@askalf/dario 4.8.22 → 4.8.24

package/dist/cc-template.d.ts

@@ -264,6 +264,10 @@ export declare function resolveMaxTokens(flag: number | 'client' | undefined, cl
 /** Valid values for the `--effort` flag. Mirrors CC's effort set (`low|medium|high|xhigh|max`) plus CC's `ultracode` mode and dario's pseudo-value `'client'` for passthrough. `'ultracode'` is CC's xhigh-plus-dynamic-workflow-orchestration mode (CC 2.1.154); the Messages API accepts only low|medium|high|xhigh|max, so dario normalizes ultracode → 'xhigh' on the wire (see normalizeEffortForWire). `'client'` passes through the client's own `output_config.effort` (falling back to `'xhigh'`). dario#87, `'max'` added in dario#190, `'ultracode'` added 2026-05-28. */
 export type EffortValue = 'low' | 'medium' | 'high' | 'xhigh' | 'ultracode' | 'max' | 'client';
 export declare const VALID_EFFORT_VALUES: ReadonlyArray<EffortValue>;
+export declare function parseEffortSuffix(model: string): {
+    model: string;
+    effort?: EffortValue;
+};
 /**
  * Resolve the outbound `output_config.effort` value.
  *

package/dist/cc-template.js

@@ -893,6 +893,27 @@ export function resolveMaxTokens(flag, clientBody) {
     return flag;
 }
 export const VALID_EFFORT_VALUES = ['low', 'medium', 'high', 'xhigh', 'ultracode', 'max', 'client'];
+/**
+ * dario#419 — strip an optional effort suffix off a model name, so OpenAI-compat
+ * clients that can't set `output_config.effort` (e.g. Cursor) can choose effort
+ * by model name: `opus-4-8:high` (colon) or Cursor-style `claude-opus-4-8-high`
+ * (hyphen). Only the wire-valid effort levels are recognized as a suffix — any
+ * other trailing token is left as part of the model name, and a bare model that
+ * IS an effort word (e.g. just "high") is left alone. Returns the model with the
+ * suffix removed plus the parsed effort (undefined when none). Exported for tests.
+ */
+const SUFFIX_EFFORTS = ['ultracode', 'medium', 'xhigh', 'high', 'low', 'max'];
+export function parseEffortSuffix(model) {
+    for (const e of SUFFIX_EFFORTS) {
+        for (const sep of [':', '-']) {
+            const tag = sep + e;
+            if (model.length > tag.length && model.endsWith(tag)) {
+                return { model: model.slice(0, -tag.length), effort: e };
+            }
+        }
+    }
+    return { model };
+}
 /**
  * Normalize an effort value to a wire-valid `output_config.effort`. The
  * Messages API accepts only low|medium|high|xhigh|max. CC's `ultracode` is a

package/dist/live-fingerprint.d.ts

@@ -282,7 +282,7 @@ export declare function _resetInstalledVersionProbeForTest(): void;
  */
 export declare const SUPPORTED_CC_RANGE: {
     readonly min: "1.0.0";
-    readonly maxTested: "2.1.159";
+    readonly maxTested: "2.1.160";
 };
 /**
  * Compare two dotted-numeric version strings. Returns negative if `a<b`,

package/dist/live-fingerprint.js

@@ -786,7 +786,7 @@ export function _resetInstalledVersionProbeForTest() {
  */
 export const SUPPORTED_CC_RANGE = {
     min: '1.0.0',
-    maxTested: '2.1.159',
+    maxTested: '2.1.160',
 };
 /**
  * Compare two dotted-numeric version strings. Returns negative if `a<b`,

package/dist/proxy.js

@@ -6,7 +6,7 @@ import { join } from 'node:path';
 import { homedir } from 'node:os';
 import { arch, platform } from 'node:process';
 import { getAccessToken, getStatus } from './oauth.js';
-import { buildCCRequest, reverseMapResponse, createStreamingReverseMapper, orderHeadersForOutbound, CC_TEMPLATE } from './cc-template.js';
+import { buildCCRequest, parseEffortSuffix, reverseMapResponse, createStreamingReverseMapper, orderHeadersForOutbound, CC_TEMPLATE } from './cc-template.js';
 import { describeTemplate, detectDrift, checkCCCompat } from './live-fingerprint.js';
 import { AccountPool, computeStickyKey, parseRateLimits, modelFamily, isInAuthCooldown, authCooldownMs } from './pool.js';
 import { Analytics, billingBucketFromClaim } from './analytics.js';
@@ -1242,6 +1242,7 @@ export async function startProxy(opts = {}) {
             // built-in `claude-*` name collision (Cursor reroutes any name it
             // recognizes through its own Anthropic gateway, bypassing localhost).
             let forcedProvider = cliProviderOverride;
+            let requestEffort; // dario#419 — per-request effort parsed from a model-name suffix (model:high / model-high)
             if (body.length > 0) {
                 try {
                     const parsed = JSON.parse(body.toString());
@@ -1249,14 +1250,26 @@ export async function startProxy(opts = {}) {
                     const prefix = parseProviderPrefix(rawModel);
                     if (prefix) {
                         forcedProvider = prefix.provider;
+                        // dario#419 — optional effort suffix (model:high / model-high). Claude
+                        // path ONLY: OpenAI backends keep their own -high/-low suffixes, so we
+                        // strip it before the alias lookup only when routing to the subscription.
+                        let providerModel = prefix.model;
+                        if (prefix.provider === 'claude') {
+                            const eff = parseEffortSuffix(providerModel);
+                            if (eff.effort) {
+                                requestEffort = eff.effort;
+                                providerModel = eff.model;
+                            }
+                        }
                         const resolvedModel = prefix.provider === 'claude'
-                            ? resolveClaudeAlias(prefix.model)
-                            : prefix.model;
+                            ? resolveClaudeAlias(providerModel)
+                            : providerModel;
                         parsed.model = resolvedModel;
                         body = Buffer.from(JSON.stringify(parsed));
                         if (verbose) {
-                            const aliasNote = resolvedModel !== prefix.model ? ` (alias: ${prefix.model} → ${resolvedModel})` : '';
-                            console.log(`[dario] provider prefix: ${rawModel} → ${prefix.provider} backend with model ${resolvedModel}${aliasNote}`);
+                            const aliasNote = resolvedModel !== providerModel ? ` (alias: ${providerModel} → ${resolvedModel})` : '';
+                            const effNote = requestEffort ? ` (effort: ${requestEffort})` : '';
+                            console.log(`[dario] provider prefix: ${rawModel} → ${prefix.provider} backend with model ${resolvedModel}${aliasNote}${effNote}`);
                         }
                     }
                     else if (cliProviderOverride === 'openai' && cliModelRaw) {
@@ -1266,6 +1279,19 @@ export async function startProxy(opts = {}) {
                         parsed.model = cliModelRaw;
                         body = Buffer.from(JSON.stringify(parsed));
                     }
+                    else if (!isOpenAIModel(rawModel) && forcedProvider !== 'openai') {
+                        // dario#419 — bare Claude model name carrying an effort suffix with no
+                        // provider prefix (e.g. Cursor's `claude-opus-4-8-high`). OpenAI-bound
+                        // models are excluded so their own suffixes pass through untouched.
+                        const eff = parseEffortSuffix(rawModel);
+                        if (eff.effort) {
+                            requestEffort = eff.effort;
+                            parsed.model = eff.model;
+                            body = Buffer.from(JSON.stringify(parsed));
+                            if (verbose)
+                                console.log(`[dario] effort suffix: ${rawModel} → model ${eff.model} (effort: ${eff.effort})`);
+                        }
+                    }
                 }
                 catch { /* not JSON — fall through */ }
             }
@@ -1388,7 +1414,7 @@ export async function startProxy(opts = {}) {
                             hybridTools: opts.hybridTools ?? false,
                             mergeTools: opts.mergeTools ?? false,
                             noAutoDetect: opts.noAutoDetect ?? false,
-                            effort: opts.effort,
+                            effort: requestEffort ?? opts.effort,
                             maxTokens: opts.maxTokens,
                             systemPrompt: opts.systemPrompt,
                             skipFields,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@askalf/dario",
-  "version": "4.8.22",
+  "version": "4.8.24",
   "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": {