@askalf/dario 4.8.78 → 4.8.80

package/dist/cc-template-data.json

@@ -1,5 +1,5 @@
 {
-  "_version": "2.1.177",
+  "_version": "2.1.178",
   "_captured": "2026-06-11T20:43:54.573Z",
   "_source": "bundled",
   "_schemaVersion": 3,
@@ -1294,7 +1294,7 @@
   "anthropic_beta": "claude-code-20250219,interleaved-thinking-2025-05-14,thinking-token-count-2026-05-13,context-management-2025-06-27,prompt-caching-scope-2026-01-05,mid-conversation-system-2026-04-07,advisor-tool-2026-03-01,effort-2025-11-24",
   "header_values": {
     "accept": "application/json",
-    "user-agent": "claude-cli/2.1.177 (external, sdk-cli)",
+    "user-agent": "claude-cli/2.1.178 (external, sdk-cli)",
     "x-stainless-arch": "x64",
     "x-stainless-lang": "js",
     "x-stainless-os": "Linux",
@@ -1319,5 +1319,5 @@
     "output_config",
     "stream"
   ],
-  "_supportedMaxTested": "2.1.177"
+  "_supportedMaxTested": "2.1.178"
 }

package/dist/proxy.d.ts

@@ -98,6 +98,21 @@ export declare function parseEffortRejection(body: string): {
     rejected: string;
     supported: string[];
 } | null;
+/**
+ * Detect upstream's HARD effort rejection — a model with NO output_config.effort
+ * support at all (it predates the parameter), distinct from the SOFT level
+ * rejection parsed by parseEffortRejection:
+ *
+ *   400 {"type":"invalid_request_error","message":"This model does not
+ *        support the effort parameter."}
+ *
+ * Observed live 2026-06-16 on claude-opus-4-1-20250805 and
+ * claude-sonnet-4-5-20250929 — the autodetected catalog exposes pre-effort
+ * models, and the box's DARIO_EFFORT=max pin hard-400s them. There is no
+ * supported tier to clamp to here, so effort must be STRIPPED — modelled as
+ * an EMPTY supported set in effortSupportByModel. Exported for tests.
+ */
+export declare function isEffortParamUnsupported(body: string): boolean;
 /**
  * Pick the strongest effort level a model says it supports. Preference is
  * descending capability — the caller asked for more than the model can do,
@@ -105,6 +120,20 @@ export declare function parseEffortRejection(body: string): {
  */
 export declare const EFFORT_PREFERENCE: readonly string[];
 export declare function bestSupportedEffort(supported: readonly string[]): string;
+/**
+ * Parse upstream's max_tokens-cap rejection:
+ *
+ *   400 {"type":"invalid_request_error","message":"max_tokens: 64000 > 32000,
+ *        which is the maximum allowed number of output tokens for
+ *        claude-opus-4-1-20250805."}
+ *
+ * Observed live 2026-06-16 — dario pins DEFAULT_MAX_TOKENS (64000, CC 2.1.143's
+ * wire default), but the autodetected catalog exposes older models with a lower
+ * per-model output cap (e.g. opus-4-1: 32000). Returns the model's cap (the
+ * clamp target) or null for any other 400. Same learn-once-and-cache shape as
+ * parseEffortRejection. Exported for tests.
+ */
+export declare function parseMaxTokensRejection(body: string): number | null;
 /**
  * Resolve an inbound API path to its upstream target + forwarding mode.
  * Allowlist semantics — anything unlisted is 403'd (prevents SSRF through

package/dist/proxy.js

@@ -324,6 +324,23 @@ export function parseEffortRejection(body) {
     const supported = m[2].split(',').map((s) => s.trim().toLowerCase()).filter((s) => s.length > 0);
     return supported.length > 0 ? { rejected: m[1], supported } : null;
 }
+/**
+ * Detect upstream's HARD effort rejection — a model with NO output_config.effort
+ * support at all (it predates the parameter), distinct from the SOFT level
+ * rejection parsed by parseEffortRejection:
+ *
+ *   400 {"type":"invalid_request_error","message":"This model does not
+ *        support the effort parameter."}
+ *
+ * Observed live 2026-06-16 on claude-opus-4-1-20250805 and
+ * claude-sonnet-4-5-20250929 — the autodetected catalog exposes pre-effort
+ * models, and the box's DARIO_EFFORT=max pin hard-400s them. There is no
+ * supported tier to clamp to here, so effort must be STRIPPED — modelled as
+ * an EMPTY supported set in effortSupportByModel. Exported for tests.
+ */
+export function isEffortParamUnsupported(body) {
+    return /does not support the effort parameter/i.test(body);
+}
 /**
  * Pick the strongest effort level a model says it supports. Preference is
  * descending capability — the caller asked for more than the model can do,
@@ -336,6 +353,26 @@ export function bestSupportedEffort(supported) {
             return e;
     return supported[0] ?? 'high';
 }
+/**
+ * Parse upstream's max_tokens-cap rejection:
+ *
+ *   400 {"type":"invalid_request_error","message":"max_tokens: 64000 > 32000,
+ *        which is the maximum allowed number of output tokens for
+ *        claude-opus-4-1-20250805."}
+ *
+ * Observed live 2026-06-16 — dario pins DEFAULT_MAX_TOKENS (64000, CC 2.1.143's
+ * wire default), but the autodetected catalog exposes older models with a lower
+ * per-model output cap (e.g. opus-4-1: 32000). Returns the model's cap (the
+ * clamp target) or null for any other 400. Same learn-once-and-cache shape as
+ * parseEffortRejection. Exported for tests.
+ */
+export function parseMaxTokensRejection(body) {
+    const m = body.match(/max_tokens:\s*\d+\s*>\s*(\d+),?\s*which is the maximum allowed/i);
+    if (!m)
+        return null;
+    const cap = Number(m[1]);
+    return Number.isFinite(cap) && cap > 0 ? cap : null;
+}
 /**
  * Resolve an inbound API path to its upstream target + forwarding mode.
  * Allowlist semantics — anything unlisted is 403'd (prevents SSRF through
@@ -977,6 +1014,12 @@ export async function startProxy(opts = {}) {
     // "does not support effort level" 400 (see parseEffortRejection); consulted
     // up front at body-build time so capped models never re-pay the rejection.
     const effortSupportByModel = new Map();
+    // Per-model max_tokens cap cache — same pay-the-round-trip-once shape as
+    // effortSupportByModel. Populated from upstream's "max_tokens: X > Y, which
+    // is the maximum allowed" 400 (see parseMaxTokensRejection); consulted up
+    // front so older models capped below dario's DEFAULT_MAX_TOKENS pin never
+    // re-pay the rejection.
+    const maxTokensCapByModel = new Map();
     // Beta flag set — sourced from the live template when the capture recorded
     // one (schema v2+), else falls back to the v2.1.104 bundled default. Same
     // fallback string shim/runtime.cjs uses (kept in sync so proxy and shim
@@ -1763,7 +1806,25 @@ export async function startProxy(opts = {}) {
                         const supportedEfforts = effortSupportByModel.get(r.model);
                         const oc = r.output_config;
                         if (supportedEfforts && oc && typeof oc.effort === 'string' && !supportedEfforts.includes(oc.effort)) {
-                            oc.effort = bestSupportedEffort(supportedEfforts);
+                            if (supportedEfforts.length === 0) {
+                                // HARD rejection learned — this model has no effort support at
+                                // all; strip the field (no tier to clamp to). See the retry
+                                // branch below + isEffortParamUnsupported.
+                                delete oc.effort;
+                                if (Object.keys(oc).length === 0)
+                                    delete r.output_config;
+                            }
+                            else {
+                                oc.effort = bestSupportedEffort(supportedEfforts);
+                            }
+                        }
+                        // max_tokens cap clamp — when a prior request taught us this model's
+                        // per-model output cap (older models cap below dario's
+                        // DEFAULT_MAX_TOKENS pin), clamp up front instead of re-paying the
+                        // 400. In-place value mutation: field order (a fingerprint) is kept.
+                        const maxTokensCap = maxTokensCapByModel.get(r.model);
+                        if (maxTokensCap !== undefined && typeof r.max_tokens === 'number' && r.max_tokens > maxTokensCap) {
+                            r.max_tokens = maxTokensCap;
                         }
                     }
                     // dario#528: overwrite the random cch placeholder with the real
@@ -2117,6 +2178,122 @@ export async function startProxy(opts = {}) {
                             return;
                         }
                     }
+                    else if (upstream.status === 400 && isEffortParamUnsupported(peekedBody) && finalBody) {
+                        // HARD effort rejection — this model has no output_config.effort
+                        // support at all (it predates the parameter; e.g. opus-4-1 /
+                        // sonnet-4-5 surfaced by the autodetected catalog under the box's
+                        // DARIO_EFFORT=max pin). Unlike the SOFT level rejection above there
+                        // is no supported tier to clamp to, so STRIP the field, retry once,
+                        // and cache an EMPTY supported set per model so the up-front consult
+                        // strips it on every later request without re-paying the round-trip.
+                        let retried = false;
+                        try {
+                            const rb = JSON.parse(finalBody.toString('utf8'));
+                            const wireModel = typeof rb.model === 'string' ? rb.model : '';
+                            const oc = rb.output_config;
+                            if (wireModel && oc && typeof oc.effort === 'string') {
+                                const firstRejection = !effortSupportByModel.has(wireModel);
+                                effortSupportByModel.set(wireModel, []); // empty set = no effort support → strip
+                                if (verbose && firstRejection)
+                                    console.log(`[dario] #${requestCount} effort parameter unsupported by ${wireModel} — retrying without output_config.effort (stripped for this model going forward)`);
+                                delete oc.effort;
+                                if (Object.keys(oc).length === 0)
+                                    delete rb.output_config;
+                                finalBody = Buffer.from(JSON.stringify(rb));
+                                const retry = await fetch(targetBase, {
+                                    method: req.method ?? 'POST',
+                                    headers: passthrough ? headers : orderHeadersForOutbound(headers),
+                                    body: new Uint8Array(finalBody),
+                                    signal: upstreamAbort.signal,
+                                });
+                                upstream = retry;
+                                peekedBody = null;
+                                retried = true;
+                                if (pool && poolAccount) {
+                                    const retrySnapshot = parseRateLimits(upstream.headers);
+                                    if (upstream.status === 429) {
+                                        pool.markRejected(poolAccount.alias, retrySnapshot);
+                                    }
+                                    else {
+                                        pool.updateRateLimits(poolAccount.alias, retrySnapshot);
+                                    }
+                                }
+                            }
+                        }
+                        catch { /* body not JSON — forward the original 400 below */ }
+                        if (!retried) {
+                            // Couldn't rebuild the body — forward the upstream 400 (already consumed).
+                            const responseHeaders = {
+                                'Content-Type': upstream.headers.get('content-type') ?? 'application/json',
+                                'Access-Control-Allow-Origin': corsOrigin,
+                                ...SECURITY_HEADERS,
+                            };
+                            for (const [key, value] of upstream.headers.entries()) {
+                                if (key === 'request-id')
+                                    responseHeaders[key] = value;
+                            }
+                            requestCount++;
+                            res.writeHead(400, responseHeaders);
+                            res.end(peekedBody);
+                            return;
+                        }
+                    }
+                    else if (upstream.status === 400 && parseMaxTokensRejection(peekedBody) !== null && finalBody) {
+                        // max_tokens-cap rejection — dario's DEFAULT_MAX_TOKENS pin exceeds
+                        // this (older) model's per-model output cap (e.g. opus-4-1 caps at
+                        // 32000 where newer models allow 64000). Clamp max_tokens to the cap
+                        // the error reports, retry once, and cache per model so the up-front
+                        // consult clamps every later request without the round-trip.
+                        const cap = parseMaxTokensRejection(peekedBody);
+                        let retried = false;
+                        try {
+                            const rb = JSON.parse(finalBody.toString('utf8'));
+                            const wireModel = typeof rb.model === 'string' ? rb.model : '';
+                            if (wireModel && typeof rb.max_tokens === 'number' && rb.max_tokens > cap) {
+                                const firstRejection = !maxTokensCapByModel.has(wireModel);
+                                maxTokensCapByModel.set(wireModel, cap);
+                                if (verbose && firstRejection)
+                                    console.log(`[dario] #${requestCount} max_tokens ${rb.max_tokens} exceeds ${wireModel} cap ${cap} — retrying clamped (cached per model)`);
+                                rb.max_tokens = cap; // in-place value mutation — field order untouched
+                                finalBody = Buffer.from(JSON.stringify(rb));
+                                const retry = await fetch(targetBase, {
+                                    method: req.method ?? 'POST',
+                                    headers: passthrough ? headers : orderHeadersForOutbound(headers),
+                                    body: new Uint8Array(finalBody),
+                                    signal: upstreamAbort.signal,
+                                });
+                                upstream = retry;
+                                peekedBody = null;
+                                retried = true;
+                                if (pool && poolAccount) {
+                                    const retrySnapshot = parseRateLimits(upstream.headers);
+                                    if (upstream.status === 429) {
+                                        pool.markRejected(poolAccount.alias, retrySnapshot);
+                                    }
+                                    else {
+                                        pool.updateRateLimits(poolAccount.alias, retrySnapshot);
+                                    }
+                                }
+                            }
+                        }
+                        catch { /* body not JSON — forward the original 400 below */ }
+                        if (!retried) {
+                            // Couldn't rebuild the body — forward the upstream 400 (already consumed).
+                            const responseHeaders = {
+                                'Content-Type': upstream.headers.get('content-type') ?? 'application/json',
+                                'Access-Control-Allow-Origin': corsOrigin,
+                                ...SECURITY_HEADERS,
+                            };
+                            for (const [key, value] of upstream.headers.entries()) {
+                                if (key === 'request-id')
+                                    responseHeaders[key] = value;
+                            }
+                            requestCount++;
+                            res.writeHead(400, responseHeaders);
+                            res.end(peekedBody);
+                            return;
+                        }
+                    }
                     else if (isLongContextError) {
                         // Cache the rejection so future requests on this account skip
                         // context-1m up front instead of re-paying the 400/429 round-trip.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@askalf/dario",
-  "version": "4.8.78",
+  "version": "4.8.80",
   "description": "Use your Claude Pro/Max subscription in any tool — Cursor, Cline, Aider, the Agent SDK, your scripts — at subscription pricing, not per-token API bills. One local Anthropic + OpenAI-compatible endpoint.",
   "type": "module",
   "bin": {