bare-agent 0.50.0 → 0.51.0

package/bareagent.context.md

@@ -513,6 +513,14 @@ new OpenAIProvider({ baseUrl: 'https://api.deepseek.com', model: 'deepseek-flash
 
 > **Fallback caveat:** per-call options (incl. `thinking`) are forwarded to EVERY provider in a `FallbackProvider` chain. For a mixed-vendor chain set `thinking` on each provider's constructor, not per-call.
 
+**Normalized `thinkingMode: 'on' | 'off'`.** Say "on"/"off"; bare-agent sends the right field for the provider/model (`AnthropicProvider`, `OpenAIProvider` incl. DeepSeek, `GeminiProvider`, `OllamaProvider`): constructor option, overridable per call (`generate(..., { thinkingMode })` / `loop.run(msgs, tools, { thinkingMode })`), `null` suppresses. It is a NEW field — the raw `thinking` option above is untouched, and setting both is a `ValidationError`. It tries the known request shapes (a 400 naming the field walks the ladder; the winner is cached), throws a `ValidationError` naming what it tried if every shape is rejected, and throws before sending when the model is measured as not switchable (`claude-opus-5-5` off, `gemini-3.1-pro-preview` off). Every reply is checked: `result.thinkingApplied` is `true` (accepted; for `'off'`, zero thinking seen), `false` (`'off'` sent but the model thought anyway — warns once, the result is kept), or `null` (cannot tell; e.g. OpenAI without `reasoning_tokens`). On `loop.run()` it is sticky-false: any round `false` ⇒ the run is `false`. Anthropic `'on'` + a non-default `temperature` (the API rejects it with thinking enabled): the temperature is dropped and retried once, `temperatureDropped: true`. `'on'` is never flagged (adaptive models may skip thinking). `thinkingSupport(provider, model)` from `bare-agent/providers` tells you, offline, what is known. DeepSeek `'on'` + a forced `toolChoice` is a `ValidationError`; CLIPipe has no mechanism and throws. Ollama's `think` is not live-verified.
+
+```js
+const p = new AnthropicProvider({ apiKey, model: 'claude-sonnet-5', thinkingMode: 'off' });
+const r = await p.generate(msgs, [], {});            // r.thinkingApplied === true
+await p.generate(msgs, [], { thinkingMode: 'on' });  // per-call override
+```
+
 > **Honesty note, and we mean it.** This is a **protocol/data-loss fix, not a capability fix.** A head-to-head with thinking fully preserved vs. stock bare-agent produced **indistinguishable** outcomes. Do not adopt 0.27 expecting better reasoning — you will not get it, and we have the measurement.
 
 Legacy `wrapTool` / `wrapTools` are retained as deprecation shims (one-shot console warning, removal in 1.0). Migration: replace `wrapTools(tools)` at `loop.run()` with `filterTools(tools)` once upfront + `onLlmResult` / `onToolResult` on `new Loop({...})` to pick up LLM-cost recording and `_ctx` threading.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bare-agent",
-  "version": "0.50.0",
+  "version": "0.51.0",
   "files": [
     "index.js",
     "index.d.ts",

package/primitives.json

@@ -442,6 +442,15 @@
       "fails": "never throws on emit; a transport write error is isolated and does not interrupt the run.",
       "example": "const stream = new Stream({ transport });\nstream.emit('loop:round', { n: 1 });"
     },
+    {
+      "name": "thinkingSupport",
+      "category": "providers",
+      "when": "you need to know, before building a request, whether a provider/model can have thinking turned on or off and which request field does it — pure lookup, no network",
+      "import": "import { thinkingSupport } from 'bare-agent/providers'",
+      "signature": "thinkingSupport(provider: 'anthropic'|'openai'|'deepseek'|'gemini'|'ollama'|'clipipe'|string, model?: string) => { switchable: boolean|'unverified', off: string|null, on: string|null, note: string }",
+      "fails": "returns switchable:false for clipipe and unknown providers (never throws); 'unverified' means not measured, not unsupported; Ollama's shapes are not live-verified.",
+      "example": "thinkingSupport('anthropic', 'claude-opus-5-5'); // { switchable: false, off: null, on: \"thinking:{type:'adaptive'}\", note: '...' }"
+    },
     {
       "name": "unitAssembler",
       "category": "context",

package/src/loop.d.ts

@@ -167,7 +167,7 @@ export class Loop {
      *   thunk is re-evaluated each round (D4/eval-assist F2) so a tool set that grows mid-run — e.g. a skill
      *   unlocking its tools — is offered on the next round; a static array is resolved once at wire time.
      * @param {Record<string, any>} [options={}] - Per-run overrides (system, temperature, ctx, etc.).
-     * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean}>}
+     * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean, thinkingApplied?: boolean|null}>}
      *   On halt the returned `error` is `halt:<rule>` (or `halt:unknown` if the
      *   thrown HaltError carried no `rule`), and `msgs` is sanitized so any
      *   dangling assistant `tool_calls` from the halted round are paired with
@@ -207,6 +207,7 @@ export class Loop {
         metrics: RunMetrics;
         malformedToolCall?: GenerateResult["malformedToolCall"];
         temperatureDropped?: boolean;
+        thinkingApplied?: boolean | null;
     }>;
     _warnedTruncated: boolean | undefined;
     /**
@@ -235,7 +236,7 @@ export class Loop {
      * @param {string} text - User message.
      * @param {ToolDef[]} [tools=[]] - Tool definitions.
      * @param {Record<string, any>} [options={}] - Per-run overrides.
-     * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean}>}
+     * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean, thinkingApplied?: boolean|null}>}
      */
     chat(text: string, tools?: ToolDef[], options?: Record<string, any>): Promise<{
         text: string;
@@ -249,6 +250,7 @@ export class Loop {
         metrics: RunMetrics;
         malformedToolCall?: GenerateResult["malformedToolCall"];
         temperatureDropped?: boolean;
+        thinkingApplied?: boolean | null;
     }>;
     /**
      * Request a clean stop. The current tool call finishes; the loop then exits at the next boundary and

package/src/loop.js

@@ -455,7 +455,7 @@ class Loop {
    *   thunk is re-evaluated each round (D4/eval-assist F2) so a tool set that grows mid-run — e.g. a skill
    *   unlocking its tools — is offered on the next round; a static array is resolved once at wire time.
    * @param {Record<string, any>} [options={}] - Per-run overrides (system, temperature, ctx, etc.).
-   * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean}>}
+   * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean, thinkingApplied?: boolean|null}>}
    *   On halt the returned `error` is `halt:<rule>` (or `halt:unknown` if the
    *   thrown HaltError carried no `rule`), and `msgs` is sanitized so any
    *   dangling assistant `tool_calls` from the halted round are paired with
@@ -565,6 +565,10 @@ class Loop {
     // unsupported/deprecated) and retried without it. Surfaced on the result so an upstream receipt
     // (recurse's refineLeaf) can report the EFFECTIVE temperature rather than the ignored request.
     let temperatureDropped = false;
+    // Normalized thinkingMode (src/thinking-mode.js): the run's `thinkingApplied` verdict — sticky-false
+    // (any round false ⇒ false), else the latest round's true/null; undefined ⇒ the provider reported none (thinkingMode unset) ⇒ key absent.
+    /** @type {boolean|null|undefined} */
+    let lastThinkingApplied;
     // BA-11: consecutive policy-deny counter (reset by any tool call that PASSES policy). When it reaches
     // this.maxConsecutiveDenials the run short-circuits cleanly — see the deny block below.
     let consecutiveDenials = 0;
@@ -597,7 +601,7 @@ class Loop {
         sealDanglingToolCalls(msgs, `[halted:${stuckTag}]`);
         this._reportError('stuck', new Error(`tool "${tc.name}" failed ${identicalErrors} times with identical arguments`), { rule: stuckTag, attempts: identicalErrors });
         this._safeEmit({ type: 'loop:done', data: { text: lastText, stuck: true, rule: stuckTag, cost: totalCost } });
-        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: stuckTag, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: stuckTag, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
       }
       return null;
     };
@@ -844,7 +848,7 @@ class Loop {
         this._reportError('provider', err, { round });
         if (this.throwOnError) throw err;
         // BA-5: a mid-run provider failure must not erase the work of the rounds that succeeded.
-        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: err.message, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: err.message, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
       }
 
       lastUsage = result.usage || lastUsage;
@@ -861,6 +865,11 @@ class Loop {
       lastMalformedToolCall = (result.malformedToolCall && typeof result.malformedToolCall === 'object')
         ? result.malformedToolCall : null;
       if (result.temperatureDropped) temperatureDropped = true;
+      // Sticky-false: once ANY round reports false (thinking happened despite 'off'), the run stays false;
+      // otherwise the latest round's true/null wins; a round without the field leaves it untouched.
+      if (result.thinkingApplied === true || result.thinkingApplied === false || result.thinkingApplied === null) {
+        if (lastThinkingApplied !== false) lastThinkingApplied = result.thinkingApplied;
+      }
       // Publish the latest measured usage to ctx (non-enumerable, fail-open) so a transcript-bound seam —
       // e.g. F2 stash auto-compaction — can read EXACT provider-counted `inputTokens` to gauge context
       // pressure on the NEXT round's trim. Symmetric with lending ctx.summarize; the Loop stays unaware of
@@ -944,7 +953,7 @@ class Loop {
         sealDanglingToolCalls(msgs, `[halted:${session.error}]`);
         this._reportError('session', new Error(`provider session terminated: ${session.error}`), { rule: session.error, sessionTurns: session.turns ?? null });
         this._safeEmit({ type: 'loop:done', data: { text: lastText, rule: session.error, sessionTurns: session.turns ?? null, cost: totalCost } });
-        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: session.error, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: session.error, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
       }
 
       // BA-13: classify this round's terminal signal against the neutral stop-reason vocabulary. BA-6
@@ -1009,7 +1018,7 @@ class Loop {
           msgs.push({ role: 'assistant', content: result.text });
         }
         this._safeEmit({ type: 'loop:done', data: { text: lastText, ...(terminal === 'truncated' && { truncated: true }), terminal, stopReason: lastStopReason, droppedToolCalls: dropped, cost: totalCost } });
-        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: errorTag, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: errorTag, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
       }
 
       // No tool calls — LLM gave a final text response
@@ -1029,7 +1038,7 @@ class Loop {
           try { await flush(msgs, ctx); }
           catch (err) { if (err instanceof HaltError) throw err; this._reportError('trim-flush', err, { round }); }
         }
-        return { text: result.text, toolCalls: [], usage: lastUsage, cost: totalCost, error: null, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+        return { text: result.text, toolCalls: [], usage: lastUsage, cost: totalCost, error: null, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
       }
 
       // Execute tool calls
@@ -1138,7 +1147,7 @@ class Loop {
               sealDanglingToolCalls(msgs, `[halted:${denyTag}]`);
               this._reportError('denied', new Error(`policy denied ${consecutiveDenials} consecutive tool calls (${tc.name})`), { rule: denyTag, denials: consecutiveDenials });
               this._safeEmit({ type: 'loop:done', data: { text: lastText, denied: true, rule: denyTag, cost: totalCost } });
-              return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: denyTag, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+              return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: denyTag, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
             }
             continue;
           }
@@ -1212,7 +1221,7 @@ class Loop {
         // BA-5: the rule tag survives on `error`; so does the work. A halt is how a bounded attempt is
         // SUPPOSED to end — the caller reads `error` to know it was bounded and `text` to learn from it.
         this._safeEmit({ type: 'loop:done', data: { text: lastText, halted: true, rule, cost: totalCost } });
-        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: `halt:${rule}`, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+        return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: `halt:${rule}`, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
       }
       throw err;
     }
@@ -1242,20 +1251,20 @@ class Loop {
             const rule = err.rule || 'unknown';
             this._reportError('halt', err, { rule, reason: err.decision?.reason ?? null });
             this._safeEmit({ type: 'loop:done', data: { text: lastText, halted: true, rule, cost: totalCost } });
-            return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: `halt:${rule}`, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+            return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: `halt:${rule}`, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
           }
           this._reportError('trim-flush', err, { phase: 'stop' });
         }
       }
       this._safeEmit({ type: 'loop:done', data: { text: lastText, stopped: true, cost: totalCost } });
-      return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: null, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+      return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: null, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
     }
 
     // Hard safety limit — should never fire under normal usage; bareguard's
     // limits.maxTurns (or the LLM's natural completion) ends the loop first.
     const warning = `[Loop] hit internal safety limit of ${HARD_ROUND_LIMIT} rounds. Wire bareguard for proper governance — see bare-agent/bareguard.`;
     this._safeEmit({ type: 'loop:done', data: { text: lastText, warning, cost: totalCost } });
-    return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: warning, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }) };
+    return { text: lastText, toolCalls: [], usage: lastUsage, cost: totalCost, error: warning, stopReason: lastStopReason, model: lastModel, msgs, metrics: finalizeMetrics(), ...(lastMalformedToolCall && { malformedToolCall: lastMalformedToolCall }), ...(temperatureDropped && { temperatureDropped: true }), ...(lastThinkingApplied !== undefined && { thinkingApplied: lastThinkingApplied }) };
   }
 
   /**
@@ -1328,7 +1337,7 @@ class Loop {
    * @param {string} text - User message.
    * @param {ToolDef[]} [tools=[]] - Tool definitions.
    * @param {Record<string, any>} [options={}] - Per-run overrides.
-   * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean}>}
+   * @returns {Promise<{text: string, toolCalls: ToolCall[], usage: Usage, cost: number, error: string|null, stopReason: string|null, model: string|null, msgs: Message[], metrics: RunMetrics, malformedToolCall?: GenerateResult['malformedToolCall'], temperatureDropped?: boolean, thinkingApplied?: boolean|null}>}
    */
   async chat(text, tools = [], options = {}) {
     this._history.push({ role: 'user', content: text });

package/src/provider-anthropic.d.ts

@@ -30,6 +30,11 @@ export type AnthropicOptions = {
      * **MEASURED CAVEAT — this option does not "turn thinking on".** On `claude-sonnet-5` adaptive thinking is ALREADY the default: sending this changed the observed thinking rate not at all (2/10 rounds with it vs 3/10 without — `poc/ba7-adaptive-default.mjs`). Its real use is pinning the mode and reaching `display`/`effort`. The change that mattered is that thinking blocks are now PRESERVED and replayed (see `Message.providerBlocks`), which happens whether or not you ever set this.
      */
     thinking?: any;
+    /**
+     * - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right Anthropic `thinking` request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+     * Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
+     */
+    thinkingMode?: "on" | "off" | null | undefined;
     /**
      * - Attach the full upstream response to `err.body` on HTTP errors (off by default to avoid leaking unexpected fields through error logs; `err.message` still carries the API error).
      */
@@ -58,6 +63,8 @@ export type AnthropicOptions = {
  *   chain set `thinking` on each provider's CONSTRUCTOR instead of per-call.
  *
  *   **MEASURED CAVEAT — this option does not "turn thinking on".** On `claude-sonnet-5` adaptive thinking is ALREADY the default: sending this changed the observed thinking rate not at all (2/10 rounds with it vs 3/10 without — `poc/ba7-adaptive-default.mjs`). Its real use is pinning the mode and reaching `display`/`effort`. The change that mattered is that thinking blocks are now PRESERVED and replayed (see `Message.providerBlocks`), which happens whether or not you ever set this.
+ * @property {'on'|'off'|null} [thinkingMode] - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right Anthropic `thinking` request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+ *   Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
  * @property {boolean} [exposeErrorBody=false] - Attach the full upstream response to `err.body` on HTTP errors (off by default to avoid leaking unexpected fields through error logs; `err.message` still carries the API error).
  * @property {number} [timeoutMs=600000] - BA-18: request/idle timeout in ms. A silent or never-answering socket (dropped by the server, or a response that never starts) was otherwise bounded only by the OS TCP timeout (~2h) — a hang, not an error, so every retry policy above it was inert. Bounds on socket INACTIVITY (the timer resets on activity, so a slow-but-streaming response is not killed); on trip, `generate()` rejects with a retryable `TimeoutError` (`code: 'ETIMEDOUT'`, `context.bound: 'idle'`) that a wired `Retry` (`Loop({ retry })`) retries. Default 10 min sits above any single non-streaming completion; `0` or `Infinity` disables it (pre-BA-18 behaviour). Overridable per call via `generate(..., { timeoutMs })`.
  * @property {number} [deadlineMs=0] - BA-19: TOTAL call-duration deadline in ms, beside `timeoutMs`. The idle bound resets on any socket activity, so a response that trickles a byte forever (a "zombie stream") never trips it and hangs the caller for hours. This is an absolute, non-resetting wall-clock ceiling; on trip, `generate()` rejects with a TERMINAL `TimeoutError` (`code: 'EDEADLINE'`, `context.bound: 'deadline'`, `retryable: false` — a hard ceiling meant to STOP, not re-spend). DISABLED by default (a deliberately long single call is legitimate); `0`/`Infinity` disable. When both are set with `timeoutMs < deadlineMs`, a silent socket trips the idle bound first. Overridable per call via `generate(..., { deadlineMs })`.
@@ -78,6 +85,9 @@ export class AnthropicProvider {
     cacheSystem: boolean;
     cacheMessages: boolean;
     thinking: any;
+    thinkingMode: "on" | "off" | null;
+    /** @type {Map<string, string>} shape that worked per mode+model (no re-probing) */
+    _thinkingShapes: Map<string, string>;
     exposeErrorBody: boolean;
     timeoutMs: number | undefined;
     deadlineMs: number | undefined;
@@ -85,7 +95,7 @@ export class AnthropicProvider {
      * Generate a response from the Anthropic API.
      * @param {Message[]} messages - Conversation messages (OpenAI format, auto-converted).
      * @param {ToolDef[]} [tools=[]] - Tool definitions.
-     * @param {Record<string, any>} [options={}] - Options (temperature, maxTokens, system, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
+     * @param {Record<string, any>} [options={}] - Options (thinkingMode — per-call override of the constructor's, `null` suppresses; temperature, maxTokens, system, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
      * @returns {Promise<GenerateResult>}
      * @throws {Error} `[AnthropicProvider] ...` — on HTTP errors (4xx/5xx) or invalid JSON response.
      */

package/src/provider-anthropic.js

@@ -7,6 +7,7 @@ const { requestWithTemperatureFallback } = require('./provider-temperature');
 const { normalizeStopReason } = require('./provider-stop-reason');
 const { resolveTimeoutMs, applyRequestBounds, guardResponseSettles } = require('./provider-http');
 const { hasUsageSignal } = require('./provider-usage');
+const { normalizeThinkingMode, resolveThinkingMode, requestWithThinkingMode, assessThinkingApplied } = require('./thinking-mode');
 
 // BA-24: the raw Anthropic usage field names. Presence of any (even value 0) means the API reported a
 // usage signal → build the object; absence of all means no signal → surface null (unpriceable).
@@ -34,6 +35,8 @@ function isLoopbackHost(hostname) {
  *   chain set `thinking` on each provider's CONSTRUCTOR instead of per-call.
  *
  *   **MEASURED CAVEAT — this option does not "turn thinking on".** On `claude-sonnet-5` adaptive thinking is ALREADY the default: sending this changed the observed thinking rate not at all (2/10 rounds with it vs 3/10 without — `poc/ba7-adaptive-default.mjs`). Its real use is pinning the mode and reaching `display`/`effort`. The change that mattered is that thinking blocks are now PRESERVED and replayed (see `Message.providerBlocks`), which happens whether or not you ever set this.
+ * @property {'on'|'off'|null} [thinkingMode] - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right Anthropic `thinking` request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+ *   Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
  * @property {boolean} [exposeErrorBody=false] - Attach the full upstream response to `err.body` on HTTP errors (off by default to avoid leaking unexpected fields through error logs; `err.message` still carries the API error).
  * @property {number} [timeoutMs=600000] - BA-18: request/idle timeout in ms. A silent or never-answering socket (dropped by the server, or a response that never starts) was otherwise bounded only by the OS TCP timeout (~2h) — a hang, not an error, so every retry policy above it was inert. Bounds on socket INACTIVITY (the timer resets on activity, so a slow-but-streaming response is not killed); on trip, `generate()` rejects with a retryable `TimeoutError` (`code: 'ETIMEDOUT'`, `context.bound: 'idle'`) that a wired `Retry` (`Loop({ retry })`) retries. Default 10 min sits above any single non-streaming completion; `0` or `Infinity` disables it (pre-BA-18 behaviour). Overridable per call via `generate(..., { timeoutMs })`.
  * @property {number} [deadlineMs=0] - BA-19: TOTAL call-duration deadline in ms, beside `timeoutMs`. The idle bound resets on any socket activity, so a response that trickles a byte forever (a "zombie stream") never trips it and hangs the caller for hours. This is an absolute, non-resetting wall-clock ceiling; on trip, `generate()` rejects with a TERMINAL `TimeoutError` (`code: 'EDEADLINE'`, `context.bound: 'deadline'`, `retryable: false` — a hard ceiling meant to STOP, not re-spend). DISABLED by default (a deliberately long single call is legitimate); `0`/`Infinity` disable. When both are set with `timeoutMs < deadlineMs`, a silent socket trips the idle bound first. Overridable per call via `generate(..., { deadlineMs })`.
@@ -64,6 +67,12 @@ class AnthropicProvider {
     // release every time Anthropic moves; passing it through means the caller can always express
     // the current API. Omit it to keep today's body byte-identical.
     this.thinking = options.thinking != null ? options.thinking : null;
+    // Normalized on/off switch (src/thinking-mode.js). Validated here; conflict with raw `thinking` is a
+    // ValidationError (no silent precedence). Unset ⇒ null ⇒ the request path is byte-identical.
+    this.thinkingMode = normalizeThinkingMode(options.thinkingMode, 'AnthropicProvider');
+    resolveThinkingMode({ label: 'AnthropicProvider', instanceMode: this.thinkingMode, options: {}, instanceThinking: this.thinking, hasRawThinking: true });
+    /** @type {Map<string, string>} shape that worked per mode+model (no re-probing) */
+    this._thinkingShapes = new Map();
     // See OpenAIProvider: attach full upstream body to err.body only on opt-in.
     this.exposeErrorBody = options.exposeErrorBody === true;
     // BA-18: request/idle timeout (ms). Resolved at call time (default 600000; 0/Infinity disable).
@@ -76,7 +85,7 @@ class AnthropicProvider {
    * Generate a response from the Anthropic API.
    * @param {Message[]} messages - Conversation messages (OpenAI format, auto-converted).
    * @param {ToolDef[]} [tools=[]] - Tool definitions.
-   * @param {Record<string, any>} [options={}] - Options (temperature, maxTokens, system, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
+   * @param {Record<string, any>} [options={}] - Options (thinkingMode — per-call override of the constructor's, `null` suppresses; temperature, maxTokens, system, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
    * @returns {Promise<GenerateResult>}
    * @throws {Error} `[AnthropicProvider] ...` — on HTTP errors (4xx/5xx) or invalid JSON response.
    */
@@ -149,6 +158,7 @@ class AnthropicProvider {
     // above, which applies whether or not you ever set this.
     const thinking = options.thinking !== undefined ? options.thinking : this.thinking;
     if (thinking) body.thinking = thinking;
+    const thinkingMode = resolveThinkingMode({ label: 'AnthropicProvider', instanceMode: this.thinkingMode, options, instanceThinking: this.thinking, hasRawThinking: true });
     if (tools.length > 0) {
       body.tools = tools.map(t => ({
         name: t.name,
@@ -163,9 +173,12 @@ class AnthropicProvider {
     const timeoutMs = resolveTimeoutMs(this.timeoutMs, options.timeoutMs);
     const deadlineMs = resolveTimeoutMs(this.deadlineMs, options.deadlineMs, 0, 'deadlineMs');
     const { data, temperatureDropped } = await requestWithTemperatureFallback({
-      request: () => this._request(body, timeoutMs, deadlineMs),
+      request: () => thinkingMode
+        ? requestWithThinkingMode({ kind: 'anthropic', mode: thinkingMode, model: this.model, label: 'AnthropicProvider', body, cache: this._thinkingShapes, request: (b) => this._request(b, timeoutMs, deadlineMs) })
+        : this._request(body, timeoutMs, deadlineMs),
       hadTemperature: () => body.temperature != null,
       stripTemperature: () => { delete body.temperature; },
+      thinkingOn: thinkingMode === 'on',
       warnOnce: () => this._warnTemperatureDropped(),
     });
 
@@ -213,6 +226,7 @@ class AnthropicProvider {
           }
         : null,
       ...(temperatureDropped && { temperatureDropped: true }),
+      ...(thinkingMode && { thinkingApplied: assessThinkingApplied({ kind: 'anthropic', mode: thinkingMode, data, owner: this, label: 'AnthropicProvider', model: this.model }) }),
     };
   }
 

package/src/provider-clipipe.d.ts

@@ -19,6 +19,10 @@ export type CLIPipeOptions = {
      * - Environment variables for the child process.
      */
     env?: Record<string, string> | undefined;
+    /**
+     * - NOT supported: any non-null value throws `ValidationError` (CLIPipe has no thinking switch). Declared so the loud failure is typed, not silent.
+     */
+    thinkingMode?: any;
     /**
      * - Timeout in milliseconds.
      */
@@ -172,7 +176,7 @@ export class CLIPipeProvider {
      * failure for a genuinely tool-incapable model lives in the tool-mode capability probe.
      * @param {Message[]} messages - Conversation messages in OpenAI format.
      * @param {ToolDef[]} [tools=[]] - Caller tools. Honored only in tool mode (`toolProtocol` set).
-     * @param {Record<string, any>} [options={}] - Unused.
+     * @param {Record<string, any>} [options={}] - Unused, except `thinkingMode` (any non-null value throws `ValidationError`: CLIPipe has no thinking switch).
      * @returns {Promise<GenerateResult>}
      * @throws {Error} `[CLIPipeProvider] failed to spawn "cmd": ...` — when the command cannot be found or executed.
      * @throws {Error} `[CLIPipeProvider] process exited with code N: ...` — on non-zero exit.

package/src/provider-clipipe.js

@@ -1,7 +1,7 @@
 'use strict';
 
 const { spawn } = require('child_process');
-const { ProviderError, HaltError } = require('./errors');
+const { ProviderError, HaltError, ValidationError } = require('./errors');
 const { buildToolSystemPrompt, renderTranscript, resolveToolProtocol, mapClaudeMeta } = require('./provider-clipipe-tools');
 const { createBridge, resolveSessionError, runSession } = require('./provider-clipipe-mcp');
 
@@ -21,6 +21,7 @@ const { createBridge, resolveSessionError, runSession } = require('./provider-cl
  * @property {string[]} [args=[]] - Arguments to pass to the command.
  * @property {string} [cwd] - Working directory for the child process.
  * @property {Record<string, string>} [env] - Environment variables for the child process.
+ * @property {any} [thinkingMode] - NOT supported: any non-null value throws `ValidationError` (CLIPipe has no thinking switch). Declared so the loud failure is typed, not silent.
  * @property {number} [timeout=30000] - Timeout in milliseconds.
  * @property {string} [systemPromptFlag] - CLI flag for system prompt (e.g. '--system'). When set, system messages are extracted and passed via this flag instead of stdin.
  * @property {(chunk: string) => void} [onChunk] - Called with each stdout chunk as it streams.
@@ -107,6 +108,10 @@ function subtractUsage(total, streamed) {
   return out;
 }
 
+const CLIPIPE_NO_THINKING_MODE = () => new ValidationError(
+  '[CLIPipeProvider] thinkingMode is not supported — a CLI subprocess has no thinking switch bare-agent can set. Remove thinkingMode (see thinkingSupport(\'clipipe\')).',
+  { context: { lib: 'bare-agent', provider: 'clipipe' } });
+
 class CLIPipeProvider {
   /**
    * Provider that pipes prompts to a CLI command via stdin and reads stdout.
@@ -119,6 +124,8 @@ class CLIPipeProvider {
    */
   constructor(options = {}) {
     if (!options.command) throw new Error('[CLIPipeProvider] requires command');
+    // thinkingMode has NO mechanism here — loud, never silently ignored.
+    if (options.thinkingMode != null) throw CLIPIPE_NO_THINKING_MODE();
     this.command = options.command;
     this.args = options.args || [];
     this.cwd = options.cwd || undefined;
@@ -177,7 +184,7 @@ class CLIPipeProvider {
    * failure for a genuinely tool-incapable model lives in the tool-mode capability probe.
    * @param {Message[]} messages - Conversation messages in OpenAI format.
    * @param {ToolDef[]} [tools=[]] - Caller tools. Honored only in tool mode (`toolProtocol` set).
-   * @param {Record<string, any>} [options={}] - Unused.
+   * @param {Record<string, any>} [options={}] - Unused, except `thinkingMode` (any non-null value throws `ValidationError`: CLIPipe has no thinking switch).
    * @returns {Promise<GenerateResult>}
    * @throws {Error} `[CLIPipeProvider] failed to spawn "cmd": ...` — when the command cannot be found or executed.
    * @throws {Error} `[CLIPipeProvider] process exited with code N: ...` — on non-zero exit.
@@ -185,6 +192,7 @@ class CLIPipeProvider {
    * @throws {Error} `[CLIPipeProvider] process produced no output` — when stdout is empty.
    */
   async generate(messages, tools = [], options = {}) {
+    if (options && options.thinkingMode != null) throw CLIPIPE_NO_THINKING_MODE();
     if (Array.isArray(tools) && tools.length > 0) {
       if (this.nativeTools) return this._generateWithMcp(messages, tools, options);
       if (this.toolProtocol) return this._generateWithTools(messages, tools);

package/src/provider-fallback.d.ts

@@ -25,6 +25,9 @@ export type FallbackOptions = {
  * NOTE: per-call `generate()` options (incl. `thinking`, `toolChoice`) are forwarded UNCHANGED to EVERY provider in the chain. For a
  * mixed-vendor chain (e.g. OpenAI-compat DeepSeek + Anthropic) set `thinking` on each provider's CONSTRUCTOR instead of per-call,
  * or a value shaped for one vendor lands in another's `body.thinking` on fallback.
+ * The normalized `thinkingMode: 'on'|'off'` is the exception: reaching every provider is CORRECT for it (each maps it to its own
+ * request shape); a provider that cannot honor it (CLIPipe, a measured non-switchable model) throws `ValidationError`, which the chain
+ * treats like any failure.
  */
 export class FallbackProvider {
     /**

package/src/provider-fallback.js

@@ -15,6 +15,9 @@
  * NOTE: per-call `generate()` options (incl. `thinking`, `toolChoice`) are forwarded UNCHANGED to EVERY provider in the chain. For a
  * mixed-vendor chain (e.g. OpenAI-compat DeepSeek + Anthropic) set `thinking` on each provider's CONSTRUCTOR instead of per-call,
  * or a value shaped for one vendor lands in another's `body.thinking` on fallback.
+ * The normalized `thinkingMode: 'on'|'off'` is the exception: reaching every provider is CORRECT for it (each maps it to its own
+ * request shape); a provider that cannot honor it (CLIPipe, a measured non-switchable model) throws `ValidationError`, which the chain
+ * treats like any failure.
  */
 class FallbackProvider {
   /**

package/src/provider-gemini.d.ts

@@ -15,6 +15,11 @@ export type GeminiOptions = {
      * - API base (override for proxies; posts to `${baseUrl}/models/${model}:generateContent`).
      */
     baseUrl?: string | undefined;
+    /**
+     * - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right `generationConfig.thinkingConfig.thinkingBudget` request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+     * Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
+     */
+    thinkingMode?: "on" | "off" | null | undefined;
     /**
      * - Attach the full upstream response to `err.body` on HTTP errors (off by default; `err.message` still carries the API error).
      */
@@ -33,6 +38,8 @@ export type GeminiOptions = {
  * @property {string} [apiKey] - Google AI Studio (Gemini) API key.
  * @property {string} [model='gemini-2.5-flash'] - Model ID.
  * @property {string} [baseUrl='https://generativelanguage.googleapis.com/v1beta'] - API base (override for proxies; posts to `${baseUrl}/models/${model}:generateContent`).
+ * @property {'on'|'off'|null} [thinkingMode] - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right `generationConfig.thinkingConfig.thinkingBudget` request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+ *   Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
  * @property {boolean} [exposeErrorBody=false] - Attach the full upstream response to `err.body` on HTTP errors (off by default; `err.message` still carries the API error).
  * @property {number} [timeoutMs=600000] - BA-18: request/idle timeout in ms. Bounds a silent or never-answering socket on inactivity so `generate()` rejects with a retryable `TimeoutError` (`code: 'ETIMEDOUT'`, `context.bound: 'idle'`) instead of hanging until the OS TCP timeout (~2h). `0`/`Infinity` disables it (pre-BA-18 behaviour). Overridable per call via `generate(..., { timeoutMs })`.
  * @property {number} [deadlineMs=0] - BA-19: TOTAL call-duration deadline in ms, beside `timeoutMs`. The idle bound resets on any socket activity, so a response that trickles a byte forever never trips it and hangs for hours. This is an absolute, non-resetting ceiling; on trip, `generate()` rejects with a TERMINAL `TimeoutError` (`code: 'EDEADLINE'`, `context.bound: 'deadline'`, `retryable: false`). DISABLED by default; `0`/`Infinity` disable. Overridable per call via `generate(..., { deadlineMs })`.
@@ -57,13 +64,16 @@ export class GeminiProvider {
     model: string;
     baseUrl: string;
     exposeErrorBody: boolean;
+    thinkingMode: "on" | "off" | null;
+    /** @type {Map<string, string>} shape that worked per mode+model (no re-probing) */
+    _thinkingShapes: Map<string, string>;
     timeoutMs: number | undefined;
     deadlineMs: number | undefined;
     /**
      * Generate a response from the Gemini API.
      * @param {Message[]} messages - Conversation messages (OpenAI format, auto-converted).
      * @param {ToolDef[]} [tools=[]] - Tool definitions.
-     * @param {Record<string, any>} [options={}] - Options (temperature, maxTokens, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
+     * @param {Record<string, any>} [options={}] - Options (thinkingMode — per-call override of the constructor's, `null` suppresses; temperature, maxTokens, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
      * @returns {Promise<GenerateResult>}
      * @throws {Error} `[GeminiProvider] ...` — on HTTP errors (4xx/5xx) or invalid JSON response.
      */

package/src/provider-gemini.js

@@ -7,6 +7,7 @@ const { requestWithTemperatureFallback } = require('./provider-temperature');
 const { normalizeStopReason } = require('./provider-stop-reason');
 const { resolveTimeoutMs, applyRequestBounds, guardResponseSettles } = require('./provider-http');
 const { hasUsageSignal } = require('./provider-usage');
+const { normalizeThinkingMode, resolveThinkingMode, requestWithThinkingMode, assessThinkingApplied } = require('./thinking-mode');
 
 // BA-24: raw Gemini usageMetadata fields. Any present (even 0) ⇒ a usage signal; none ⇒ null.
 const GEMINI_USAGE_KEYS = ['promptTokenCount', 'candidatesTokenCount', 'thoughtsTokenCount', 'cachedContentTokenCount'];
@@ -27,6 +28,8 @@ function isLoopbackHost(hostname) {
  * @property {string} [apiKey] - Google AI Studio (Gemini) API key.
  * @property {string} [model='gemini-2.5-flash'] - Model ID.
  * @property {string} [baseUrl='https://generativelanguage.googleapis.com/v1beta'] - API base (override for proxies; posts to `${baseUrl}/models/${model}:generateContent`).
+ * @property {'on'|'off'|null} [thinkingMode] - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right `generationConfig.thinkingConfig.thinkingBudget` request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+ *   Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
  * @property {boolean} [exposeErrorBody=false] - Attach the full upstream response to `err.body` on HTTP errors (off by default; `err.message` still carries the API error).
  * @property {number} [timeoutMs=600000] - BA-18: request/idle timeout in ms. Bounds a silent or never-answering socket on inactivity so `generate()` rejects with a retryable `TimeoutError` (`code: 'ETIMEDOUT'`, `context.bound: 'idle'`) instead of hanging until the OS TCP timeout (~2h). `0`/`Infinity` disables it (pre-BA-18 behaviour). Overridable per call via `generate(..., { timeoutMs })`.
  * @property {number} [deadlineMs=0] - BA-19: TOTAL call-duration deadline in ms, beside `timeoutMs`. The idle bound resets on any socket activity, so a response that trickles a byte forever never trips it and hangs for hours. This is an absolute, non-resetting ceiling; on trip, `generate()` rejects with a TERMINAL `TimeoutError` (`code: 'EDEADLINE'`, `context.bound: 'deadline'`, `retryable: false`). DISABLED by default; `0`/`Infinity` disable. Overridable per call via `generate(..., { deadlineMs })`.
@@ -52,6 +55,10 @@ class GeminiProvider {
     this.model = options.model || 'gemini-2.5-flash';
     this.baseUrl = options.baseUrl || 'https://generativelanguage.googleapis.com/v1beta';
     this.exposeErrorBody = options.exposeErrorBody === true;
+    // Normalized on/off switch (src/thinking-mode.js). Gemini has no raw `thinking` option to conflict with.
+    this.thinkingMode = normalizeThinkingMode(options.thinkingMode, 'GeminiProvider');
+    /** @type {Map<string, string>} shape that worked per mode+model (no re-probing) */
+    this._thinkingShapes = new Map();
     // BA-18: request/idle timeout (ms). Resolved at call time (default 600000; 0/Infinity disable).
     this.timeoutMs = options.timeoutMs;
     // BA-19: total call-duration deadline (ms). Resolved at call time (default 0 = disabled).
@@ -62,7 +69,7 @@ class GeminiProvider {
    * Generate a response from the Gemini API.
    * @param {Message[]} messages - Conversation messages (OpenAI format, auto-converted).
    * @param {ToolDef[]} [tools=[]] - Tool definitions.
-   * @param {Record<string, any>} [options={}] - Options (temperature, maxTokens, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
+   * @param {Record<string, any>} [options={}] - Options (thinkingMode — per-call override of the constructor's, `null` suppresses; temperature, maxTokens, timeoutMs — a per-call override of the constructor's `timeoutMs`, see BA-18; deadlineMs — a per-call override of the constructor's `deadlineMs`, see BA-19).
    * @returns {Promise<GenerateResult>}
    * @throws {Error} `[GeminiProvider] ...` — on HTTP errors (4xx/5xx) or invalid JSON response.
    */
@@ -124,10 +131,13 @@ class GeminiProvider {
 
     // BA-10: graceful degrade if a model rejects a non-default `temperature` (Gemini nests it under
     // generationConfig). Keyed off the API error text, so dormant on models that accept temperature.
+    const thinkingMode = resolveThinkingMode({ label: 'GeminiProvider', instanceMode: this.thinkingMode, options });
     const timeoutMs = resolveTimeoutMs(this.timeoutMs, options.timeoutMs);
     const deadlineMs = resolveTimeoutMs(this.deadlineMs, options.deadlineMs, 0, 'deadlineMs');
     const { data, temperatureDropped } = await requestWithTemperatureFallback({
-      request: () => this._request(`/models/${this.model}:generateContent`, body, timeoutMs, deadlineMs),
+      request: () => thinkingMode
+        ? requestWithThinkingMode({ kind: 'gemini', mode: thinkingMode, model: this.model, label: 'GeminiProvider', body, cache: this._thinkingShapes, request: (b) => this._request(`/models/${this.model}:generateContent`, b, timeoutMs, deadlineMs) })
+        : this._request(`/models/${this.model}:generateContent`, body, timeoutMs, deadlineMs),
       hadTemperature: () => body.generationConfig?.temperature != null,
       stripTemperature: () => { if (body.generationConfig) delete body.generationConfig.temperature; },
       warnOnce: () => this._warnTemperatureDropped(),
@@ -161,6 +171,7 @@ class GeminiProvider {
       stopReason: normalizeStopReason(data.candidates?.[0]?.finishReason, 'gemini', { hasToolCalls: toolCalls.length > 0 }),
       usage: this._normalizeUsage(data.usageMetadata),
       ...(temperatureDropped && { temperatureDropped: true }),
+      ...(thinkingMode && { thinkingApplied: assessThinkingApplied({ kind: 'gemini', mode: thinkingMode, data, owner: this, label: 'GeminiProvider', model: this.model }) }),
     };
   }
 

package/src/provider-ollama.d.ts

@@ -4,6 +4,11 @@ export type GenerateResult = import("../types").GenerateResult;
 export type OllamaOptions = {
     model?: string | undefined;
     url?: string | undefined;
+    /**
+     * - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right `think` (`/api/chat`; NOT live-verified against a real Ollama thinking model — reply signal `message.thinking`) request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+     * Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
+     */
+    thinkingMode?: "on" | "off" | null | undefined;
     exposeErrorBody?: boolean | undefined;
     /**
      * - BA-18: request/idle timeout in ms. Bounds a silent or never-answering socket on inactivity so `generate()` rejects with a retryable `TimeoutError` (`code: 'ETIMEDOUT'`, `context.bound: 'idle'`) instead of hanging until the OS TCP timeout. `0`/`Infinity` disables it. Overridable per call via `generate(..., { timeoutMs })`. (A local Ollama that is loading a large model cold can be slow to first byte — raise this or disable it for very large local models.)
@@ -28,6 +33,8 @@ export type OllamaOptions = {
  * @typedef {object} OllamaOptions
  * @property {string} [model='llama3.2']
  * @property {string} [url='http://localhost:11434']
+ * @property {'on'|'off'|null} [thinkingMode] - Normalized thinking switch: `'on'` or `'off'`, and bare-agent sends the right `think` (`/api/chat`; NOT live-verified against a real Ollama thinking model — reply signal `message.thinking`) request field for the provider/model, retries the known alternative shapes if the API rejects the field by name, and THROWS a `ValidationError` (naming what it tried) if every shape is rejected — or before any request when the model is measured as not switchable. Every reply is checked: `GenerateResult.thinkingApplied` is `true`/`false`/`null` (`false` = 'off' was sent but the reply shows thinking happened — warns once per instance, result kept). Unset ⇒ request body byte-identical to before. Overridable per call via `generate(..., { thinkingMode })`; `null` suppresses an instance default. Any value but 'on'/'off' ⇒ `ValidationError`. A NEW field: it does not change the raw `thinking` option; setting both (after per-call override) ⇒ `ValidationError`. See `src/thinking-mode.js` / `thinkingSupport()`.
+ *   Per-call `thinkingMode` reaches EVERY provider in a `FallbackProvider` chain — correct for a normalized field (each provider maps it to its own shape; one that cannot — CLIPipe, a non-switchable model — throws `ValidationError`, which the chain treats like any failure and falls through).
  * @property {boolean} [exposeErrorBody=false]
  * @property {number} [timeoutMs=600000] - BA-18: request/idle timeout in ms. Bounds a silent or never-answering socket on inactivity so `generate()` rejects with a retryable `TimeoutError` (`code: 'ETIMEDOUT'`, `context.bound: 'idle'`) instead of hanging until the OS TCP timeout. `0`/`Infinity` disables it. Overridable per call via `generate(..., { timeoutMs })`. (A local Ollama that is loading a large model cold can be slow to first byte — raise this or disable it for very large local models.)
  * @property {number} [deadlineMs=0] - BA-19: TOTAL call-duration deadline in ms, beside `timeoutMs`. The idle bound resets on any socket activity, so a response that trickles a byte forever never trips it and hangs for hours. This is an absolute, non-resetting ceiling; on trip, `generate()` rejects with a TERMINAL `TimeoutError` (`code: 'EDEADLINE'`, `context.bound: 'deadline'`, `retryable: false`). DISABLED by default; `0`/`Infinity` disable. Overridable per call via `generate(..., { deadlineMs })`. (A cold large-model load is a legitimate long single call — leave this disabled or set it generously for such models.)
@@ -48,6 +55,9 @@ export class OllamaProvider {
     model: string;
     url: string;
     exposeErrorBody: boolean;
+    thinkingMode: "on" | "off" | null;
+    /** @type {Map<string, string>} shape that worked per mode+model (no re-probing) */
+    _thinkingShapes: Map<string, string>;
     timeoutMs: number | undefined;
     deadlineMs: number | undefined;
     exposeMalformedArgs: boolean;
@@ -55,7 +65,7 @@ export class OllamaProvider {
      * Generate a response from a local Ollama instance.
      * @param {Message[]} messages - Conversation messages.
      * @param {ToolDef[]} [tools=[]] - Tool definitions.
-     * @param {Record<string, any>} [options={}] - Options (`temperature`, `maxTokens`, `timeoutMs` — a per-call override of the constructor's `timeoutMs`, see BA-18; `deadlineMs` — a per-call override of the constructor's `deadlineMs`, see BA-19).
+     * @param {Record<string, any>} [options={}] - Options (`thinkingMode` — per-call override of the constructor's, `null` suppresses; `temperature`, `maxTokens`, `timeoutMs` — a per-call override of the constructor's `timeoutMs`, see BA-18; `deadlineMs` — a per-call override of the constructor's `deadlineMs`, see BA-19).
      * @returns {Promise<GenerateResult>}
      * @throws {Error} `[OllamaProvider] ...` — on HTTP errors or invalid JSON response.
      */