@animalabs/membrane 0.5.54 → 0.5.63

package/dist/types/provider.d.ts

@@ -126,6 +126,8 @@ export interface ProviderRequest {
     presencePenalty?: number;
     /** Frequency penalty */
     frequencyPenalty?: number;
+    /** Repetition penalty (multiplicative, vLLM/HuggingFace style) */
+    repetitionPenalty?: number;
     /** Stop sequences */
     stopSequences?: string[];
     /** Tools in provider format */
@@ -140,6 +142,13 @@ export interface ProviderRequestOptions {
     idleTimeoutMs?: number;
     /** Called with the raw API request body right before fetch */
     onRequest?: (rawRequest: unknown) => void;
+    /**
+     * Wrap native thinking deltas in <thinking>...</thinking> tags on the
+     * onChunk stream. Used by the XML formatter path so its tag-based parser
+     * tracks thinking blocks; without this, native thinking content streams
+     * indistinguishably from visible text.
+     */
+    wrapThinkingTags?: boolean;
 }
 export interface ProviderResponse {
     /** Raw response content */

package/dist/types/provider.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"provider.d.ts","sourceRoot":"","sources":["../../src/types/provider.ts"],"names":[],"mappings":"AAAA;;GAEG;AAMH,MAAM,WAAW,cAAc;IAC7B,uEAAuE;IACvE,+BAA+B,CAAC,EAAE,OAAO,CAAC;IAE1C,+DAA+D;IAC/D,wBAAwB,CAAC,EAAE,OAAO,CAAC;IAEnC,iDAAiD;IACjD,sBAAsB,CAAC,EAAE,OAAO,CAAC;IAEjC,mEAAmE;IACnE,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAE/B,+DAA+D;IAC/D,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IAExB,8BAA8B;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAMD,MAAM,WAAW,iBAAiB;IAEhC,UAAU,EAAE,OAAO,CAAC;IACpB,QAAQ,EAAE,OAAO,CAAC;IAClB,UAAU,EAAE,OAAO,CAAC;IACpB,UAAU,EAAE,OAAO,CAAC;IAGpB,eAAe,EAAE,OAAO,CAAC;IAGzB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,kBAAkB,CAAC,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC;IACvD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAG7B,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;CACzB;AAMD,MAAM,WAAW,oBAAoB;IAEnC,eAAe,EAAE,OAAO,CAAC;IACzB,YAAY,EAAE,OAAO,CAAC;IACtB,eAAe,EAAE,OAAO,CAAC;IACzB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,iBAAiB,EAAE,OAAO,CAAC;IAG3B,KAAK,EAAE,iBAAiB,CAAC;IAGzB,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,CAAC;IACxB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAG7B,MAAM,EAAE,cAAc,CAAC;CACxB;AAMD,MAAM,WAAW,YAAY;IAC3B,oCAAoC;IACpC,eAAe,EAAE,MAAM,CAAC;IAExB,qCAAqC;IACrC,gBAAgB,EAAE,MAAM,CAAC;IAEzB,0CAA0C;IAC1C,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAE9B,yCAAyC;IACzC,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAE7B,oBAAoB;IACpB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAMD,MAAM,WAAW,eAAe;IAC9B,8BAA8B;IAC9B,EAAE,EAAE,MAAM,CAAC;IAEX,qDAAqD;IACrD,QAAQ,EAAE,MAAM,CAAC;IAEjB,0BAA0B;IAC1B,WAAW,EAAE,MAAM,CAAC;IAEpB,mBAAmB;IACnB,YAAY,EAAE,oBAAoB,CAAC;IAEnC,yBAAyB;IACzB,OAAO,CAAC,EAAE,YAAY,CAAC;IAEvB,yCAAyC;IACzC,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IAEnB,kCAAkC;IAClC,UAAU,CAAC,EAAE,OAAO,CAAC;IAErB,oCAAoC;IACpC,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAMD,MAAM,WAAW,aAAa;IAC5B,mCAAmC;IACnC,eAAe,CAAC,OAAO,EAAE,MAAM,GAAG,oBAAoB,GAAG,SAAS,CAAC;IAEnE,8BAA8B;IAC9B,UAAU,CAAC,OAAO,EAAE,MAAM,GAAG,YAAY,GAAG,SAAS,CAAC;IAEtD,6BAA6B;IAC7B,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAAC;IAEvD,gCAAgC;IAChC,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS,CAAC;IAEvD,0CAA0C;IAC1C,YAAY,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAAC;IAExC,4CAA4C;IAC5C,UAAU,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,eAAe,EAAE,CAAC;CACrD;AAED,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,uBAAuB,CAAC,EAAE,OAAO,CAAC;IAClC,iBAAiB,CAAC,EAAE,OAAO,CAAC;CAC7B;AAMD,MAAM,WAAW,eAAe;IAC9B,oBAAoB;IACpB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAEtB,4CAA4C;IAC5C,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAExC,gDAAgD;IAChD,QAAQ,CACN,OAAO,EAAE,eAAe,EACxB,OAAO,CAAC,EAAE,sBAAsB,GAC/B,OAAO,CAAC,gBAAgB,CAAC,CAAC;IAE7B,+BAA+B;IAC/B,MAAM,CACJ,OAAO,EAAE,eAAe,EACxB,SAAS,EAAE,eAAe,EAC1B,OAAO,CAAC,EAAE,sBAAsB,GAC/B,OAAO,CAAC,gBAAgB,CAAC,CAAC;CAC9B;AAGD,MAAM,WAAW,eAAe;IAC9B,sCAAsC;IACtC,QAAQ,EAAE,OAAO,EAAE,CAAC;IAEpB,yEAAyE;IACzE,MAAM,CAAC,EAAE,MAAM,GAAG,OAAO,EAAE,CAAC;IAE5B,eAAe;IACf,KAAK,EAAE,MAAM,CAAC;IAEd,iBAAiB;IACjB,SAAS,EAAE,MAAM,CAAC;IAElB,kBAAkB;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,6BAA6B;IAC7B,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,qBAAqB;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,uBAAuB;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IAEzB,wBAAwB;IACxB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAE1B,qBAAqB;IACrB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IAEzB,+BAA+B;IAC/B,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;IAElB,0CAA0C;IAC1C,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC;AAED,MAAM,WAAW,sBAAsB;IACrC,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0EAA0E;IAC1E,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,8DAA8D;IAC9D,SAAS,CAAC,EAAE,CAAC,UAAU,EAAE,OAAO,KAAK,IAAI,CAAC;CAC3C;AAED,MAAM,WAAW,gBAAgB;IAC/B,2BAA2B;IAC3B,OAAO,EAAE,OAAO,CAAC;IAEjB,qCAAqC;IACrC,UAAU,EAAE,MAAM,CAAC;IAEnB,oCAAoC;IACpC,YAAY,CAAC,EAAE,MAAM,CAAC;IAEtB,+BAA+B;IAC/B,KAAK,EAAE;QACL,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;QACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;IAEF,8BAA8B;IAC9B,KAAK,EAAE,MAAM,CAAC;IAEd,oDAAoD;IACpD,UAAU,EAAE,OAAO,CAAC;IAEpB,iCAAiC;IACjC,GAAG,EAAE,OAAO,CAAC;CACd;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;IACjC,cAAc,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,KAAK,IAAI,CAAC;CAC1D"}
+{"version":3,"file":"provider.d.ts","sourceRoot":"","sources":["../../src/types/provider.ts"],"names":[],"mappings":"AAAA;;GAEG;AAMH,MAAM,WAAW,cAAc;IAC7B,uEAAuE;IACvE,+BAA+B,CAAC,EAAE,OAAO,CAAC;IAE1C,+DAA+D;IAC/D,wBAAwB,CAAC,EAAE,OAAO,CAAC;IAEnC,iDAAiD;IACjD,sBAAsB,CAAC,EAAE,OAAO,CAAC;IAEjC,mEAAmE;IACnE,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAE/B,+DAA+D;IAC/D,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IAExB,8BAA8B;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAMD,MAAM,WAAW,iBAAiB;IAEhC,UAAU,EAAE,OAAO,CAAC;IACpB,QAAQ,EAAE,OAAO,CAAC;IAClB,UAAU,EAAE,OAAO,CAAC;IACpB,UAAU,EAAE,OAAO,CAAC;IAGpB,eAAe,EAAE,OAAO,CAAC;IAGzB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,kBAAkB,CAAC,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC;IACvD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAG7B,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;CACzB;AAMD,MAAM,WAAW,oBAAoB;IAEnC,eAAe,EAAE,OAAO,CAAC;IACzB,YAAY,EAAE,OAAO,CAAC;IACtB,eAAe,EAAE,OAAO,CAAC;IACzB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,iBAAiB,EAAE,OAAO,CAAC;IAG3B,KAAK,EAAE,iBAAiB,CAAC;IAGzB,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,CAAC;IACxB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAG7B,MAAM,EAAE,cAAc,CAAC;CACxB;AAMD,MAAM,WAAW,YAAY;IAC3B,oCAAoC;IACpC,eAAe,EAAE,MAAM,CAAC;IAExB,qCAAqC;IACrC,gBAAgB,EAAE,MAAM,CAAC;IAEzB,0CAA0C;IAC1C,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAE9B,yCAAyC;IACzC,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAE7B,oBAAoB;IACpB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAMD,MAAM,WAAW,eAAe;IAC9B,8BAA8B;IAC9B,EAAE,EAAE,MAAM,CAAC;IAEX,qDAAqD;IACrD,QAAQ,EAAE,MAAM,CAAC;IAEjB,0BAA0B;IAC1B,WAAW,EAAE,MAAM,CAAC;IAEpB,mBAAmB;IACnB,YAAY,EAAE,oBAAoB,CAAC;IAEnC,yBAAyB;IACzB,OAAO,CAAC,EAAE,YAAY,CAAC;IAEvB,yCAAyC;IACzC,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IAEnB,kCAAkC;IAClC,UAAU,CAAC,EAAE,OAAO,CAAC;IAErB,oCAAoC;IACpC,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAMD,MAAM,WAAW,aAAa;IAC5B,mCAAmC;IACnC,eAAe,CAAC,OAAO,EAAE,MAAM,GAAG,oBAAoB,GAAG,SAAS,CAAC;IAEnE,8BAA8B;IAC9B,UAAU,CAAC,OAAO,EAAE,MAAM,GAAG,YAAY,GAAG,SAAS,CAAC;IAEtD,6BAA6B;IAC7B,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAAC;IAEvD,gCAAgC;IAChC,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS,CAAC;IAEvD,0CAA0C;IAC1C,YAAY,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAAC;IAExC,4CAA4C;IAC5C,UAAU,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,eAAe,EAAE,CAAC;CACrD;AAED,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,uBAAuB,CAAC,EAAE,OAAO,CAAC;IAClC,iBAAiB,CAAC,EAAE,OAAO,CAAC;CAC7B;AAMD,MAAM,WAAW,eAAe;IAC9B,oBAAoB;IACpB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAEtB,4CAA4C;IAC5C,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAExC,gDAAgD;IAChD,QAAQ,CACN,OAAO,EAAE,eAAe,EACxB,OAAO,CAAC,EAAE,sBAAsB,GAC/B,OAAO,CAAC,gBAAgB,CAAC,CAAC;IAE7B,+BAA+B;IAC/B,MAAM,CACJ,OAAO,EAAE,eAAe,EACxB,SAAS,EAAE,eAAe,EAC1B,OAAO,CAAC,EAAE,sBAAsB,GAC/B,OAAO,CAAC,gBAAgB,CAAC,CAAC;CAC9B;AAGD,MAAM,WAAW,eAAe;IAC9B,sCAAsC;IACtC,QAAQ,EAAE,OAAO,EAAE,CAAC;IAEpB,yEAAyE;IACzE,MAAM,CAAC,EAAE,MAAM,GAAG,OAAO,EAAE,CAAC;IAE5B,eAAe;IACf,KAAK,EAAE,MAAM,CAAC;IAEd,iBAAiB;IACjB,SAAS,EAAE,MAAM,CAAC;IAElB,kBAAkB;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,6BAA6B;IAC7B,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,qBAAqB;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,uBAAuB;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IAEzB,wBAAwB;IACxB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAE1B,kEAAkE;IAClE,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAE3B,qBAAqB;IACrB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IAEzB,+BAA+B;IAC/B,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;IAElB,0CAA0C;IAC1C,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC;AAED,MAAM,WAAW,sBAAsB;IACrC,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0EAA0E;IAC1E,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,8DAA8D;IAC9D,SAAS,CAAC,EAAE,CAAC,UAAU,EAAE,OAAO,KAAK,IAAI,CAAC;IAC1C;;;;;OAKG;IACH,gBAAgB,CAAC,EAAE,OAAO,CAAC;CAC5B;AAED,MAAM,WAAW,gBAAgB;IAC/B,2BAA2B;IAC3B,OAAO,EAAE,OAAO,CAAC;IAEjB,qCAAqC;IACrC,UAAU,EAAE,MAAM,CAAC;IAEnB,oCAAoC;IACpC,YAAY,CAAC,EAAE,MAAM,CAAC;IAEtB,+BAA+B;IAC/B,KAAK,EAAE;QACL,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;QACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;IAEF,8BAA8B;IAC9B,KAAK,EAAE,MAAM,CAAC;IAEd,oDAAoD;IACpD,UAAU,EAAE,OAAO,CAAC;IAEpB,iCAAiC;IACjC,GAAG,EAAE,OAAO,CAAC;CACd;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;IACjC,cAAc,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,KAAK,IAAI,CAAC;CAC1D"}

package/dist/types/request.d.ts

@@ -18,10 +18,20 @@ export interface GenerationConfig {
     presencePenalty?: number;
     /** Frequency penalty (provider-specific) */
     frequencyPenalty?: number;
+    /** Repetition penalty — multiplicative (vLLM/HuggingFace style, typically 1.0-1.2) */
+    repetitionPenalty?: number;
     /** Enable thinking/reasoning mode */
     thinking?: {
         enabled: boolean;
         budgetTokens?: number;
+        /** Thinking type for the API: 'enabled' (default, explicit budget) or 'adaptive' (model-managed) */
+        type?: 'enabled' | 'adaptive';
+        /**
+         * Controls how thinking content is returned: 'summarized' (readable summary)
+         * or 'omitted' (empty thinking field, signature only). Models like Fable 5 /
+         * Opus 4.7+ default to 'omitted' — set 'summarized' to receive thinking text.
+         */
+        display?: 'summarized' | 'omitted';
     };
     /** Image generation config (Gemini) */
     imageGeneration?: {

package/dist/types/request.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"request.d.ts","sourceRoot":"","sources":["../../src/types/request.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAMjD,MAAM,WAAW,gBAAgB;IAC/B,uBAAuB;IACvB,KAAK,EAAE,MAAM,CAAC;IAEd,iCAAiC;IACjC,SAAS,EAAE,MAAM,CAAC;IAElB,wBAAwB;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,6BAA6B;IAC7B,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,yCAAyC;IACzC,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,2CAA2C;IAC3C,eAAe,CAAC,EAAE,MAAM,CAAC;IAEzB,4CAA4C;IAC5C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAE1B,qCAAqC;IACrC,QAAQ,CAAC,EAAE;QACT,OAAO,EAAE,OAAO,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,CAAC;KACvB,CAAC;IAEF,uCAAuC;IACvC,eAAe,CAAC,EAAE;QAChB,OAAO,EAAE,OAAO,CAAC;QACjB,UAAU,EAAE,CAAC,MAAM,GAAG,OAAO,CAAC,EAAE,CAAC;QACjC,WAAW,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,GAAG,KAAK,GAAG,KAAK,CAAC;QACtD,SAAS,CAAC,EAAE,OAAO,GAAG,QAAQ,GAAG,OAAO,CAAC;KAC1C,CAAC;CACH;AAMD,MAAM,MAAM,oBAAoB,GAC5B,MAAM,GACN,YAAY,GACZ,oBAAoB,CAAC;AAEzB,MAAM,WAAW,kBAAkB;IACjC,4BAA4B;IAC5B,SAAS,EAAE,MAAM,EAAE,CAAC;IAEpB,4CAA4C;IAC5C,QAAQ,CAAC,EAAE,oBAAoB,CAAC;IAEhC,wDAAwD;IACxD,cAAc,CAAC,EAAE,MAAM,CAAC;IAExB,qEAAqE;IACrE,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAMD,MAAM,WAAW,cAAc;IAC7B,oCAAoC;IACpC,MAAM,CAAC,EAAE,WAAW,CAAC;IAErB,sCAAsC;IACtC,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,iCAAiC;IACjC,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,qCAAqC;IACrC,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC/B;AAMD,MAAM,MAAM,QAAQ,GAChB,KAAK,GACL,QAAQ,GACR,MAAM,CAAC;AAMX,MAAM,WAAW,iBAAiB;IAChC,4BAA4B;IAC5B,QAAQ,EAAE,iBAAiB,EAAE,CAAC;IAE9B,oBAAoB;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAEhB,+BAA+B;IAC/B,MAAM,EAAE,gBAAgB,CAAC;IAEzB,uBAAuB;IACvB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IAEzB,4CAA4C;IAC5C,QAAQ,CAAC,EAAE,QAAQ,CAAC;IAEpB,kCAAkC;IAClC,aAAa,CAAC,EAAE,kBAAkB,GAAG,MAAM,EAAE,CAAC;IAE9C;;;;OAIG;IACH,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAEhC;;;;OAIG;IACH,aAAa,CAAC,EAAE,OAAO,CAAC;IAExB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC;IAEvB;;;;OAIG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;IAEvB;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAE5B;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAE9B;;;;;;OAMG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,kDAAkD;IAClD,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAC1C"}
+{"version":3,"file":"request.d.ts","sourceRoot":"","sources":["../../src/types/request.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAMjD,MAAM,WAAW,gBAAgB;IAC/B,uBAAuB;IACvB,KAAK,EAAE,MAAM,CAAC;IAEd,iCAAiC;IACjC,SAAS,EAAE,MAAM,CAAC;IAElB,wBAAwB;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,6BAA6B;IAC7B,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,yCAAyC;IACzC,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,2CAA2C;IAC3C,eAAe,CAAC,EAAE,MAAM,CAAC;IAEzB,4CAA4C;IAC5C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAE1B,sFAAsF;IACtF,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAE3B,qCAAqC;IACrC,QAAQ,CAAC,EAAE;QACT,OAAO,EAAE,OAAO,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,oGAAoG;QACpG,IAAI,CAAC,EAAE,SAAS,GAAG,UAAU,CAAC;QAC9B;;;;WAIG;QACH,OAAO,CAAC,EAAE,YAAY,GAAG,SAAS,CAAC;KACpC,CAAC;IAEF,uCAAuC;IACvC,eAAe,CAAC,EAAE;QAChB,OAAO,EAAE,OAAO,CAAC;QACjB,UAAU,EAAE,CAAC,MAAM,GAAG,OAAO,CAAC,EAAE,CAAC;QACjC,WAAW,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,GAAG,KAAK,GAAG,KAAK,CAAC;QACtD,SAAS,CAAC,EAAE,OAAO,GAAG,QAAQ,GAAG,OAAO,CAAC;KAC1C,CAAC;CACH;AAMD,MAAM,MAAM,oBAAoB,GAC5B,MAAM,GACN,YAAY,GACZ,oBAAoB,CAAC;AAEzB,MAAM,WAAW,kBAAkB;IACjC,4BAA4B;IAC5B,SAAS,EAAE,MAAM,EAAE,CAAC;IAEpB,4CAA4C;IAC5C,QAAQ,CAAC,EAAE,oBAAoB,CAAC;IAEhC,wDAAwD;IACxD,cAAc,CAAC,EAAE,MAAM,CAAC;IAExB,qEAAqE;IACrE,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAMD,MAAM,WAAW,cAAc;IAC7B,oCAAoC;IACpC,MAAM,CAAC,EAAE,WAAW,CAAC;IAErB,sCAAsC;IACtC,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,iCAAiC;IACjC,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,qCAAqC;IACrC,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC/B;AAMD,MAAM,MAAM,QAAQ,GAChB,KAAK,GACL,QAAQ,GACR,MAAM,CAAC;AAMX,MAAM,WAAW,iBAAiB;IAChC,4BAA4B;IAC5B,QAAQ,EAAE,iBAAiB,EAAE,CAAC;IAE9B,oBAAoB;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAEhB,+BAA+B;IAC/B,MAAM,EAAE,gBAAgB,CAAC;IAEzB,uBAAuB;IACvB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IAEzB,4CAA4C;IAC5C,QAAQ,CAAC,EAAE,QAAQ,CAAC;IAEpB,kCAAkC;IAClC,aAAa,CAAC,EAAE,kBAAkB,GAAG,MAAM,EAAE,CAAC;IAE9C;;;;OAIG;IACH,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAEhC;;;;OAIG;IACH,aAAa,CAAC,EAAE,OAAO,CAAC;IAExB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC;IAEvB;;;;OAIG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;IAEvB;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAE5B;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAE9B;;;;;;OAMG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,kDAAkD;IAClD,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAC1C"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@animalabs/membrane",
-  "version": "0.5.54",
+  "version": "0.5.63",
   "description": "LLM middleware - a selective boundary that transforms what passes through",
   "repository": {
     "type": "git",

package/src/formatters/native.ts

@@ -385,10 +385,20 @@ export class NativeFormatter implements PrefillFormatter {
           is_error: block.isError,
         });
       } else if (block.type === 'thinking') {
+        // Round-trip thinking blocks verbatim, including the signature — the
+        // API validates it and (on display:'omitted' models) decrypts it to
+        // reconstruct the original reasoning. Signature-only blocks (empty
+        // thinking field) are valid and must be passed back unchanged.
         result.push({
           type: 'thinking',
           thinking: block.thinking,
+          ...((block as { signature?: string }).signature
+            ? { signature: (block as { signature?: string }).signature }
+            : {}),
         });
+      } else if (block.type === 'redacted_thinking') {
+        // Pass through verbatim (carries encrypted data field)
+        result.push({ ...(block as unknown as Record<string, unknown>) });
       } else if (block.type === 'document' || block.type === 'audio') {
         hasUnsupportedMedia = true;
       }

package/src/formatters/normalize-tool-pairs.ts

@@ -16,11 +16,13 @@
  * output is shipped, so producer-side bugs cannot leak the same 400 family
  * (compression-bug 5/6/7/8/9, agent-framework #37, 2026-05-22 miner stall).
  *
- * Algorithm overview (six phases): reclassify blocks by required role,
+ * Algorithm overview (eight phases): reclassify blocks by required role,
  * reflow into role-correct envelopes, hoist matching tool_results across
  * the assistant→user boundary, evict interlopers wedged between use and
  * result, synthesize `[pending]` results for trailing orphans (or signal
- * not-ready when the id is in the caller-supplied pending set), validate.
+ * not-ready when the id is in the caller-supplied pending set), drop
+ * empty envelopes, prepend a synthetic `[continuing]` user envelope when
+ * the first envelope ended up assistant-role, validate.
  */
 
 import type { ProviderMessage as LooseProviderMessage } from './types.js';
@@ -123,36 +125,108 @@ export function normalizeToolPairs(
   // Phase 5.5: suppress cache_control on/after any envelope containing
   // a synthetic block, so cache keys don't get invalidated when the
   // real result arrives in a later round.
+  //
+  // The suppression itself happens here (we know which envelope is the
+  // first synthetic by its position in the current array), but the
+  // `cache_suppressed_for_synthetic` telemetry is deferred until after
+  // phase 6 (which can drop empty envelopes before the synthetic) and
+  // phase 7 (which can prepend a `[continuing]` envelope, shifting
+  // everything by +1). The event's `envelopeIndex` must refer to the
+  // final output array so consumers can index back into it reliably.
+  // We pin the envelope by reference and recompute the index after
+  // those phases settle.
   // ---------------------------------------------------------------------
+  let pendingCacheSuppressionRef: Envelope | null = null;
   if (orphanRes.firstSyntheticEnvelope !== null) {
-    suppressCacheControlFrom(envelopes, orphanRes.firstSyntheticEnvelope, onEvent);
+    const ref = envelopes[orphanRes.firstSyntheticEnvelope]!;
+    const suppressed = suppressCacheControlFrom(envelopes, orphanRes.firstSyntheticEnvelope);
+    if (suppressed) {
+      pendingCacheSuppressionRef = ref;
+    }
   }
 
   // ---------------------------------------------------------------------
   // Phase 6: drop empty envelopes (can arise from phase 4 dropping or
-  // phase 3 hoisting), repair first-message-must-be-user, validate. We
-  // deliberately do NOT merge consecutive same-role envelopes here —
-  // that's the formatter's job.
+  // phase 3 hoisting). We deliberately do NOT merge consecutive
+  // same-role envelopes here — that's the formatter's job.
+  //
+  // The synthetic-bearing envelope (held by `pendingCacheSuppressionRef`)
+  // cannot be dropped here — phase 5 unshifts its synthetic block onto
+  // that envelope's content, so it's guaranteed non-empty.
   // ---------------------------------------------------------------------
   envelopes = envelopes.filter((e) => e.content.length > 0);
 
-  // First-message-must-be-user repair: only repair the case where the
-  // original input's first message WAS user, but re-roling moved blocks
-  // to a leading assistant envelope (e.g. misplaced thinking block).
-  // If the producer genuinely shipped an assistant-first conversation,
-  // that's a real bug and validate() will throw.
-  const originalFirstRole = input.length > 0 ? input[0]!.role : 'user';
-  if (
-    envelopes.length > 0 &&
-    envelopes[0]!.role === 'assistant' &&
-    originalFirstRole === 'user'
-  ) {
+  // ---------------------------------------------------------------------
+  // Phase 7: ensure first envelope is user-role.
+  //
+  // Anthropic requires `messages[0].role === 'user'`. The leading
+  // envelope can become assistant for two distinct reasons:
+  //
+  //   (a) Re-roling artifact — a strict-role block (thinking, tool_use)
+  //       lived under a user-role input message and phase 1+2 moved it
+  //       to a new leading assistant envelope. `originalFirstRole`
+  //       is `'user'`.
+  //
+  //   (b) Producer bug — a context strategy genuinely selected an
+  //       assistant message as the first message of its compiled view
+  //       (the 2026-05-26 reviewer postmortem: PassthroughStrategy
+  //       `selectFromEnd` cut on an assistant turn). `originalFirstRole`
+  //       is `'assistant'`.
+  //
+  // Both cases get the same repair (prepend a `[continuing]` user
+  // envelope) because deletion would lose content in case (a) — the
+  // re-roled blocks are real conversation content the producer
+  // expected to ship. The synthetic costs a leading cache miss
+  // (deterministic literal, so idempotent across identical inputs)
+  // but preserves API correctness and producer simplicity. We emit
+  // a warn-level event so telemetry can distinguish the causes and
+  // alert on (b) without coupling control flow to attribution.
+  //
+  // Idempotency: the synthetic content is a fixed literal. Running
+  // normalize twice on the same input produces identical output the
+  // second time (envelope[0] is user, gate doesn't fire).
+  // ---------------------------------------------------------------------
+  if (envelopes.length > 0 && envelopes[0]!.role === 'assistant') {
+    // `input` is guaranteed non-empty here: rebuildEnvelopes only
+    // produces envelopes when iterating input messages, so a non-empty
+    // envelopes implies a non-empty input.
+    const originalFirstRole = input[0]!.role;
+    const leadingBlockTypes = envelopes[0]!.content.map((b) => b.type);
     envelopes.unshift({ role: 'user', content: [{ type: 'text', text: '[continuing]' }] });
+    onEvent({ kind: 'leading_user_synthesized', originalFirstRole, leadingBlockTypes });
+  }
+
+  // ---------------------------------------------------------------------
+  // Deferred phase 5.5 telemetry: emit `cache_suppressed_for_synthetic`
+  // now that index-mutating phases (6, 7) have settled. The envelope
+  // reference pinned in phase 5.5 survives both — phase 6 can't drop it
+  // (the synthetic block keeps it non-empty), and phase 7 either leaves
+  // it in place or shifts it by +1 via unshift.
+  // ---------------------------------------------------------------------
+  if (pendingCacheSuppressionRef !== null) {
+    const envelopeIndex = envelopes.indexOf(pendingCacheSuppressionRef);
+    // Assertion: indexOf must succeed. Phases 6 and 7 only filter/prepend;
+    // neither can remove an envelope holding a synthetic block.
+    if (envelopeIndex < 0) {
+      throw new MembraneNormalizerError(
+        `Phase 5.5 envelope reference vanished between phase 6 and phase 7 — ` +
+          `internal bug: synthetic-bearing envelope should be reachable after both phases.`,
+        input.map(cloneMsg),
+        envelopes.map(toProviderMessage),
+      );
+    }
+    onEvent({ kind: 'cache_suppressed_for_synthetic', envelopeIndex });
   }
 
-  // Validate. When `ready === false` we intentionally have unmatched
-  // tool_uses — but ONLY the ones in `pending` are allowed to remain
-  // unsynthesized. Any other gap is a bug in phase 5 and must throw.
+  // ---------------------------------------------------------------------
+  // Phase 8: validate. When `ready === false` we intentionally have
+  // unmatched tool_uses — but ONLY the ones in `pending` are allowed to
+  // remain unsynthesized. Any other gap is a bug in phase 5 and must
+  // throw. The first-message-must-be-user branch should be unreachable
+  // after phase 7; it remains as defense-in-depth against a future
+  // phase introducing a leading assistant envelope without firing
+  // phase 7.
+  // ---------------------------------------------------------------------
   validate(envelopes, input, pending);
 
   return { messages: envelopes.map(toProviderMessage), ready };
@@ -459,8 +533,7 @@ function resolveOrphans(
 function suppressCacheControlFrom(
   envelopes: Envelope[],
   startIndex: number,
-  onEvent: (e: NormalizeEvent) => void,
-): void {
+): boolean {
   // Strip cache_control from blocks at-or-after startIndex. We must NOT
   // mutate the caller's input blocks (envelopes share references with
   // the input via rebuildEnvelopes), so clone-on-write: replace any
@@ -468,6 +541,10 @@ function suppressCacheControlFrom(
   // The envelope's content array is replaced wholesale via .map; this
   // is the only place in the normalizer that creates new block objects
   // out of existing ones (synthetics aside).
+  //
+  // Returns whether any block was actually suppressed, so the caller
+  // can decide whether to emit telemetry. Emission is deferred until
+  // after phases 6 and 7 settle the final envelope ordering.
   let suppressed = false;
   for (let i = startIndex; i < envelopes.length; i++) {
     const env = envelopes[i]!;
@@ -480,9 +557,7 @@ function suppressCacheControlFrom(
       return rest as ProviderBlock;
     });
   }
-  if (suppressed) {
-    onEvent({ kind: 'cache_suppressed_for_synthetic', envelopeIndex: startIndex });
-  }
+  return suppressed;
 }
 
 function validate(

package/src/formatters/types.ts

@@ -111,7 +111,34 @@ export type NormalizeEvent =
   | { kind: 'synthetic_pending_result'; toolUseId: string; reason: 'trailing' | 'mid_stream' }
   | { kind: 'orphan_tool_result_textified'; toolUseId: string }
   | { kind: 'pending_in_flight'; toolUseId: string }
-  | { kind: 'cache_suppressed_for_synthetic'; envelopeIndex: number };
+  | { kind: 'cache_suppressed_for_synthetic'; envelopeIndex: number }
+  | {
+      /**
+       * Fires when the first envelope after re-roling is assistant and a
+       * synthetic `[continuing]` user envelope had to be prepended to
+       * satisfy Anthropic's `messages[0].role === 'user'` requirement.
+       *
+       * `originalFirstRole` distinguishes the two causes a consumer might
+       * want to alert on separately:
+       *   - `'user'`     → re-roling artifact (a strict-role block lived
+       *                    under a user-role message and was moved to a
+       *                    new assistant envelope). Usually benign.
+       *   - `'assistant'`→ producer shipped an assistant-first messages
+       *                    list. Real producer bug worth investigating.
+       *
+       * (Empty input never reaches this event: `rebuildEnvelopes([])`
+       * returns `[]`, and the phase-7 gate requires a non-empty envelope
+       * list. So `input[0]` is always defined when this fires.)
+       *
+       * `leadingBlockTypes` is the block-type list of the now-second
+       * envelope (i.e. what came right after the synthesized user turn),
+       * useful for classifying re-roling causes (e.g. `['thinking']`
+       * vs. `['text', 'tool_use']`).
+       */
+      kind: 'leading_user_synthesized';
+      originalFirstRole: 'user' | 'assistant';
+      leadingBlockTypes: string[];
+    };
 
 // ============================================================================
 // Build Result

package/src/membrane.ts

@@ -292,6 +292,12 @@ export class Membrane {
     // These can't be handled by the text-based XML parser, so we capture and append them
     const extraContentBlocks: ContentBlock[] = [];
 
+    // Native thinking blocks from the provider (with signatures). The parser
+    // derives signature-less thinking blocks from <thinking> text (via
+    // wrapThinkingTags); signatures from these are merged into those after
+    // parsing, and signature-only blocks are prepended.
+    const providerThinkingBlocks: ContentBlock[] = [];
+
     // Transform initial request using the formatter
     let { providerRequest, prefillResult } = this.transformRequest(request, formatter);
 
@@ -385,6 +391,10 @@ export class Membrane {
           {
             signal,
             normalizedRequest: request,
+            // The tag-based parser tracks thinking via <thinking> tags — ask the
+            // provider to wrap native thinking deltas so they don't stream as
+            // visible text (see ProviderRequestOptions.wrapThinkingTags)
+            wrapThinkingTags: true,
             onRequest: (req) => {
               rawRequest = req;
               onRequest?.(req);
@@ -410,6 +420,18 @@ export class Membrane {
                 data: (block as any).data,
                 mimeType: (block as any).mimeType,
               } as ContentBlock);
+            } else if (block.type === 'thinking') {
+              // Native thinking block from the provider — carries the signature
+              // (encrypted full reasoning). Captured so consumers can persist and
+              // round-trip it for reasoning continuity. Includes signature-only
+              // blocks (display:'omitted' returns an empty thinking field).
+              providerThinkingBlocks.push({
+                type: 'thinking',
+                thinking: (block as any).thinking ?? '',
+                ...((block as any).signature ? { signature: (block as any).signature } : {}),
+              } as ContentBlock);
+            } else if (block.type === 'redacted_thinking') {
+              providerThinkingBlocks.push({ ...(block as any) } as ContentBlock);
             }
           }
         }
@@ -700,6 +722,33 @@ export class Membrane {
         response.content.push(...extraContentBlocks);
       }
 
+      // Merge provider thinking signatures into parser-derived thinking blocks
+      // (matched in stream order), and prepend any leftover provider blocks —
+      // signature-only thinking (display:'omitted') never appears in the text
+      // stream, so the parser produces no block for it. redacted_thinking
+      // blocks are always prepended verbatim.
+      if (providerThinkingBlocks.length > 0) {
+        const parsedThinking = response.content.filter(
+          (b) => b.type === 'thinking'
+        ) as Array<{ type: 'thinking'; thinking: string; signature?: string }>;
+
+        const providerThinking = providerThinkingBlocks.filter((b) => b.type === 'thinking');
+        const redacted = providerThinkingBlocks.filter((b) => b.type === 'redacted_thinking');
+
+        const matched = Math.min(providerThinking.length, parsedThinking.length);
+        for (let i = 0; i < matched; i++) {
+          const sig = (providerThinking[i] as { signature?: string }).signature;
+          if (sig) {
+            parsedThinking[i]!.signature = sig;
+          }
+        }
+
+        const leftover = providerThinking.slice(matched);
+        if (leftover.length > 0 || redacted.length > 0) {
+          response.content.unshift(...leftover, ...redacted);
+        }
+      }
+
       return response;
     } catch (error) {
       // Check if this is an abort error
@@ -1005,6 +1054,19 @@ export class Membrane {
             content: block.content,
             is_error: block.isError,
           });
+        } else if (block.type === 'thinking') {
+          // Round-trip thinking blocks verbatim including the signature — the
+          // API validates it and (on display:'omitted' models) decrypts it to
+          // reconstruct prior reasoning. Empty thinking + signature is valid.
+          content.push({
+            type: 'thinking',
+            thinking: (block as { thinking?: string }).thinking ?? '',
+            ...((block as { signature?: string }).signature
+              ? { signature: (block as { signature?: string }).signature }
+              : {}),
+          });
+        } else if (block.type === 'redacted_thinking') {
+          content.push({ ...(block as unknown as Record<string, unknown>) });
         } else if (block.type === 'image') {
           if (block.source.type === 'base64') {
             const imageBlock: Record<string, unknown> = {
@@ -1081,13 +1143,8 @@ export class Membrane {
       );
     }
 
-    // Build thinking config for native extended thinking
-    const thinking = request.config.thinking?.enabled
-      ? {
-          type: 'enabled' as const,
-          budget_tokens: request.config.thinking.budgetTokens ?? 5000,
-        }
-      : undefined;
+    // Build thinking config for native extended thinking (budget clamped to max_tokens)
+    const thinking = this.buildThinkingParam(request.config);
 
     // Anthropic requires temperature=1 when extended thinking is enabled
     const temperature = thinking ? 1 : request.config.temperature;
@@ -1172,8 +1229,10 @@ export class Membrane {
    * Used by transformRequest, buildContinuationRequest, and buildContinuationRequestWithImages.
    */
   private getBaseProviderParams(config: NormalizedRequest['config']) {
+    // Build thinking config for native extended thinking
+    const thinking = this.buildThinkingParam(config);
     // Anthropic requires temperature=1 when extended thinking is enabled
-    const temperature = config.thinking?.enabled ? 1 : config.temperature;
+    const temperature = thinking ? 1 : config.temperature;
     return {
       model: config.model,
       maxTokens: config.maxTokens,
@@ -1182,9 +1241,41 @@ export class Membrane {
       topK: config.topK,
       presencePenalty: config.presencePenalty,
       frequencyPenalty: config.frequencyPenalty,
+      repetitionPenalty: config.repetitionPenalty,
+      thinking,
     };
   }
 
+  /**
+   * Build the provider thinking parameter from config.
+   *
+   * For type 'enabled', the API requires max_tokens > budget_tokens and a
+   * minimum budget of 1024 — a misconfigured budget (e.g., default 10000 with
+   * max_tokens 4096) is clamped to fit. If no valid budget fits (max_tokens
+   * too small), thinking is omitted entirely rather than sending a request
+   * the API will reject.
+   */
+  private buildThinkingParam(config: NormalizedRequest['config']):
+    | { type: 'adaptive'; display?: 'summarized' | 'omitted' }
+    | { type: 'enabled'; budget_tokens: number; display?: 'summarized' | 'omitted' }
+    | undefined {
+    if (!config.thinking?.enabled) return undefined;
+
+    const display = config.thinking.display;
+    if ((config.thinking.type ?? 'enabled') === 'adaptive') {
+      return { type: 'adaptive', ...(display ? { display } : {}) };
+    }
+
+    const requested = config.thinking.budgetTokens ?? 5000;
+    const maxTokens = typeof config.maxTokens === 'number' ? config.maxTokens : undefined;
+    const budget = maxTokens !== undefined ? Math.min(requested, maxTokens - 1024) : requested;
+    if (budget < 1024) {
+      // Can't fit a valid thinking budget under max_tokens — skip thinking
+      return undefined;
+    }
+    return { type: 'enabled', budget_tokens: budget, ...(display ? { display } : {}) };
+  }
+
   /**
    * Transform a normalized request into provider format using the formatter
    */
@@ -1232,6 +1323,15 @@ export class Membrane {
       },
     };
 
+    // The API rejects extended thinking combined with an assistant prefill.
+    // Prefill-style builds (XML formatter) use the thinking config for the
+    // literal `<thinking>` text prefix instead of the API feature — drop the
+    // API param when the built request actually ends in an assistant prefill.
+    // Chat-style builds (no prefill) keep it.
+    if (buildResult.assistantPrefill && providerRequest.thinking) {
+      delete providerRequest.thinking;
+    }
+
     return { providerRequest, prefillResult: buildResult };
   }
 
@@ -1243,6 +1343,8 @@ export class Membrane {
       timeoutMs?: number;
       idleTimeoutMs?: number;
       onRequest?: (rawRequest: unknown) => void;
+      /** See ProviderRequestOptions.wrapThinkingTags */
+      wrapThinkingTags?: boolean;
       /**
        * The original NormalizedRequest, threaded through so the
        * `beforeRequest` hook can see both shapes (normalized + provider).
@@ -1292,6 +1394,9 @@ export class Membrane {
     
     return {
       ...this.getBaseProviderParams(originalRequest.config),
+      // Continuations always end in an assistant prefill — the API rejects
+      // extended thinking combined with prefill, so never send the param here
+      thinking: undefined,
       messages,
       system: prefillResult.systemContent
         ? (Array.isArray(prefillResult.systemContent) && prefillResult.systemContent.length > 0
@@ -1362,6 +1467,9 @@ export class Membrane {
 
     return {
       ...this.getBaseProviderParams(originalRequest.config),
+      // Continuations always end in an assistant prefill — the API rejects
+      // extended thinking combined with prefill, so never send the param here
+      thinking: undefined,
       messages,
       system: prefillResult.systemContent
         ? (Array.isArray(prefillResult.systemContent) && prefillResult.systemContent.length > 0
@@ -1595,6 +1703,11 @@ export class Membrane {
         return 'stop_sequence';
       case 'tool_use':
         return 'tool_use';
+      case 'refusal':
+        // Safety refusal (e.g., Fable 5 reasoning_extraction). Must survive
+        // mapping — downstream consumers react to refusals (chapterx adds a
+        // Discord reaction). Defaulting this to end_turn silently hid them.
+        return 'refusal';
       default:
         return 'end_turn';
     }
@@ -2483,13 +2596,16 @@ export class Membrane {
 }
 
 // Native tool names must match ^[a-zA-Z0-9_-]{1,128}$.
-// The framework uses module:tool namespacing, so we round-trip colons
-// through an escape encoding for the API wire format.
-// Lossless: escape underscores first (_u), then encode colons (_c).
+// Tool names use `--` namespacing, which is already API-valid; the only
+// character that ever needs escaping is a literal colon, encoded losslessly as
+// `__` and back. We deliberately do NOT escape underscores — they are valid,
+// and escaping them (the previous `_u`/`_c` scheme) garbled every
+// underscore-containing tool name in the request the model actually sees
+// (`send_message` → `send_umessage`), polluting its reasoning for no benefit.
 function sanitizeToolName(name: string): string {
-  return name.replace(/_/g, '_u').replace(/:/g, '_c');
+  return name.replace(/:/g, '__');
 }
 
 function unsanitizeToolName(name: string): string {
-  return name.replace(/_c/g, ':').replace(/_u/g, '_');
+  return name.replace(/__/g, ':');
 }