@jsonstudio/rcc 0.89.1086 → 0.89.1136

package/dist/tools/provider-update/probe-context.js

@@ -0,0 +1,199 @@
+import { encoding_for_model } from 'tiktoken';
+const encoderCache = new Map();
+function getEncoder(model) {
+    const key = model.trim() || 'gpt-4o';
+    const existing = encoderCache.get(key);
+    if (existing) {
+        return existing;
+    }
+    // tiktoken's TS types restrict this to a known model union; allow override for local experiments.
+    const created = encoding_for_model(key);
+    encoderCache.set(key, created);
+    return created;
+}
+const UNIT_TEXT = 'a ';
+export function buildTextForExactTokenCount(targetTokens, encoderModel = 'gpt-4o') {
+    if (!Number.isFinite(targetTokens) || targetTokens <= 0) {
+        throw new Error(`Invalid targetTokens: ${targetTokens}`);
+    }
+    // We deliberately generate text from plain strings (not decode(tokens)) because
+    // decode() is not guaranteed to round-trip back to the same token count via encode().
+    //
+    // Empirically for tiktoken (e.g. 'gpt-4o'), UNIT_TEXT repeat has stable linear growth:
+    //  encode('a '.repeat(n)).length == n + k
+    // We detect k at runtime and compute n to hit targetTokens exactly.
+    const encoder = getEncoder(encoderModel);
+    if (Math.floor(targetTokens) === 1) {
+        return 'a';
+    }
+    const f1 = encoder.encode(UNIT_TEXT).length;
+    const f99 = encoder.encode(UNIT_TEXT.repeat(99)).length;
+    const f100 = encoder.encode(UNIT_TEXT.repeat(100)).length;
+    const slope = f100 - f99;
+    const offset = slope === 1 ? (f1 - 1) : 0;
+    let repeats = slope === 1 ? Math.max(1, Math.floor(targetTokens) - offset) : Math.max(1, Math.floor(targetTokens));
+    // Fine-tune with minimal extra encodes (should converge immediately for slope=1).
+    for (let i = 0; i < 20; i++) {
+        const text = UNIT_TEXT.repeat(repeats);
+        const tokens = encoder.encode(text).length;
+        if (tokens === Math.floor(targetTokens)) {
+            return text;
+        }
+        if (tokens < Math.floor(targetTokens)) {
+            repeats += Math.max(1, Math.floor(targetTokens) - tokens);
+        }
+        else {
+            repeats -= Math.max(1, tokens - Math.floor(targetTokens));
+            if (repeats <= 0) {
+                repeats = 1;
+            }
+        }
+    }
+    // Last resort: return best-effort (caller will validate).
+    return UNIT_TEXT.repeat(repeats);
+}
+export function countTokens(text, encoderModel = 'gpt-4o') {
+    const encoder = getEncoder(encoderModel);
+    return encoder.encode(text).length;
+}
+function sleep(ms) {
+    return new Promise((resolve) => setTimeout(resolve, ms));
+}
+async function requestOnce(modelId, threshold, text, opts) {
+    const fetcher = opts.fetcher ?? fetch;
+    const controller = new AbortController();
+    const timeout = setTimeout(() => controller.abort(), opts.timeoutMs);
+    try {
+        const headers = {
+            'Content-Type': 'application/json',
+            Accept: 'application/json'
+        };
+        const key = typeof opts.apiKey === 'string' ? opts.apiKey.trim() : '';
+        if (key) {
+            headers.Authorization = `Bearer ${key}`;
+        }
+        const body = {
+            model: modelId,
+            input: [
+                {
+                    role: 'user',
+                    content: [
+                        {
+                            type: 'input_text',
+                            text
+                        }
+                    ]
+                }
+            ],
+            stream: false,
+            max_output_tokens: 1
+        };
+        return await fetcher(opts.endpoint, {
+            method: 'POST',
+            headers,
+            body: JSON.stringify(body),
+            signal: controller.signal
+        });
+    }
+    finally {
+        clearTimeout(timeout);
+    }
+}
+async function requestWithRetries(modelId, threshold, text, opts) {
+    let lastError = null;
+    for (let attempt = 0; attempt <= opts.maxRetries; attempt++) {
+        try {
+            const resp = await requestOnce(modelId, threshold, text, opts);
+            if (resp.status !== 429 && resp.status !== 503 && resp.status !== 504) {
+                return resp;
+            }
+            lastError = new Error(`HTTP ${resp.status}`);
+            // Drain body to free sockets.
+            try {
+                await resp.text();
+            }
+            catch { /* ignore */ }
+        }
+        catch (err) {
+            lastError = err;
+        }
+        if (attempt < opts.maxRetries) {
+            await sleep(500 * (attempt + 1));
+        }
+    }
+    throw lastError instanceof Error ? lastError : new Error(String(lastError ?? 'request failed'));
+}
+export async function probeContextForModel(modelId, thresholds, { endpoint, apiKey, timeoutMs = 60_000, fetcher, maxRetries = 2, encoderModel = 'gpt-4o' }) {
+    const model = (modelId || '').trim();
+    if (!model) {
+        throw new Error('probeContextForModel: modelId is required');
+    }
+    const normalizedEndpoint = (endpoint || '').trim();
+    if (!normalizedEndpoint) {
+        throw new Error('probeContextForModel: endpoint is required');
+    }
+    const sorted = thresholds
+        .map((value) => Math.floor(Number(value)))
+        .filter((value) => Number.isFinite(value) && value > 0);
+    if (!sorted.length) {
+        throw new Error('probeContextForModel: thresholds must be non-empty');
+    }
+    const passed = [];
+    let firstFailure;
+    for (const threshold of sorted) {
+        const text = buildTextForExactTokenCount(threshold, encoderModel);
+        const check = countTokens(text, encoderModel);
+        if (check !== threshold) {
+            throw new Error(`probeContextForModel: internal token mismatch: want=${threshold} got=${check}`);
+        }
+        let resp;
+        try {
+            resp = await requestWithRetries(model, threshold, text, {
+                endpoint: normalizedEndpoint,
+                apiKey,
+                timeoutMs,
+                fetcher,
+                maxRetries
+            });
+        }
+        catch (err) {
+            firstFailure = {
+                threshold,
+                status: 0,
+                message: err instanceof Error ? err.message : String(err)
+            };
+            break;
+        }
+        if (resp.ok) {
+            passed.push(threshold);
+            // Drain body to reuse socket; but keep it small.
+            try {
+                await resp.text();
+            }
+            catch { /* ignore */ }
+            continue;
+        }
+        let responseText = '';
+        try {
+            responseText = await resp.text();
+        }
+        catch {
+            responseText = '';
+        }
+        firstFailure = {
+            threshold,
+            status: resp.status,
+            statusText: resp.statusText,
+            responseSnippet: responseText ? responseText.slice(0, 2000) : undefined
+        };
+        break;
+    }
+    return {
+        modelId: model,
+        thresholds: sorted,
+        passed,
+        maxPassedTokens: passed.length ? passed[passed.length - 1] : null,
+        ...(firstFailure ? { firstFailure } : {})
+    };
+}
+//# sourceMappingURL=probe-context.js.map

package/dist/tools/provider-update/probe-context.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"probe-context.js","sourceRoot":"","sources":["../../../src/tools/provider-update/probe-context.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,kBAAkB,EAAE,MAAM,UAAU,CAAC;AA0B9C,MAAM,YAAY,GAAG,IAAI,GAAG,EAAiD,CAAC;AAC9E,SAAS,UAAU,CAAC,KAAa;IAC/B,MAAM,GAAG,GAAG,KAAK,CAAC,IAAI,EAAE,IAAI,QAAQ,CAAC;IACrC,MAAM,QAAQ,GAAG,YAAY,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IACvC,IAAI,QAAQ,EAAE,CAAC;QACb,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,kGAAkG;IAClG,MAAM,OAAO,GAAG,kBAAkB,CAAC,GAAU,CAAC,CAAC;IAC/C,YAAY,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;IAC/B,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,MAAM,SAAS,GAAG,IAAI,CAAC;AAEvB,MAAM,UAAU,2BAA2B,CAAC,YAAoB,EAAE,YAAY,GAAG,QAAQ;IACvF,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,YAAY,CAAC,IAAI,YAAY,IAAI,CAAC,EAAE,CAAC;QACxD,MAAM,IAAI,KAAK,CAAC,yBAAyB,YAAY,EAAE,CAAC,CAAC;IAC3D,CAAC;IACD,gFAAgF;IAChF,sFAAsF;IACtF,EAAE;IACF,uFAAuF;IACvF,0CAA0C;IAC1C,oEAAoE;IACpE,MAAM,OAAO,GAAG,UAAU,CAAC,YAAY,CAAC,CAAC;IAEzC,IAAI,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,KAAK,CAAC,EAAE,CAAC;QACnC,OAAO,GAAG,CAAC;IACb,CAAC;IAED,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC;IAC5C,MAAM,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC;IACxD,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;IAC1D,MAAM,KAAK,GAAG,IAAI,GAAG,GAAG,CAAC;IACzB,MAAM,MAAM,GAAG,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1C,IAAI,OAAO,GAAG,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC;IAEnH,kFAAkF;IAClF,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5B,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACvC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;QAC3C,IAAI,MAAM,KAAK,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,EAAE,CAAC;YACxC,OAAO,IAAI,CAAC;QACd,CAAC;QACD,IAAI,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,EAAE,CAAC;YACtC,OAAO,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,GAAG,MAAM,CAAC,CAAC;QAC5D,CAAC;aAAM,CAAC;YACN,OAAO,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC;YAC1D,IAAI,OAAO,IAAI,CAAC,EAAE,CAAC;gBACjB,OAAO,GAAG,CAAC,CAAC;YACd,CAAC;QACH,CAAC;IACH,CAAC;IAED,0DAA0D;IAC1D,OAAO,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;AACnC,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,IAAY,EAAE,YAAY,GAAG,QAAQ;IAC/D,MAAM,OAAO,GAAG,UAAU,CAAC,YAAY,CAAC,CAAC;IACzC,OAAO,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;AACrC,CAAC;AAED,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;AAC3D,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,OAAe,EAAE,SAAiB,EAAE,IAAY,EAAE,IAAyB;IACpG,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC;IACtC,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;IACrE,IAAI,CAAC;QACH,MAAM,OAAO,GAA2B;YACtC,cAAc,EAAE,kBAAkB;YAClC,MAAM,EAAE,kBAAkB;SAC3B,CAAC;QACF,MAAM,GAAG,GAAG,OAAO,IAAI,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACtE,IAAI,GAAG,EAAE,CAAC;YACR,OAAO,CAAC,aAAa,GAAG,UAAU,GAAG,EAAE,CAAC;QAC1C,CAAC;QAED,MAAM,IAAI,GAAG;YACX,KAAK,EAAE,OAAO;YACd,KAAK,EAAE;gBACL;oBACE,IAAI,EAAE,MAAM;oBACZ,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,YAAY;4BAClB,IAAI;yBACL;qBACF;iBACF;aACF;YACD,MAAM,EAAE,KAAK;YACb,iBAAiB,EAAE,CAAC;SACrB,CAAC;QAEF,OAAO,MAAM,OAAO,CAAC,IAAI,CAAC,QAAQ,EAAE;YAClC,MAAM,EAAE,MAAM;YACd,OAAO;YACP,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;YAC1B,MAAM,EAAE,UAAU,CAAC,MAAM;SAC1B,CAAC,CAAC;IACL,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,OAAO,CAAC,CAAC;IACxB,CAAC;AACH,CAAC;AAED,KAAK,UAAU,kBAAkB,CAAC,OAAe,EAAE,SAAiB,EAAE,IAAY,EAAE,IAAyB;IAC3G,IAAI,SAAS,GAAY,IAAI,CAAC;IAC9B,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,IAAI,CAAC,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QAC5D,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;YAC/D,IAAI,IAAI,CAAC,MAAM,KAAK,GAAG,IAAI,IAAI,CAAC,MAAM,KAAK,GAAG,IAAI,IAAI,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBACtE,OAAO,IAAI,CAAC;YACd,CAAC;YACD,SAAS,GAAG,IAAI,KAAK,CAAC,QAAQ,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;YAC7C,8BAA8B;YAC9B,IAAI,CAAC;gBAAC,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;YAAC,CAAC;YAAC,MAAM,CAAC,CAAC,YAAY,CAAC,CAAC;QACnD,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,SAAS,GAAG,GAAG,CAAC;QAClB,CAAC;QACD,IAAI,OAAO,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;YAC9B,MAAM,KAAK,CAAC,GAAG,GAAG,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC;QACnC,CAAC;IACH,CAAC;IACD,MAAM,SAAS,YAAY,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,SAAS,IAAI,gBAAgB,CAAC,CAAC,CAAC;AAClG,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACxC,OAAe,EACf,UAAoB,EACpB,EACE,QAAQ,EACR,MAAM,EACN,SAAS,GAAG,MAAM,EAClB,OAAO,EACP,UAAU,GAAG,CAAC,EACd,YAAY,GAAG,QAAQ,EAQxB;IAED,MAAM,KAAK,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACrC,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,MAAM,IAAI,KAAK,CAAC,2CAA2C,CAAC,CAAC;IAC/D,CAAC;IACD,MAAM,kBAAkB,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACnD,IAAI,CAAC,kBAAkB,EAAE,CAAC;QACxB,MAAM,IAAI,KAAK,CAAC,4CAA4C,CAAC,CAAC;IAChE,CAAC;IAED,MAAM,MAAM,GAAG,UAAU;SACtB,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;SACzC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC;IAC1D,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;QACnB,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;IACxE,CAAC;IAED,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,YAA6C,CAAC;IAElD,KAAK,MAAM,SAAS,IAAI,MAAM,EAAE,CAAC;QAC/B,MAAM,IAAI,GAAG,2BAA2B,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;QAClE,MAAM,KAAK,GAAG,WAAW,CAAC,IAAI,EAAE,YAAY,CAAC,CAAC;QAC9C,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;YACxB,MAAM,IAAI,KAAK,CAAC,uDAAuD,SAAS,QAAQ,KAAK,EAAE,CAAC,CAAC;QACnG,CAAC;QAED,IAAI,IAAc,CAAC;QACnB,IAAI,CAAC;YACH,IAAI,GAAG,MAAM,kBAAkB,CAAC,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE;gBACtD,QAAQ,EAAE,kBAAkB;gBAC5B,MAAM;gBACN,SAAS;gBACT,OAAO;gBACP,UAAU;aACX,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,YAAY,GAAG;gBACb,SAAS;gBACT,MAAM,EAAE,CAAC;gBACT,OAAO,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;aAC1D,CAAC;YACF,MAAM;QACR,CAAC;QAED,IAAI,IAAI,CAAC,EAAE,EAAE,CAAC;YACZ,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YACvB,iDAAiD;YACjD,IAAI,CAAC;gBAAC,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;YAAC,CAAC;YAAC,MAAM,CAAC,CAAC,YAAY,CAAC,CAAC;YACjD,SAAS;QACX,CAAC;QAED,IAAI,YAAY,GAAG,EAAE,CAAC;QACtB,IAAI,CAAC;YACH,YAAY,GAAG,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;QACnC,CAAC;QAAC,MAAM,CAAC;YACP,YAAY,GAAG,EAAE,CAAC;QACpB,CAAC;QACD,YAAY,GAAG;YACb,SAAS;YACT,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,eAAe,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS;SACxE,CAAC;QACF,MAAM;IACR,CAAC;IAED,OAAO;QACL,OAAO,EAAE,KAAK;QACd,UAAU,EAAE,MAAM;QAClB,MAAM;QACN,eAAe,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC,CAAC,CAAC,IAAI;QAClE,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC1C,CAAC;AACJ,CAAC"}

package/dist/tools/provider-update/types.d.ts

@@ -3,6 +3,7 @@ export type AuthConfig = {
     apiKey?: string;
     headerName?: string;
     prefix?: string;
+    tokenFile?: string;
     clientId?: string;
     clientSecret?: string;
     tokenUrl?: string;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jsonstudio/rcc",
-  "version": "0.89.1086",
+  "version": "0.89.1136",
   "description": "Multi-provider OpenAI proxy server with anthropic/responses/chat support (release)",
   "type": "module",
   "main": "dist/index.js",
@@ -28,7 +28,7 @@
   },
   "scripts": {
     "build": "npm run llmswitch:ensure && node scripts/build-core.mjs && node scripts/vendor-core.mjs && npm run clean && node scripts/gen-build-info.mjs && tsc && node scripts/copy-compat-assets.mjs && node scripts/copy-modules-config.mjs",
-    "build:dev": "BUILD_MODE=dev npm run build && npm run verify:e2e-toolcall && npm run verify:apply-patch && npm run verify:exec-command && npm run test:routing-instructions && npm run install:global && npm run mock:regressions && npm run verify:errorsamples",
+    "build:dev": "BUILD_MODE=dev npm run build && npm run verify:e2e-toolcall && npm run verify:apply-patch && npm run verify:apply-patch-regressions && npm run verify:exec-command && npm run test:routing-instructions && npm run install:global && npm run mock:regressions && npm run verify:errorsamples",
     "build:min": "npm run llmswitch:ensure && node scripts/build-core.mjs && node scripts/vendor-core.mjs && npm run clean && node scripts/gen-build-info.mjs && tsc && node scripts/copy-compat-assets.mjs && node scripts/copy-modules-config.mjs",
     "prepack": "echo skip-prepack",
     "postbuild": "chmod +x dist/cli.js || true",
@@ -37,7 +37,7 @@
     "dev": "tsx watch src/index.ts",
     "jest:run": "node --experimental-vm-modules ./node_modules/jest/bin/jest.js",
     "test": "npm run test:routing-instructions && npm run mock:regressions",
-    "test:routing-instructions": "npm run jest:run -- --runTestsByPath tests/server/runtime/request-executor.single-attempt.spec.ts tests/servertool/virtual-router-series-cooldown.spec.ts tests/utils/is-direct-execution.test.ts tests/utils/windows-netstat.test.ts",
+    "test:routing-instructions": "npm run jest:run -- --runTestsByPath tests/server/runtime/request-executor.single-attempt.spec.ts tests/server/runtime/executor-provider.retryable.spec.ts tests/servertool/virtual-router-series-cooldown.spec.ts tests/servertool/virtual-router-quota-routing.spec.ts tests/servertool/virtual-router-engine-update-deps.spec.ts tests/manager/quota/provider-quota-center.spec.ts tests/manager/quota/provider-quota-store.spec.ts tests/server/http-server/daemon-admin.e2e.spec.ts tests/server/http-server/quota-view-injection.spec.ts tests/server/handlers/sse-timeout.spec.ts tests/utils/is-direct-execution.test.ts tests/utils/windows-netstat.test.ts",
     "test:watch": "npm run jest:run -- --watch",
     "test:coverage": "npm run jest:run -- --coverage",
     "test:integration": "npm run jest:run -- --testPathPattern=integration",
@@ -54,6 +54,7 @@
     "prepare": "",
     "postinstall": "chmod +x dist/cli.js || true",
     "verify:apply-patch": "node scripts/verify-apply-patch.mjs",
+    "verify:apply-patch-regressions": "node scripts/verify-apply-patch-regressions.mjs",
     "verify:exec-command": "node scripts/tests/exec-command-loop.mjs",
     "verify:errorsamples": "node scripts/verify-codex-error-samples.mjs",
     "install:global": "./scripts/install-global.sh",
@@ -119,6 +120,7 @@
     "test:golden": "node scripts/tests/golden-provider-cycle.mjs",
     "test:sharedmodule": "npm run jest:run -- tests/sharedmodule/*.spec.ts",
     "sync:ci-goldens": "node scripts/tools/sync-ci-goldens.mjs",
+    "sync:apply-patch-regressions": "node scripts/tools/sync-apply-patch-regressions.mjs",
     "mock:extract": "node scripts/mock-provider/extract.mjs",
     "mock:validate": "node scripts/mock-provider/validate.mjs",
     "mock:regressions": "node scripts/mock-provider/run-regressions.mjs",
@@ -129,7 +131,7 @@
   },
   "dependencies": {
     "@anthropic-ai/sdk": "^0.65.0",
-    "@jsonstudio/llms": "^0.6.753",
+    "@jsonstudio/llms": "^0.6.802",
     "@jsonstudio/rcc": "^0.89.555",
     "@lmstudio/sdk": "^1.5.0",
     "@radix-ui/react-switch": "^1.2.6",

package/scripts/scan-apply-patch-samples.mjs

@@ -0,0 +1,362 @@
+#!/usr/bin/env node
+
+/**
+ * Scan apply_patch tool call shapes in:
+ * 1) repo samples: samples/ci-goldens/** (including _regressions/apply_patch)
+ * 2) user samples: ~/.routecodex/codex-samples/**
+ *
+ * Goal: identify remaining non-context shape failures (invalid_json/missing_changes/etc.)
+ * and optionally capture failures into repo ci-goldens regressions.
+ *
+ * Usage:
+ *   node scripts/scan-apply-patch-samples.mjs
+ *   node scripts/scan-apply-patch-samples.mjs --user-all
+ *   ROUTECODEX_APPLY_PATCH_REGRESSION_TO_REPO=1 node scripts/scan-apply-patch-samples.mjs --user-all
+ */
+
+import fs from 'node:fs/promises';
+import path from 'node:path';
+import os from 'node:os';
+import crypto from 'node:crypto';
+import { fileURLToPath, pathToFileURL } from 'node:url';
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+const repoRoot = path.resolve(__dirname, '..');
+
+const coreLoaderPath = path.join(repoRoot, 'dist', 'modules', 'llmswitch', 'core-loader.js');
+const coreLoaderUrl = pathToFileURL(coreLoaderPath).href;
+
+const HOME = os.homedir();
+const USER_CODEX_ROOT = path.join(HOME, '.routecodex', 'codex-samples');
+const USER_CODEX_ROOT_ALT = path.join(HOME, '.routecodex', 'codex samples');
+const USER_CODEX_PENDING = path.join(USER_CODEX_ROOT, 'openai-chat', '__pending__');
+const REPO_GOLDENS_ROOT = path.join(repoRoot, 'samples', 'ci-goldens');
+const ERROR_SAMPLES_ROOT = path.join(HOME, '.routecodex', 'errorsamples', 'apply_patch');
+
+const DEFAULT_CAPTURE_TOTAL_LIMIT = 5000;
+const DEFAULT_CAPTURE_PER_REASON_LIMIT = 250;
+
+async function fileExists(p) {
+  try {
+    await fs.access(p);
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+async function* walkJsonFiles(root) {
+  const stack = [root];
+  while (stack.length) {
+    const current = stack.pop();
+    let entries;
+    try {
+      entries = await fs.readdir(current, { withFileTypes: true });
+    } catch {
+      continue;
+    }
+    for (const entry of entries) {
+      const full = path.join(current, entry.name);
+      if (entry.isDirectory()) {
+        stack.push(full);
+      } else if (entry.isFile() && entry.name.toLowerCase().endsWith('.json')) {
+        yield full;
+      }
+    }
+  }
+}
+
+function extractApplyPatchArgs(doc, { allowRegressionOriginalArgs } = { allowRegressionOriginalArgs: false }) {
+  const out = [];
+
+  function visit(node) {
+    if (!node) return;
+    if (Array.isArray(node)) {
+      for (const item of node) visit(item);
+      return;
+    }
+    if (typeof node !== 'object') return;
+
+    const any = node;
+
+    // OpenAI chat tool_calls: { function: { name, arguments } }
+    if (any.function && typeof any.function === 'object') {
+      const fn = any.function;
+      if (fn && fn.name === 'apply_patch') {
+        const args = fn.arguments;
+        if (typeof args === 'string' && args.trim()) {
+          out.push(args);
+        }
+      }
+    }
+
+    // Responses output: { type:"function_call", name:"apply_patch", arguments:"..." }
+    if (any.type === 'function_call' && any.name === 'apply_patch') {
+      const args = any.arguments;
+      if (typeof args === 'string' && args.trim()) {
+        out.push(args);
+      }
+    }
+
+    // Regression sample shape: { originalArgs: "...", errorType: "..." }
+    if (allowRegressionOriginalArgs) {
+      if (
+        typeof any.originalArgs === 'string' &&
+        any.originalArgs.trim() &&
+        typeof any.errorType === 'string' &&
+        any.errorType.trim()
+      ) {
+        out.push(any.originalArgs);
+      }
+    }
+
+    for (const value of Object.values(any)) visit(value);
+  }
+
+  visit(doc);
+  return out;
+}
+
+function isContextMismatchReason(reason) {
+  // We only capture/repair shape issues; context mismatches are not actionable here.
+  // Keep this conservative: if a new reason is introduced, we still capture it unless
+  // it clearly indicates a context mismatch.
+  if (!reason) return false;
+  const r = String(reason);
+  return (
+    r.includes('context') ||
+    r.includes('expected') ||
+    r.includes('hunk') ||
+    r.includes('no_match') ||
+    r.includes('not_found')
+  );
+}
+
+function stableHash(input) {
+  return crypto.createHash('sha256').update(input).digest('hex').slice(0, 16);
+}
+
+async function captureFailure({
+  destRoot,
+  label,
+  sourceFile,
+  reason,
+  originalArgs,
+  validationResult,
+  captureState,
+}) {
+  if (isContextMismatchReason(reason)) return;
+  if (captureState?.reasonAllowList && !captureState.reasonAllowList.has(reason)) return;
+  if (!destRoot) return;
+
+  const totalLimit = captureState.totalLimit ?? DEFAULT_CAPTURE_TOTAL_LIMIT;
+  const perReasonLimit = captureState.perReasonLimit ?? DEFAULT_CAPTURE_PER_REASON_LIMIT;
+  if (captureState.totalCaptured >= totalLimit) return;
+
+  const currentReasonCount = captureState.byReason.get(reason) ?? 0;
+  if (currentReasonCount >= perReasonLimit) return;
+
+  const key = `${label}\n${reason}\n${sourceFile}\n${originalArgs}`;
+  const filename = `sample_${stableHash(key)}.json`;
+  const folder = path.join(destRoot, reason);
+  const outPath = path.join(folder, filename);
+
+  try {
+    await fs.access(outPath);
+    return;
+  } catch {
+    // continue
+  }
+
+  await fs.mkdir(folder, { recursive: true });
+  const payload = {
+    tool: 'apply_patch',
+    label,
+    reason,
+    capturedAt: new Date().toISOString(),
+    sourceFile,
+    argsSha256: crypto.createHash('sha256').update(originalArgs).digest('hex'),
+    originalArgs,
+    validationResult,
+  };
+  await fs.writeFile(outPath, JSON.stringify(payload, null, 2), 'utf-8');
+
+  captureState.totalCaptured += 1;
+  captureState.byReason.set(reason, currentReasonCount + 1);
+}
+
+async function loadValidator() {
+  if (!(await fileExists(coreLoaderPath))) {
+    throw new Error(`core-loader missing at ${coreLoaderPath} (run npm run build:dev first)`);
+  }
+  const { importCoreModule } = await import(coreLoaderUrl);
+  const { validateToolCall } = await importCoreModule('tools/tool-registry');
+  if (typeof validateToolCall !== 'function') {
+    throw new Error('validateToolCall not found in llmswitch-core tools/tool-registry');
+  }
+  return { validateToolCall };
+}
+
+async function scanRoot(label, rootDir, validateToolCall, capture) {
+  if (!(await fileExists(rootDir))) {
+    return { label, rootDir, files: 0, calls: 0, ok: 0, byReason: new Map(), examples: [] };
+  }
+
+  const byReason = new Map();
+  const examples = [];
+  let files = 0;
+  let calls = 0;
+  let ok = 0;
+
+  for await (const filePath of walkJsonFiles(rootDir)) {
+    files += 1;
+    let raw;
+    try {
+      raw = await fs.readFile(filePath, 'utf-8');
+    } catch {
+      continue;
+    }
+    // Fast prefilter: skip JSON that doesn't even mention apply_patch.
+    if (!raw.includes('apply_patch')) continue;
+
+    let doc;
+    try {
+      doc = JSON.parse(raw);
+    } catch {
+      continue;
+    }
+
+    const isRegressionFile = filePath.includes(`${path.sep}_regressions${path.sep}`);
+    const argsList = extractApplyPatchArgs(doc, { allowRegressionOriginalArgs: isRegressionFile });
+    if (!argsList.length) continue;
+
+    const dedup = Array.from(new Set(argsList));
+    for (const args of dedup) {
+      calls += 1;
+      const res = validateToolCall('apply_patch', args);
+      if (res && res.ok) {
+        ok += 1;
+        continue;
+      }
+      const reason = (res && res.reason) || 'unknown';
+      const cur = byReason.get(reason) || { count: 0 };
+      cur.count += 1;
+      byReason.set(reason, cur);
+      if (capture?.enabled) {
+        await captureFailure({
+          destRoot: capture.destRoot,
+          label,
+          sourceFile: filePath,
+          reason,
+          originalArgs: args,
+          validationResult: res,
+          captureState: capture.state,
+        });
+      }
+      if (examples.length < 12) {
+        examples.push({ file: filePath, reason });
+      }
+    }
+  }
+
+  return { label, rootDir, files, calls, ok, byReason, examples };
+}
+
+function printReport(report) {
+  const failures = report.calls - report.ok;
+  console.log(`\n[scan] ${report.label}`);
+  console.log(`root: ${report.rootDir}`);
+  console.log(`files: ${report.files}  apply_patch_calls: ${report.calls}  ok: ${report.ok}  fail: ${failures}`);
+  const entries = Array.from(report.byReason.entries()).sort((a, b) => b[1].count - a[1].count);
+  if (entries.length) {
+    console.log('failures_by_reason:');
+    for (const [reason, v] of entries.slice(0, 20)) {
+      console.log(`  - ${reason}: ${v.count}`);
+    }
+  }
+  if (report.examples.length) {
+    console.log('examples:');
+    for (const ex of report.examples) {
+      console.log(`  - ${path.relative(repoRoot, ex.file)}  reason=${ex.reason}`);
+    }
+  }
+}
+
+async function main() {
+  const { validateToolCall } = await loadValidator();
+
+  const captureEnabled = process.argv.slice(2).includes('--capture') || process.env.ROUTECODEX_SCAN_CAPTURE === '1';
+  const captureRepo = process.argv.slice(2).includes('--capture-repo') || process.env.ROUTECODEX_SCAN_CAPTURE_REPO === '1';
+  const captureRoot = process.env.ROUTECODEX_ERRORSAMPLES_DIR || ERROR_SAMPLES_ROOT;
+  const totalLimit = Number.parseInt(process.env.ROUTECODEX_CAPTURE_TOTAL_LIMIT || '', 10);
+  const perReasonLimit = Number.parseInt(process.env.ROUTECODEX_CAPTURE_PER_REASON_LIMIT || '', 10);
+  const reasonsArg = process.argv
+    .slice(2)
+    .find((arg) => arg.startsWith('--capture-reasons='))
+    ?.split('=')[1];
+  const reasonsEnv = process.env.ROUTECODEX_CAPTURE_REASONS;
+  const reasonAllowListRaw = (reasonsArg || reasonsEnv || '').trim();
+  const reasonAllowList = reasonAllowListRaw
+    ? new Set(
+        reasonAllowListRaw
+          .split(',')
+          .map((s) => s.trim())
+          .filter(Boolean)
+      )
+    : null;
+  const captureState = {
+    totalCaptured: 0,
+    totalLimit: Number.isFinite(totalLimit) ? totalLimit : DEFAULT_CAPTURE_TOTAL_LIMIT,
+    perReasonLimit: Number.isFinite(perReasonLimit) ? perReasonLimit : DEFAULT_CAPTURE_PER_REASON_LIMIT,
+    byReason: new Map(),
+    reasonAllowList,
+  };
+
+  const repo = await scanRoot('repo samples/ci-goldens', REPO_GOLDENS_ROOT, validateToolCall, {
+    enabled: captureEnabled && captureRepo,
+    destRoot: captureRoot,
+    state: captureState,
+  });
+  printReport(repo);
+
+  const userAll = process.argv.slice(2).includes('--user-all');
+  const userRoots = [];
+  if (userAll) {
+    if (await fileExists(USER_CODEX_ROOT)) userRoots.push({ label: 'user ~/.routecodex/codex-samples (all)', root: USER_CODEX_ROOT });
+    if (await fileExists(USER_CODEX_ROOT_ALT))
+      userRoots.push({ label: 'user ~/.routecodex/codex samples (all)', root: USER_CODEX_ROOT_ALT });
+  } else {
+    const pending = (await fileExists(USER_CODEX_PENDING)) ? USER_CODEX_PENDING : null;
+    if (pending) userRoots.push({ label: 'user ~/.routecodex/codex-samples/openai-chat/__pending__', root: pending });
+    else if (await fileExists(USER_CODEX_ROOT)) userRoots.push({ label: 'user ~/.routecodex/codex-samples', root: USER_CODEX_ROOT });
+    else if (await fileExists(USER_CODEX_ROOT_ALT)) userRoots.push({ label: 'user ~/.routecodex/codex samples', root: USER_CODEX_ROOT_ALT });
+    else userRoots.push({ label: 'user ~/.routecodex/codex-samples', root: USER_CODEX_ROOT });
+  }
+
+  const userReports = [];
+  for (const entry of userRoots) {
+    const report = await scanRoot(entry.label, entry.root, validateToolCall, {
+      enabled: captureEnabled,
+      destRoot: captureRoot,
+      state: captureState,
+    });
+    userReports.push(report);
+    printReport(report);
+  }
+
+  const totalCalls = repo.calls + userReports.reduce((sum, r) => sum + r.calls, 0);
+  const totalOk = repo.ok + userReports.reduce((sum, r) => sum + r.ok, 0);
+  console.log(`\n[scan] TOTAL apply_patch_calls=${totalCalls} ok=${totalOk} fail=${totalCalls - totalOk}`);
+  if (captureEnabled) {
+    console.log(
+      `[scan] captured_failures=${captureState.totalCaptured} dest=${captureRoot} (per_reason<=${captureState.perReasonLimit}, total<=${captureState.totalLimit})`
+    );
+  }
+
+  process.exitCode = totalCalls - totalOk > 0 ? 1 : 0;
+}
+
+main().catch((error) => {
+  console.error('[scan-apply-patch-samples] failed:', error?.stack || error?.message || String(error ?? 'unknown'));
+  process.exit(2);
+});