const { url, init } = live.request(env, credential);
+            let body;
+            try {
+                const response = await fetchImpl(url, {
+                    ...init,
+                    signal: controller.signal
+                });
+                body = await readJsonBody(response, entry);
+            }
+            catch (err) {
+                throw err instanceof NetworkProbeError
+                    ? err
+                    : new NetworkProbeError(err.message);
+            }
+            shape = summarizeShape(body);
+            entry.check(body);
+        }
+        catch (err) {
+            status = err instanceof NetworkProbeError ? "network-failure" : "schema-failure";
+            reason = redactText(err.message);
+        }
+        finally {
+            clearTimeout(timer);
+        }
+        spent.requests += 1;
+        spent.costUsd += live.estimatedCostUsd;
+        results.push({
+            ...base,
+            status,
+            reason,
+            durationMs: now() - startedAt,
+            requests: 1,
+            costUsd: live.estimatedCostUsd,
+            shape
+        });
+    }
+    const totals = {
+        passed: 0,
+        schemaFailures: 0,
+        networkFailures: 0,
+        skipped: 0,
+        requests: 0,
+        costUsd: 0
+    };
+    for (const r of results) {
+        if (r.status === "passed")
+            totals.passed++;
+        else if (r.status === "schema-failure")
+            totals.schemaFailures++;
+        else if (r.status === "network-failure")
+            totals.networkFailures++;
+        else if (r.status === "skipped")
+            totals.skipped++;
+        totals.requests += r.requests;
+        totals.costUsd += r.costUsd;
+    }
+    return {
+        startedAt: new Date(now()).toISOString(),
+        results,
+        totals,
+        perProvider
+    };
+}
+/** One line per probe, for a CI log. */
+export function formatProbeReport(report) {
+    const icon = {
+        passed: "✓",
+        "schema-failure": "✗",
+        "network-failure": "~",
+        skipped: "·"
+    };
+    const lines = report.results.map((r) => `${icon[r.status]} ${r.id.padEnd(28)} ${r.status}` +
+        (r.reason ? ` — ${r.reason}` : ""));
+    lines.push(`\n${report.totals.passed} passed, ${report.totals.schemaFailures} schema failures, ` +
+        `${report.totals.networkFailures} network failures, ${report.totals.skipped} skipped ` +
+        `(${report.totals.requests} requests, ~USD ${report.totals.costUsd.toFixed(4)})`);
+    return lines.join("\n");
+}
+//# sourceMappingURL=probe-runner.js.map

package/dist/providers/contract/probe-runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"probe-runner.js","sourceRoot":"","sources":["../../../src/providers/contract/probe-runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,UAAU,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AACzD,OAAO,EACL,cAAc,EAGf,MAAM,qBAAqB,CAAC;AA8C7B,2EAA2E;AAC3E,MAAM,iBAAkB,SAAQ,KAAK;CAAG;AAExC,KAAK,UAAU,YAAY,CACzB,QAAkB,EAClB,KAAyB;IAEzB,MAAM,gBAAgB,GAAG,KAAK,CAAC,IAAI,EAAE,gBAAgB,IAAI,KAAK,CAAC;IAC/D,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;IACnC,IAAI,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACtC,MAAM,IAAI,iBAAiB,CACzB,QAAQ,QAAQ,CAAC,MAAM,KAAK,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,EAAE,CACpD,CAAC;IACJ,CAAC;IACD,IAAI,CAAC;QACH,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAY,CAAC;IACrC,CAAC;IAAC,MAAM,CAAC;QACP,0EAA0E;QAC1E,MAAM,IAAI,iBAAiB,CACzB,mCAAmC,QAAQ,CAAC,MAAM,MAAM,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,EAAE,CAChF,CAAC;IACJ,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,SAAS,CAC7B,OAAO,GAAqB,EAAE;IAE9B,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,IAAI,EAAE,CAAC;IAC9B,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IACzE,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,IAAI,cAAc,CAAC;IAClD,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,MAAM,CAAC;IAC9C,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;IAE9C,MAAM,OAAO,GAAkB,EAAE,CAAC;IAClC,MAAM,WAAW,GAA0D,EAAE,CAAC;IAE9E,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,MAAM,KAAK,GAAG,CAAC,WAAW,CAAC,KAAK,CAAC,QAAQ,CAAC,KAAK,EAAE,QAAQ,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC,CAAC;QAC5E,MAAM,IAAI,GAAG;YACX,EAAE,EAAE,KAAK,CAAC,EAAE;YACZ,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,OAAO,EAAE,KAAK,CAAC,OAAO;SACvB,CAAC;QACF,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC;QACxB,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,OAAO,CAAC,IAAI,CAAC;gBACX,GAAG,IAAI;gBACP,MAAM,EAAE,SAAS;gBACjB,MAAM,EAAE,iBAAiB,KAAK,CAAC,OAAO,IAAI,wBAAwB,EAAE;gBACpE,UAAU,EAAE,CAAC;gBACb,QAAQ,EAAE,CAAC;gBACX,OAAO,EAAE,CAAC;aACX,CAAC,CAAC;YACH,SAAS;QACX,CAAC;QAED,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QACvE,IAAI,IAAI,CAAC,UAAU,IAAI,CAAC,UAAU,EAAE,CAAC;YACnC,OAAO,CAAC,IAAI,CAAC;gBACX,GAAG,IAAI;gBACP,MAAM,EAAE,SAAS;gBACjB,MAAM,EAAE,GAAG,IAAI,CAAC,UAAU,aAAa;gBACvC,UAAU,EAAE,CAAC;gBACb,QAAQ,EAAE,CAAC;gBACX,OAAO,EAAE,CAAC;aACX,CAAC,CAAC;YACH,SAAS;QACX,CAAC;QACD,IAAI,KAAK,CAAC,QAAQ,GAAG,CAAC,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;YAC1C,OAAO,CAAC,IAAI,CAAC;gBACX,GAAG,IAAI;gBACP,MAAM,EAAE,SAAS;gBACjB,MAAM,EAAE,sBAAsB,KAAK,CAAC,QAAQ,cAAc,IAAI,CAAC,WAAW,GAAG;gBAC7E,UAAU,EAAE,CAAC;gBACb,QAAQ,EAAE,CAAC;gBACX,OAAO,EAAE,CAAC;aACX,CAAC,CAAC;YACH,SAAS;QACX,CAAC;QACD,IAAI,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;YAC5D,OAAO,CAAC,IAAI,CAAC;gBACX,GAAG,IAAI;gBACP,MAAM,EAAE,SAAS;gBACjB,MAAM,EAAE,mBAAmB,KAAK,CAAC,QAAQ,kBAAkB,IAAI,CAAC,UAAU,GAAG;gBAC7E,UAAU,EAAE,CAAC;gBACb,QAAQ,EAAE,CAAC;gBACX,OAAO,EAAE,CAAC;aACX,CAAC,CAAC;YACH,SAAS;QACX,CAAC;QAED,MAAM,SAAS,GAAG,GAAG,EAAE,CAAC;QACxB,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;QAC9D,IAAI,MAAM,GAAgB,QAAQ,CAAC;QACnC,IAAI,MAA0B,CAAC;QAC/B,IAAI,KAAc,CAAC;QACnB,IAAI,CAAC;YACH,MAAM,EAAE,GAAG,EAAE,IAAI,EAAE,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YACpD,IAAI,IAAa,CAAC;YAClB,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,MAAM,SAAS,CAAC,GAAG,EAAE;oBACpC,GAAG,IAAI;oBACP,MAAM,EAAE,UAAU,CAAC,MAAM;iBAC1B,CAAC,CAAC;gBACH,IAAI,GAAG,MAAM,YAAY,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YAC7C,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,MAAM,GAAG,YAAY,iBAAiB;oBACpC,CAAC,CAAC,GAAG;oBACL,CAAC,CAAC,IAAI,iBAAiB,CAAE,GAAa,CAAC,OAAO,CAAC,CAAC;YACpD,CAAC;YACD,KAAK,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;YAC7B,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACpB,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,GAAG,GAAG,YAAY,iBAAiB,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,gBAAgB,CAAC;YACjF,MAAM,GAAG,UAAU,CAAE,GAAa,CAAC,OAAO,CAAC,CAAC;QAC9C,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;QAED,KAAK,CAAC,QAAQ,IAAI,CAAC,CAAC;QACpB,KAAK,CAAC,OAAO,IAAI,IAAI,CAAC,gBAAgB,CAAC;QACvC,OAAO,CAAC,IAAI,CAAC;YACX,GAAG,IAAI;YACP,MAAM;YACN,MAAM;YACN,UAAU,EAAE,GAAG,EAAE,GAAG,SAAS;YAC7B,QAAQ,EAAE,CAAC;YACX,OAAO,EAAE,IAAI,CAAC,gBAAgB;YAC9B,KAAK;SACN,CAAC,CAAC;IACL,CAAC;IAED,MAAM,MAAM,GAAG;QACb,MAAM,EAAE,CAAC;QACT,cAAc,EAAE,CAAC;QACjB,eAAe,EAAE,CAAC;QAClB,OAAO,EAAE,CAAC;QACV,QAAQ,EAAE,CAAC;QACX,OAAO,EAAE,CAAC;KACX,CAAC;IACF,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,IAAI,CAAC,CAAC,MAAM,KAAK,QAAQ;YAAE,MAAM,CAAC,MAAM,EAAE,CAAC;aACtC,IAAI,CAAC,CAAC,MAAM,KAAK,gBAAgB;YAAE,MAAM,CAAC,cAAc,EAAE,CAAC;aAC3D,IAAI,CAAC,CAAC,MAAM,KAAK,iBAAiB;YAAE,MAAM,CAAC,eAAe,EAAE,CAAC;aAC7D,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS;YAAE,MAAM,CAAC,OAAO,EAAE,CAAC;QAClD,MAAM,CAAC,QAAQ,IAAI,CAAC,CAAC,QAAQ,CAAC;QAC9B,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC;IAC9B,CAAC;IAED,OAAO;QACL,SAAS,EAAE,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,WAAW,EAAE;QACxC,OAAO;QACP,MAAM;QACN,WAAW;KACZ,CAAC;AACJ,CAAC;AAED,wCAAwC;AACxC,MAAM,UAAU,iBAAiB,CAAC,MAAmB;IACnD,MAAM,IAAI,GAAgC;QACxC,MAAM,EAAE,GAAG;QACX,gBAAgB,EAAE,GAAG;QACrB,iBAAiB,EAAE,GAAG;QACtB,OAAO,EAAE,GAAG;KACb,CAAC;IACF,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAC9B,CAAC,CAAC,EAAE,EAAE,CACJ,GAAG,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE;QAClD,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CACrC,CAAC;IACF,KAAK,CAAC,IAAI,CACR,KAAK,MAAM,CAAC,MAAM,CAAC,MAAM,YAAY,MAAM,CAAC,MAAM,CAAC,cAAc,oBAAoB;QACnF,GAAG,MAAM,CAAC,MAAM,CAAC,eAAe,sBAAsB,MAAM,CAAC,MAAM,CAAC,OAAO,WAAW;QACtF,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,mBAAmB,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CACnF,CAAC;IACF,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}

package/dist/providers/contract/redact.d.ts

@@ -0,0 +1,29 @@
+/**
+ * Artifact redaction for the provider contract probes.
+ *
+ * A probe run keeps something about the live response so a human can read why
+ * it failed, and a live response carries credentials in echoed headers, the
+ * prompt, model output, request ids, and pre-signed result URLs. Rather than
+ * denylisting those one by one — a list that is wrong the first time a
+ * provider adds a field — the retained artifact is a *shape*: every leaf
+ * becomes its type name, and only a small allowlist of enum-like keys keeps
+ * its literal value. A probe checks structure, so structure is all it needs.
+ */
+/**
+ * Keys whose values stay literal. Each is an enum or a small closed set that
+ * carries no prompt, credential, identifier, or URL — and each is exactly what
+ * a contract failure is usually about.
+ */
+export declare const SHAPE_LITERAL_KEYS: ReadonlySet<string>;
+/**
+ * Reduce a value to its shape. Strings become `"string(len)"`, numbers
+ * `"number"`, and so on; an allowlisted key keeps a short literal instead.
+ */
+export declare function summarizeShape(value: unknown, depth?: number): unknown;
+/**
+ * Redact free text a probe wants to keep — an error message, which can quote
+ * a body. Strips anything that looks like a credential or a URL query string,
+ * then truncates.
+ */
+export declare function redactText(text: string, maxLength?: number): string;
+//# sourceMappingURL=redact.d.ts.map

package/dist/providers/contract/redact.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"redact.d.ts","sourceRoot":"","sources":["../../../src/providers/contract/redact.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH;;;;GAIG;AACH,eAAO,MAAM,kBAAkB,EAAE,WAAW,CAAC,MAAM,CAejD,CAAC;AAaH;;;GAGG;AACH,wBAAgB,cAAc,CAAC,KAAK,EAAE,OAAO,EAAE,KAAK,SAAI,GAAG,OAAO,CAsBjE;AAcD;;;;GAIG;AACH,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,SAAM,GAAG,MAAM,CA0BhE"}

package/dist/providers/contract/redact.js

@@ -0,0 +1,113 @@
+/**
+ * Artifact redaction for the provider contract probes.
+ *
+ * A probe run keeps something about the live response so a human can read why
+ * it failed, and a live response carries credentials in echoed headers, the
+ * prompt, model output, request ids, and pre-signed result URLs. Rather than
+ * denylisting those one by one — a list that is wrong the first time a
+ * provider adds a field — the retained artifact is a *shape*: every leaf
+ * becomes its type name, and only a small allowlist of enum-like keys keeps
+ * its literal value. A probe checks structure, so structure is all it needs.
+ */
+/**
+ * Keys whose values stay literal. Each is an enum or a small closed set that
+ * carries no prompt, credential, identifier, or URL — and each is exactly what
+ * a contract failure is usually about.
+ */
+export const SHAPE_LITERAL_KEYS = new Set([
+    "object",
+    "role",
+    "type",
+    "state",
+    "status",
+    "code",
+    "finish_reason",
+    "finishReason",
+    "blockReason",
+    "mime_type",
+    "mimeType",
+    "content_type",
+    "modality",
+    "successFlag"
+]);
+/** Longest array the artifact keeps; the rest is reported as a count. */
+const MAX_ARRAY_ITEMS = 3;
+/** Deepest nesting the artifact keeps. */
+const MAX_DEPTH = 8;
+function literalKind(value) {
+    if (value === null)
+        return "null";
+    if (Array.isArray(value))
+        return "array";
+    return typeof value;
+}
+/**
+ * Reduce a value to its shape. Strings become `"string(len)"`, numbers
+ * `"number"`, and so on; an allowlisted key keeps a short literal instead.
+ */
+export function summarizeShape(value, depth = 0) {
+    if (depth > MAX_DEPTH)
+        return "…";
+    if (value === null)
+        return "null";
+    if (Array.isArray(value)) {
+        const head = value
+            .slice(0, MAX_ARRAY_ITEMS)
+            .map((item) => summarizeShape(item, depth + 1));
+        return value.length > MAX_ARRAY_ITEMS
+            ? [...head, `…${value.length - MAX_ARRAY_ITEMS} more`]
+            : head;
+    }
+    if (typeof value === "object") {
+        const out = {};
+        for (const [key, item] of Object.entries(value)) {
+            out[key] = SHAPE_LITERAL_KEYS.has(key)
+                ? shortLiteral(item)
+                : summarizeShape(item, depth + 1);
+        }
+        return out;
+    }
+    if (typeof value === "string")
+        return `string(${value.length})`;
+    return literalKind(value);
+}
+/**
+ * An allowlisted key's value, kept literal but bounded: a long string under an
+ * allowlisted name is a value someone reused that key for, not an enum.
+ */
+function shortLiteral(value) {
+    if (typeof value === "string") {
+        return value.length <= 32 ? value : `string(${value.length})`;
+    }
+    if (typeof value === "number" || typeof value === "boolean")
+        return value;
+    return summarizeShape(value);
+}
+/**
+ * Redact free text a probe wants to keep — an error message, which can quote
+ * a body. Strips anything that looks like a credential or a URL query string,
+ * then truncates.
+ */
+export function redactText(text, maxLength = 400) {
+    const scrubbed = text
+        // Bearer/Key headers and `"api_key": "…"`-style pairs.
+        .replace(/\b(Bearer|Key)\s+[A-Za-z0-9._-]{8,}/gi, "$1 [redacted]")
+        .replace(/("?(?:api[_-]?key|apikey|token|secret|password|authorization|credential)"?\s*[:=]\s*"?)[^",\s}]+/gi, "$1[redacted]")
+        // Credential-shaped tokens wherever they sit, including inside a message
+        // a provider echoed back ("Incorrect API key provided: sk-…").
+        .replace(/\b(?:sk|pk|rk|ak|gsk|xai|fal|kie|hf|nvapi|AIza)[-_][A-Za-z0-9_-]{8,}/gi, "[redacted-credential]")
+        .replace(/\b[A-Za-z0-9_-]{32,}\b/g, "[redacted-token]")
+        // Any URL: the path may identify a user and the query may sign it.
+        .replace(/https?:\/\/[^\s"']+/gi, (url) => {
+        try {
+            return `${new URL(url).origin}/[redacted]`;
+        }
+        catch {
+            return "[redacted-url]";
+        }
+    });
+    return scrubbed.length > maxLength
+        ? `${scrubbed.slice(0, maxLength)}…`
+        : scrubbed;
+}
+//# sourceMappingURL=redact.js.map

package/dist/providers/contract/redact.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"redact.js","sourceRoot":"","sources":["../../../src/providers/contract/redact.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH;;;;GAIG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAwB,IAAI,GAAG,CAAC;IAC7D,QAAQ;IACR,MAAM;IACN,MAAM;IACN,OAAO;IACP,QAAQ;IACR,MAAM;IACN,eAAe;IACf,cAAc;IACd,aAAa;IACb,WAAW;IACX,UAAU;IACV,cAAc;IACd,UAAU;IACV,aAAa;CACd,CAAC,CAAC;AAEH,yEAAyE;AACzE,MAAM,eAAe,GAAG,CAAC,CAAC;AAC1B,0CAA0C;AAC1C,MAAM,SAAS,GAAG,CAAC,CAAC;AAEpB,SAAS,WAAW,CAAC,KAAc;IACjC,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,MAAM,CAAC;IAClC,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC;IACzC,OAAO,OAAO,KAAK,CAAC;AACtB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,cAAc,CAAC,KAAc,EAAE,KAAK,GAAG,CAAC;IACtD,IAAI,KAAK,GAAG,SAAS;QAAE,OAAO,GAAG,CAAC;IAClC,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,MAAM,CAAC;IAClC,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzB,MAAM,IAAI,GAAG,KAAK;aACf,KAAK,CAAC,CAAC,EAAE,eAAe,CAAC;aACzB,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;QAClD,OAAO,KAAK,CAAC,MAAM,GAAG,eAAe;YACnC,CAAC,CAAC,CAAC,GAAG,IAAI,EAAE,IAAI,KAAK,CAAC,MAAM,GAAG,eAAe,OAAO,CAAC;YACtD,CAAC,CAAC,IAAI,CAAC;IACX,CAAC;IACD,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC9B,MAAM,GAAG,GAA4B,EAAE,CAAC;QACxC,KAAK,MAAM,CAAC,GAAG,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,KAAgC,CAAC,EAAE,CAAC;YAC3E,GAAG,CAAC,GAAG,CAAC,GAAG,kBAAkB,CAAC,GAAG,CAAC,GAAG,CAAC;gBACpC,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC;gBACpB,CAAC,CAAC,cAAc,CAAC,IAAI,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;QACtC,CAAC;QACD,OAAO,GAAG,CAAC;IACb,CAAC;IACD,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,UAAU,KAAK,CAAC,MAAM,GAAG,CAAC;IAChE,OAAO,WAAW,CAAC,KAAK,CAAC,CAAC;AAC5B,CAAC;AAED;;;GAGG;AACH,SAAS,YAAY,CAAC,KAAc;IAClC,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC9B,OAAO,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,KAAK,CAAC,MAAM,GAAG,CAAC;IAChE,CAAC;IACD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,OAAO,KAAK,KAAK,SAAS;QAAE,OAAO,KAAK,CAAC;IAC1E,OAAO,cAAc,CAAC,KAAK,CAAC,CAAC;AAC/B,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,UAAU,CAAC,IAAY,EAAE,SAAS,GAAG,GAAG;IACtD,MAAM,QAAQ,GAAG,IAAI;QACnB,uDAAuD;SACtD,OAAO,CAAC,uCAAuC,EAAE,eAAe,CAAC;SACjE,OAAO,CACN,oGAAoG,EACpG,cAAc,CACf;QACD,yEAAyE;QACzE,+DAA+D;SAC9D,OAAO,CACN,wEAAwE,EACxE,uBAAuB,CACxB;SACA,OAAO,CAAC,yBAAyB,EAAE,kBAAkB,CAAC;QACvD,mEAAmE;SAClE,OAAO,CAAC,uBAAuB,EAAE,CAAC,GAAG,EAAE,EAAE;QACxC,IAAI,CAAC;YACH,OAAO,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC,MAAM,aAAa,CAAC;QAC7C,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,gBAAgB,CAAC;QAC1B,CAAC;IACH,CAAC,CAAC,CAAC;IACL,OAAO,QAAQ,CAAC,MAAM,GAAG,SAAS;QAChC,CAAC,CAAC,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,GAAG;QACpC,CAAC,CAAC,QAAQ,CAAC;AACf,CAAC"}

package/dist/providers/core-tools.d.ts

@@ -0,0 +1,74 @@
+/**
+ * The NodeTool tools that mirror a Claude Agent SDK built-in.
+ *
+ * Two rules follow from that mirroring, and this module is the one place both
+ * read:
+ *
+ * 1. **Top level, not CodeAct.** Every provider gets these as ordinary tool
+ *    definitions instead of finding them behind `execute_code`. They are the
+ *    shapes every frontier model is trained on, so a tool call costs less than
+ *    a sandbox round trip that only forwards one.
+ * 2. **The built-in wins on the Claude Agent SDK.** That provider runs the
+ *    SDK's own agent loop, which already ships `Read`/`Write`/`Grep`/… . A
+ *    NodeTool copy next to the built-in is a second surface for one capability,
+ *    so {@link SDK_NATIVE_TOOL_REPLACEMENTS} names the ones the built-in
+ *    replaces outright and the provider drops them from its MCP toolset.
+ */
+/**
+ * Tools lifted out of the CodeAct sandbox and offered top level, for every
+ * provider. Membership is decided by "the Claude Agent SDK has a built-in for
+ * this", not by how often the tool is used.
+ */
+export declare const CORE_TOOL_NAMES: ReadonlySet<string>;
+/**
+ * Discovery: which providers, models and node types this install actually has.
+ *
+ * These are lookups whose answer decides what the model writes next, and a
+ * wrong answer is expensive — a hallucinated model id or node type fails at
+ * generation time, after the run has been paid for. A lookup is also exactly
+ * the shape a tool call fits: one question, one answer, nothing to compose.
+ * Routing it through a code action costs a sandbox round trip and hides the
+ * result behind whatever the action chose to return.
+ *
+ * They stay on the belt, so `nodetool.models.pick()` inside an action and the
+ * graph planner's own searches keep working; what changes is the documented
+ * path.
+ */
+export declare const DISCOVERY_TOOL_NAMES: ReadonlySet<string>;
+/**
+ * Every tool offered as an ordinary tool call next to `execute_code`:
+ * {@link CORE_TOOL_NAMES} plus {@link DISCOVERY_TOOL_NAMES}. The two sets are
+ * kept apart because only the core set has SDK built-ins behind it — the
+ * substitution in {@link SDK_NATIVE_TOOL_REPLACEMENTS} must never reach a
+ * discovery tool, which no built-in covers.
+ */
+export declare const DIRECT_TOOL_NAMES: ReadonlySet<string>;
+/**
+ * NodeTool tool name → the Claude Agent SDK built-in that replaces it. Only
+ * true replacements are listed. Three members of {@link CORE_TOOL_NAMES} are
+ * deliberately absent:
+ *
+ * - `list_directory` — `Glob` matches a pattern; it does not list a directory.
+ * - `browser` / `http_request` / `download_file` — `WebFetch` is a prompted GET
+ *   that writes nothing, so it covers none of the three in full.
+ * - `run_subtask` — `Task` spawns an SDK agent holding SDK tools. The child
+ *   would lose the NodeTool toolbelt the parent passed it.
+ */
+export declare const SDK_NATIVE_TOOL_REPLACEMENTS: ReadonlyMap<string, string>;
+/**
+ * The replacements that resolve paths. NodeTool's versions are contained to the
+ * run's workspace directory; the SDK's built-ins resolve against the session
+ * `cwd`. Substituting one for the other is only equivalent when that `cwd` is
+ * the workspace, so a caller that passes no workspace directory keeps the
+ * NodeTool tools.
+ */
+export declare const SDK_NATIVE_WORKSPACE_SCOPED: ReadonlySet<string>;
+/**
+ * The SDK built-ins that replace a NodeTool tool for this call.
+ *
+ * @param toolNames Tool names the caller offered.
+ * @param workspaceDir The session `cwd`. Without it the path-scoped
+ *   replacements are skipped — see {@link SDK_NATIVE_WORKSPACE_SCOPED}.
+ */
+export declare function sdkNativeReplacements(toolNames: Iterable<string>, workspaceDir?: string): Set<string>;
+//# sourceMappingURL=core-tools.d.ts.map

package/dist/providers/core-tools.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"core-tools.d.ts","sourceRoot":"","sources":["../../src/providers/core-tools.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH;;;;GAIG;AACH,eAAO,MAAM,eAAe,EAAE,WAAW,CAAC,MAAM,CAiB9C,CAAC;AAEH;;;;;;;;;;;;;GAaG;AACH,eAAO,MAAM,oBAAoB,EAAE,WAAW,CAAC,MAAM,CASnD,CAAC;AAEH;;;;;;GAMG;AACH,eAAO,MAAM,iBAAiB,EAAE,WAAW,CAAC,MAAM,CAGhD,CAAC;AAEH;;;;;;;;;;GAUG;AACH,eAAO,MAAM,4BAA4B,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAUpE,CAAC;AAEF;;;;;;GAMG;AACH,eAAO,MAAM,2BAA2B,EAAE,WAAW,CAAC,MAAM,CAM1D,CAAC;AAEH;;;;;;GAMG;AACH,wBAAgB,qBAAqB,CACnC,SAAS,EAAE,QAAQ,CAAC,MAAM,CAAC,EAC3B,YAAY,CAAC,EAAE,MAAM,GACpB,GAAG,CAAC,MAAM,CAAC,CAQb"}

package/dist/providers/core-tools.js

@@ -0,0 +1,127 @@
+/**
+ * The NodeTool tools that mirror a Claude Agent SDK built-in.
+ *
+ * Two rules follow from that mirroring, and this module is the one place both
+ * read:
+ *
+ * 1. **Top level, not CodeAct.** Every provider gets these as ordinary tool
+ *    definitions instead of finding them behind `execute_code`. They are the
+ *    shapes every frontier model is trained on, so a tool call costs less than
+ *    a sandbox round trip that only forwards one.
+ * 2. **The built-in wins on the Claude Agent SDK.** That provider runs the
+ *    SDK's own agent loop, which already ships `Read`/`Write`/`Grep`/… . A
+ *    NodeTool copy next to the built-in is a second surface for one capability,
+ *    so {@link SDK_NATIVE_TOOL_REPLACEMENTS} names the ones the built-in
+ *    replaces outright and the provider drops them from its MCP toolset.
+ */
+/**
+ * Tools lifted out of the CodeAct sandbox and offered top level, for every
+ * provider. Membership is decided by "the Claude Agent SDK has a built-in for
+ * this", not by how often the tool is used.
+ */
+export const CORE_TOOL_NAMES = new Set([
+    // Read / Write / Edit / Glob / Grep.
+    "read_file",
+    "write_file",
+    "edit_file",
+    "list_directory",
+    "glob",
+    "grep",
+    // WebSearch / WebFetch.
+    "web_search",
+    "browser",
+    "http_request",
+    "download_file",
+    // TodoWrite.
+    "todo_write",
+    // Task.
+    "run_subtask"
+]);
+/**
+ * Discovery: which providers, models and node types this install actually has.
+ *
+ * These are lookups whose answer decides what the model writes next, and a
+ * wrong answer is expensive — a hallucinated model id or node type fails at
+ * generation time, after the run has been paid for. A lookup is also exactly
+ * the shape a tool call fits: one question, one answer, nothing to compose.
+ * Routing it through a code action costs a sandbox round trip and hides the
+ * result behind whatever the action chose to return.
+ *
+ * They stay on the belt, so `nodetool.models.pick()` inside an action and the
+ * graph planner's own searches keep working; what changes is the documented
+ * path.
+ */
+export const DISCOVERY_TOOL_NAMES = new Set([
+    // Providers and models.
+    "find_model",
+    "list_models",
+    "list_provider_models",
+    // Node types.
+    "search_nodes",
+    "get_node_info",
+    "list_nodes"
+]);
+/**
+ * Every tool offered as an ordinary tool call next to `execute_code`:
+ * {@link CORE_TOOL_NAMES} plus {@link DISCOVERY_TOOL_NAMES}. The two sets are
+ * kept apart because only the core set has SDK built-ins behind it — the
+ * substitution in {@link SDK_NATIVE_TOOL_REPLACEMENTS} must never reach a
+ * discovery tool, which no built-in covers.
+ */
+export const DIRECT_TOOL_NAMES = new Set([
+    ...CORE_TOOL_NAMES,
+    ...DISCOVERY_TOOL_NAMES
+]);
+/**
+ * NodeTool tool name → the Claude Agent SDK built-in that replaces it. Only
+ * true replacements are listed. Three members of {@link CORE_TOOL_NAMES} are
+ * deliberately absent:
+ *
+ * - `list_directory` — `Glob` matches a pattern; it does not list a directory.
+ * - `browser` / `http_request` / `download_file` — `WebFetch` is a prompted GET
+ *   that writes nothing, so it covers none of the three in full.
+ * - `run_subtask` — `Task` spawns an SDK agent holding SDK tools. The child
+ *   would lose the NodeTool toolbelt the parent passed it.
+ */
+export const SDK_NATIVE_TOOL_REPLACEMENTS = new Map([
+    ["read_file", "Read"],
+    ["write_file", "Write"],
+    ["edit_file", "Edit"],
+    ["glob", "Glob"],
+    ["grep", "Grep"],
+    ["web_search", "WebSearch"],
+    ["todo_write", "TodoWrite"]
+]);
+/**
+ * The replacements that resolve paths. NodeTool's versions are contained to the
+ * run's workspace directory; the SDK's built-ins resolve against the session
+ * `cwd`. Substituting one for the other is only equivalent when that `cwd` is
+ * the workspace, so a caller that passes no workspace directory keeps the
+ * NodeTool tools.
+ */
+export const SDK_NATIVE_WORKSPACE_SCOPED = new Set([
+    "read_file",
+    "write_file",
+    "edit_file",
+    "glob",
+    "grep"
+]);
+/**
+ * The SDK built-ins that replace a NodeTool tool for this call.
+ *
+ * @param toolNames Tool names the caller offered.
+ * @param workspaceDir The session `cwd`. Without it the path-scoped
+ *   replacements are skipped — see {@link SDK_NATIVE_WORKSPACE_SCOPED}.
+ */
+export function sdkNativeReplacements(toolNames, workspaceDir) {
+    const replaced = new Set();
+    for (const name of toolNames) {
+        if (!SDK_NATIVE_TOOL_REPLACEMENTS.has(name))
+            continue;
+        if (!workspaceDir && SDK_NATIVE_WORKSPACE_SCOPED.has(name))
+            continue;
+        replaced.add(name);
+    }
+    return replaced;
+}
+//# sourceMappingURL=core-tools.js.map

package/dist/providers/core-tools.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"core-tools.js","sourceRoot":"","sources":["../../src/providers/core-tools.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH;;;;GAIG;AACH,MAAM,CAAC,MAAM,eAAe,GAAwB,IAAI,GAAG,CAAC;IAC1D,qCAAqC;IACrC,WAAW;IACX,YAAY;IACZ,WAAW;IACX,gBAAgB;IAChB,MAAM;IACN,MAAM;IACN,wBAAwB;IACxB,YAAY;IACZ,SAAS;IACT,cAAc;IACd,eAAe;IACf,aAAa;IACb,YAAY;IACZ,QAAQ;IACR,aAAa;CACd,CAAC,CAAC;AAEH;;;;;;;;;;;;;GAaG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAwB,IAAI,GAAG,CAAC;IAC/D,wBAAwB;IACxB,YAAY;IACZ,aAAa;IACb,sBAAsB;IACtB,cAAc;IACd,cAAc;IACd,eAAe;IACf,YAAY;CACb,CAAC,CAAC;AAEH;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAwB,IAAI,GAAG,CAAC;IAC5D,GAAG,eAAe;IAClB,GAAG,oBAAoB;CACxB,CAAC,CAAC;AAEH;;;;;;;;;;GAUG;AACH,MAAM,CAAC,MAAM,4BAA4B,GAAgC,IAAI,GAAG,CAC9E;IACE,CAAC,WAAW,EAAE,MAAM,CAAC;IACrB,CAAC,YAAY,EAAE,OAAO,CAAC;IACvB,CAAC,WAAW,EAAE,MAAM,CAAC;IACrB,CAAC,MAAM,EAAE,MAAM,CAAC;IAChB,CAAC,MAAM,EAAE,MAAM,CAAC;IAChB,CAAC,YAAY,EAAE,WAAW,CAAC;IAC3B,CAAC,YAAY,EAAE,WAAW,CAAC;CAC5B,CACF,CAAC;AAEF;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,2BAA2B,GAAwB,IAAI,GAAG,CAAC;IACtE,WAAW;IACX,YAAY;IACZ,WAAW;IACX,MAAM;IACN,MAAM;CACP,CAAC,CAAC;AAEH;;;;;;GAMG;AACH,MAAM,UAAU,qBAAqB,CACnC,SAA2B,EAC3B,YAAqB;IAErB,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAU,CAAC;IACnC,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,IAAI,CAAC,4BAA4B,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,SAAS;QACtD,IAAI,CAAC,YAAY,IAAI,2BAA2B,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,SAAS;QACrE,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC"}