cognitive-core 0.2.4 → 0.3.1

package/dist/eval/live/azure-openai-llm-provider.js

@@ -0,0 +1,92 @@
+const DEFAULT_BASE_URL = process.env.AZURE_OPENAI_BASE_URL ?? 'https://staging-west-us3.openai.azure.com/openai';
+const DEFAULT_API_VERSION = process.env.AZURE_OPENAI_API_VERSION ?? '2025-04-01-preview';
+/** Pull the assistant text out of a responses-API payload (robust to shape variants). */
+export function extractResponsesText(json) {
+    const j = json;
+    // 1) SDK convenience field, when present.
+    if (typeof j.output_text === 'string' && j.output_text.trim())
+        return j.output_text;
+    // 2) Walk output[].content[]: collect any text parts (skip reasoning items).
+    const out = j.output;
+    if (Array.isArray(out)) {
+        const parts = [];
+        for (const item of out) {
+            const content = item?.content;
+            if (!Array.isArray(content))
+                continue;
+            for (const part of content) {
+                if (typeof part?.text === 'string')
+                    parts.push(part.text);
+            }
+        }
+        if (parts.length)
+            return parts.join('');
+    }
+    return '';
+}
+export class AzureOpenAILLMProvider {
+    name = 'azure-openai';
+    opts;
+    constructor(opts) {
+        this.opts = opts;
+    }
+    async complete(messages, options) {
+        const apiKey = this.opts.apiKey ?? process.env.AZURE_API_KEY;
+        if (!apiKey)
+            throw new Error('Azure OpenAI: AZURE_API_KEY not set (and no apiKey option)');
+        if (!this.opts.model)
+            throw new Error('Azure OpenAI: model (deployment name) is required');
+        const baseUrl = (this.opts.baseUrl ?? DEFAULT_BASE_URL).replace(/\/$/, '');
+        const apiVersion = this.opts.apiVersion ?? DEFAULT_API_VERSION;
+        const url = `${baseUrl}/responses?api-version=${encodeURIComponent(apiVersion)}`;
+        const timeoutMs = this.opts.timeoutMs ?? 120_000;
+        const retries = this.opts.retries ?? 2;
+        const body = {
+            model: this.opts.model,
+            input: messages.map((m) => ({ role: m.role, content: m.content })),
+        };
+        if (options?.temperature != null)
+            body.temperature = options.temperature;
+        const maxOut = this.opts.maxOutputTokens ?? options?.maxTokens;
+        if (maxOut != null)
+            body.max_output_tokens = maxOut;
+        let lastErr;
+        for (let attempt = 0; attempt <= retries; attempt++) {
+            const controller = new AbortController();
+            const timer = setTimeout(() => controller.abort(), timeoutMs);
+            try {
+                const res = await fetch(url, {
+                    method: 'POST',
+                    headers: { 'api-key': apiKey, 'content-type': 'application/json' },
+                    body: JSON.stringify(body),
+                    signal: controller.signal,
+                });
+                if (!res.ok) {
+                    const text = (await res.text()).slice(0, 500);
+                    throw new Error(`Azure OpenAI HTTP ${res.status}: ${text}`);
+                }
+                const json = await res.json();
+                if (json.error) {
+                    throw new Error(`Azure OpenAI error: ${JSON.stringify(json.error).slice(0, 300)}`);
+                }
+                return extractResponsesText(json).trim();
+            }
+            catch (err) {
+                lastErr = err;
+                const msg = controller.signal.aborted ? `timed out after ${timeoutMs}ms` : lastErr.message;
+                // eslint-disable-next-line no-console
+                console.error(`[azure-openai] attempt ${attempt + 1}/${retries + 1} failed: ${msg.slice(0, 300)}`);
+                if (attempt < retries)
+                    await new Promise((r) => setTimeout(r, 5000 * (attempt + 1)));
+            }
+            finally {
+                clearTimeout(timer);
+            }
+        }
+        throw lastErr ?? new Error('Azure OpenAI: request failed');
+    }
+}
+export function createAzureOpenAILLMProvider(opts) {
+    return new AzureOpenAILLMProvider(opts);
+}
+//# sourceMappingURL=azure-openai-llm-provider.js.map

package/dist/eval/live/azure-openai-llm-provider.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"azure-openai-llm-provider.js","sourceRoot":"","sources":["../../../src/eval/live/azure-openai-llm-provider.ts"],"names":[],"mappings":"AAkCA,MAAM,gBAAgB,GACpB,OAAO,CAAC,GAAG,CAAC,qBAAqB,IAAI,kDAAkD,CAAC;AAC1F,MAAM,mBAAmB,GAAG,OAAO,CAAC,GAAG,CAAC,wBAAwB,IAAI,oBAAoB,CAAC;AAEzF,yFAAyF;AACzF,MAAM,UAAU,oBAAoB,CAAC,IAAa;IAChD,MAAM,CAAC,GAAG,IAA+B,CAAC;IAC1C,0CAA0C;IAC1C,IAAI,OAAO,CAAC,CAAC,WAAW,KAAK,QAAQ,IAAI,CAAC,CAAC,WAAW,CAAC,IAAI,EAAE;QAAE,OAAO,CAAC,CAAC,WAAW,CAAC;IACpF,6EAA6E;IAC7E,MAAM,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC;IACrB,IAAI,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;QACvB,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,MAAM,IAAI,IAAI,GAAqC,EAAE,CAAC;YACzD,MAAM,OAAO,GAAG,IAAI,EAAE,OAAO,CAAC;YAC9B,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;gBAAE,SAAS;YACtC,KAAK,MAAM,IAAI,IAAI,OAAyC,EAAE,CAAC;gBAC7D,IAAI,OAAO,IAAI,EAAE,IAAI,KAAK,QAAQ;oBAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC5D,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,MAAM;YAAE,OAAO,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,MAAM,OAAO,sBAAsB;IACxB,IAAI,GAAG,cAAc,CAAC;IACd,IAAI,CAAqB;IAE1C,YAAY,IAAwB;QAClC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,QAAmB,EAAE,OAA2B;QAC7D,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,IAAI,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC;QAC7D,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,4DAA4D,CAAC,CAAC;QAC3F,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAC;QAE3F,MAAM,OAAO,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,gBAAgB,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC3E,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,mBAAmB,CAAC;QAC/D,MAAM,GAAG,GAAG,GAAG,OAAO,0BAA0B,kBAAkB,CAAC,UAAU,CAAC,EAAE,CAAC;QACjF,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,OAAO,CAAC;QACjD,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC;QAEvC,MAAM,IAAI,GAA4B;YACpC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK;YACtB,KAAK,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;SACnE,CAAC;QACF,IAAI,OAAO,EAAE,WAAW,IAAI,IAAI;YAAE,IAAI,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;QACzE,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,eAAe,IAAI,OAAO,EAAE,SAAS,CAAC;QAC/D,IAAI,MAAM,IAAI,IAAI;YAAE,IAAI,CAAC,iBAAiB,GAAG,MAAM,CAAC;QAEpD,IAAI,OAA0B,CAAC;QAC/B,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC;YACpD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;YACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;YAC9D,IAAI,CAAC;gBACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE;oBAC3B,MAAM,EAAE,MAAM;oBACd,OAAO,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,cAAc,EAAE,kBAAkB,EAAE;oBAClE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;oBAC1B,MAAM,EAAE,UAAU,CAAC,MAAM;iBAC1B,CAAC,CAAC;gBACH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;oBACZ,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;oBAC9C,MAAM,IAAI,KAAK,CAAC,qBAAqB,GAAG,CAAC,MAAM,KAAK,IAAI,EAAE,CAAC,CAAC;gBAC9D,CAAC;gBACD,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC;gBAC9B,IAAK,IAAgC,CAAC,KAAK,EAAE,CAAC;oBAC5C,MAAM,IAAI,KAAK,CACb,uBAAuB,IAAI,CAAC,SAAS,CAAE,IAAgC,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAC/F,CAAC;gBACJ,CAAC;gBACD,OAAO,oBAAoB,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;YAC3C,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,OAAO,GAAG,GAAY,CAAC;gBACvB,MAAM,GAAG,GAAG,UAAU,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,mBAAmB,SAAS,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;gBAC3F,sCAAsC;gBACtC,OAAO,CAAC,KAAK,CACX,0BAA0B,OAAO,GAAG,CAAC,IAAI,OAAO,GAAG,CAAC,YAAY,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CACpF,CAAC;gBACF,IAAI,OAAO,GAAG,OAAO;oBAAE,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,IAAI,GAAG,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;YACvF,CAAC;oBAAS,CAAC;gBACT,YAAY,CAAC,KAAK,CAAC,CAAC;YACtB,CAAC;QACH,CAAC;QACD,MAAM,OAAO,IAAI,IAAI,KAAK,CAAC,8BAA8B,CAAC,CAAC;IAC7D,CAAC;CACF;AAED,MAAM,UAAU,4BAA4B,CAAC,IAAwB;IACnE,OAAO,IAAI,sBAAsB,CAAC,IAAI,CAAC,CAAC;AAC1C,CAAC"}

package/dist/eval/live/bridges/skillflow_e2b_bridge.d.ts

@@ -0,0 +1,3 @@
+#!/usr/bin/env node
+export {};
+//# sourceMappingURL=skillflow_e2b_bridge.d.ts.map

package/dist/eval/live/bridges/skillflow_e2b_bridge.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"skillflow_e2b_bridge.d.ts","sourceRoot":"","sources":["../../../../src/eval/live/bridges/skillflow_e2b_bridge.ts"],"names":[],"mappings":""}

package/dist/eval/live/bridges/skillflow_e2b_bridge.js

@@ -0,0 +1,12 @@
+#!/usr/bin/env node
+import { ensureSkillFlowE2BBridgeDependency } from '../skillflow.js';
+try {
+    ensureSkillFlowE2BBridgeDependency('e2b');
+    const { runSkillFlowE2BBridge } = await import('swarmkit-eval');
+    await runSkillFlowE2BBridge();
+}
+catch (err) {
+    process.stdout.write(JSON.stringify({ error: err.message }) + '\n');
+    process.exitCode = 1;
+}
+//# sourceMappingURL=skillflow_e2b_bridge.js.map

package/dist/eval/live/bridges/skillflow_e2b_bridge.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"skillflow_e2b_bridge.js","sourceRoot":"","sources":["../../../../src/eval/live/bridges/skillflow_e2b_bridge.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,kCAAkC,EAAE,MAAM,iBAAiB,CAAC;AAErE,IAAI,CAAC;IACH,kCAAkC,CAAC,KAAK,CAAC,CAAC;IAC1C,MAAM,EAAE,qBAAqB,EAAE,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,CAAC;IAChE,MAAM,qBAAqB,EAAE,CAAC;AAChC,CAAC;AAAC,OAAO,GAAG,EAAE,CAAC;IACb,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAG,GAAa,CAAC,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;IAC/E,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;AACvB,CAAC"}

package/dist/eval/live/claude-cli-llm-provider.d.ts

@@ -0,0 +1,53 @@
+import type { LLMProvider, Message, CompletionOptions } from '../../agents/types.js';
+import type { ClaudeCliOptions } from './claude-cli.js';
+/**
+ * LLMProvider adapter that routes completions through the Claude CLI (`claude -p`).
+ *
+ * Used by the eval learning arm to wire a real model into the distiller and
+ * convention-distiller so they run with an actual LLM rather than falling back
+ * to heuristics.
+ *
+ * Message flattening strategy
+ * ---------------------------
+ * The Claude CLI takes a single prompt string on stdin. Multi-message histories
+ * are flattened as:
+ *
+ *   [SYSTEM]\n<system content>\n\n[USER]\n<user content>\n\n[ASSISTANT]\n<assistant content>\n\n...
+ *
+ * A sole user message (the common distiller case) is sent directly as the
+ * prompt with no role prefix so the CLI treats it as a plain question.
+ *
+ * Temperature / maxTokens
+ * -----------------------
+ * The Claude CLI (`claude -p`) does not expose `--temperature` or `--max-tokens`
+ * flags in its current headless interface. Temperature is therefore silently
+ * ignored — this is acceptable for distillation because the distiller already
+ * requests determinism via its prompt wording and because the same batch of
+ * trajectories rarely repeats within a run. maxTokens is also not forwarded.
+ * If the CLI gains these flags in a future version, add them to the args array
+ * in runClaudeOnce and thread them from CompletionOptions here.
+ */
+export declare class ClaudeCliLLMProvider implements LLMProvider {
+    readonly name = "claude-cli";
+    private readonly opts;
+    constructor(opts?: ClaudeCliOptions);
+    complete(messages: Message[], _options?: CompletionOptions): Promise<string>;
+}
+/**
+ * Flatten a messages array into a single prompt string for the Claude CLI.
+ *
+ * - Single user message (no system, no prior assistant turns): sent as-is.
+ * - Otherwise: each role is prefixed with `[ROLE]\n` and turns are separated
+ *   by a blank line. The CLI sees the whole conversation as one text block.
+ */
+export declare function flattenMessages(messages: Message[]): string;
+/**
+ * Factory for ClaudeCliLLMProvider.
+ *
+ * @param opts  ClaudeCliOptions forwarded to runClaude on every call.
+ *              Pass `model`, `provider`, `awsRegion`, and `timeoutMs` from
+ *              the live run config so the distiller uses the same model/infra
+ *              as the eval agent.
+ */
+export declare function createClaudeCliLLMProvider(opts?: ClaudeCliOptions): ClaudeCliLLMProvider;
+//# sourceMappingURL=claude-cli-llm-provider.d.ts.map

package/dist/eval/live/claude-cli-llm-provider.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-cli-llm-provider.d.ts","sourceRoot":"","sources":["../../../src/eval/live/claude-cli-llm-provider.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,OAAO,EAAE,iBAAiB,EAAE,MAAM,uBAAuB,CAAC;AAErF,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAExD;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,qBAAa,oBAAqB,YAAW,WAAW;IACtD,QAAQ,CAAC,IAAI,gBAAgB;IAE7B,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAmB;gBAE5B,IAAI,GAAE,gBAAqB;IAIjC,QAAQ,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,QAAQ,CAAC,EAAE,iBAAiB,GAAG,OAAO,CAAC,MAAM,CAAC;CAKnF;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAY3D;AAED;;;;;;;GAOG;AACH,wBAAgB,0BAA0B,CAAC,IAAI,GAAE,gBAAqB,GAAG,oBAAoB,CAE5F"}

package/dist/eval/live/claude-cli-llm-provider.js

@@ -0,0 +1,71 @@
+import { runClaude } from './claude-cli.js';
+/**
+ * LLMProvider adapter that routes completions through the Claude CLI (`claude -p`).
+ *
+ * Used by the eval learning arm to wire a real model into the distiller and
+ * convention-distiller so they run with an actual LLM rather than falling back
+ * to heuristics.
+ *
+ * Message flattening strategy
+ * ---------------------------
+ * The Claude CLI takes a single prompt string on stdin. Multi-message histories
+ * are flattened as:
+ *
+ *   [SYSTEM]\n<system content>\n\n[USER]\n<user content>\n\n[ASSISTANT]\n<assistant content>\n\n...
+ *
+ * A sole user message (the common distiller case) is sent directly as the
+ * prompt with no role prefix so the CLI treats it as a plain question.
+ *
+ * Temperature / maxTokens
+ * -----------------------
+ * The Claude CLI (`claude -p`) does not expose `--temperature` or `--max-tokens`
+ * flags in its current headless interface. Temperature is therefore silently
+ * ignored — this is acceptable for distillation because the distiller already
+ * requests determinism via its prompt wording and because the same batch of
+ * trajectories rarely repeats within a run. maxTokens is also not forwarded.
+ * If the CLI gains these flags in a future version, add them to the args array
+ * in runClaudeOnce and thread them from CompletionOptions here.
+ */
+export class ClaudeCliLLMProvider {
+    name = 'claude-cli';
+    opts;
+    constructor(opts = {}) {
+        this.opts = opts;
+    }
+    async complete(messages, _options) {
+        const prompt = flattenMessages(messages);
+        const result = await runClaude(prompt, this.opts);
+        return result.text;
+    }
+}
+/**
+ * Flatten a messages array into a single prompt string for the Claude CLI.
+ *
+ * - Single user message (no system, no prior assistant turns): sent as-is.
+ * - Otherwise: each role is prefixed with `[ROLE]\n` and turns are separated
+ *   by a blank line. The CLI sees the whole conversation as one text block.
+ */
+export function flattenMessages(messages) {
+    // Fast path: single user-only message
+    if (messages.length === 1 && messages[0].role === 'user') {
+        return messages[0].content;
+    }
+    return messages
+        .map((m) => {
+        const label = m.role === 'system' ? '[SYSTEM]' : m.role === 'user' ? '[USER]' : '[ASSISTANT]';
+        return `${label}\n${m.content}`;
+    })
+        .join('\n\n');
+}
+/**
+ * Factory for ClaudeCliLLMProvider.
+ *
+ * @param opts  ClaudeCliOptions forwarded to runClaude on every call.
+ *              Pass `model`, `provider`, `awsRegion`, and `timeoutMs` from
+ *              the live run config so the distiller uses the same model/infra
+ *              as the eval agent.
+ */
+export function createClaudeCliLLMProvider(opts = {}) {
+    return new ClaudeCliLLMProvider(opts);
+}
+//# sourceMappingURL=claude-cli-llm-provider.js.map

package/dist/eval/live/claude-cli-llm-provider.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-cli-llm-provider.js","sourceRoot":"","sources":["../../../src/eval/live/claude-cli-llm-provider.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAG5C;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,MAAM,OAAO,oBAAoB;IACtB,IAAI,GAAG,YAAY,CAAC;IAEZ,IAAI,CAAmB;IAExC,YAAY,OAAyB,EAAE;QACrC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,QAAmB,EAAE,QAA4B;QAC9D,MAAM,MAAM,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;QACzC,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QAClD,OAAO,MAAM,CAAC,IAAI,CAAC;IACrB,CAAC;CACF;AAED;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAAC,QAAmB;IACjD,sCAAsC;IACtC,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACzD,OAAO,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IAC7B,CAAC;IAED,OAAO,QAAQ;SACZ,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;QACT,MAAM,KAAK,GAAG,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC;QAC9F,OAAO,GAAG,KAAK,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC;IAClC,CAAC,CAAC;SACD,IAAI,CAAC,MAAM,CAAC,CAAC;AAClB,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,0BAA0B,CAAC,OAAyB,EAAE;IACpE,OAAO,IAAI,oBAAoB,CAAC,IAAI,CAAC,CAAC;AACxC,CAAC"}

package/dist/eval/live/claude-cli.d.ts

@@ -0,0 +1,96 @@
+/**
+ * Thin wrapper around the headless Claude CLI (`claude -p`).
+ *
+ * Runs one non-interactive completion and returns the final assistant text.
+ * This is the real-agent substrate for live evals — the Claude Max plan in the
+ * environment means no API key wrangling. Kept tiny and dependency-free so both
+ * the single-turn delegate and the ALFWorld multi-turn loop can reuse it.
+ */
+export interface ClaudeCliOptions {
+    /** Model selector passed to `--model` (e.g. "haiku", "sonnet"). Omit for default. */
+    model?: string;
+    /** Hard timeout in ms per attempt (default 90s). */
+    timeoutMs?: number;
+    /** Resume a prior session id to preserve context across turns. */
+    resumeSessionId?: string;
+    /** Retries on timeout/spawn error (default 2) — rides through transient CLI throttling. */
+    retries?: number;
+    /** Working directory for tool-capable workspace tasks. */
+    cwd?: string;
+    /** Permission mode for tool-capable workspace tasks. */
+    permissionMode?: 'acceptEdits' | 'auto' | 'bypassPermissions' | 'default' | 'dontAsk' | 'plan';
+    /** Optional allowed tool list passed through to Claude CLI. */
+    allowedTools?: string[];
+    /** API provider override (e.g. "bedrock", "vertex"). Passed via --settings (legacy). */
+    provider?: string;
+    /** AWS region for Bedrock provider (default "us-east-1"). */
+    awsRegion?: string;
+    /**
+     * Route this call through Amazon Bedrock via env vars (the approach that works
+     * for `claude -p`): sets CLAUDE_CODE_USE_BEDROCK=1, AWS_REGION, and
+     * ANTHROPIC_MODEL=<inference-profile id>. Under Bedrock `--model` is DROPPED —
+     * CLI aliases ("haiku"/"sonnet") are not valid Bedrock ids (→ 400). `model` is
+     * mapped to its inference-profile id (see BEDROCK_MODELS). Needs ~/.aws creds.
+     */
+    bedrock?: boolean;
+    /**
+     * When true, pass `--disallowedTools` to make the step a SINGLE inference that
+     * emits text (one command/action) rather than running an autonomous agentic
+     * loop with tool access. REQUIRED for benchmarks where the agent must emit a
+     * command for the *bridge* to execute (SkillFlow/CL-Bench): with tools enabled
+     * the CLI builds artifacts in its own ephemeral sandbox, invisible to the
+     * bridge's verifier (false failures) and ~minutes/step. Harmless for ALFWorld
+     * (action-only prompts are already num_turns≈1). See DESIGN-eval-status.md.
+     */
+    disallowTools?: boolean;
+    /** Internal retry override: force a specific Claude binary for this attempt. */
+    commandOverride?: string;
+}
+export declare function resolveClaudeCommand(): string;
+export declare function claudePathEntries(command?: string): string[];
+export declare function buildClaudeChildEnv(base: NodeJS.ProcessEnv, command?: string): NodeJS.ProcessEnv;
+/**
+ * Friendly model name → Bedrock cross-region inference-profile id. Used for
+ * ANTHROPIC_MODEL when `bedrock` is set. Unknown names pass through (so a full
+ * profile id can be supplied directly). Ids verified against the account's
+ * enabled profiles (us-east-1).
+ */
+export declare const BEDROCK_MODELS: Record<string, string>;
+export declare function bedrockModelId(name: string, region?: string): string;
+/**
+ * Run the Claude CLI with retry-on-failure. A single hung/throttled `claude -p`
+ * call should not abort an entire multi-hour eval; we retry a couple of times
+ * with backoff before giving up.
+ */
+export declare function runClaude(prompt: string, opts?: ClaudeCliOptions): Promise<ClaudeCliResult>;
+/** Extra Azure Responses fields resolved from ambient credentials. */
+export interface AzureResponsesEnv {
+    apiKey?: string;
+    azureEndpoint?: string;
+    azureApiVersion?: string;
+    azureDeployment?: string;
+}
+/**
+ * Bridge the Azure credential naming gap for the Responses solver/learning path.
+ *
+ * swarmkit-eval's `generateResponsesText` only reads the canonical
+ * `AZURE_OPENAI_*` env names, but this environment commonly exports the shorter
+ * `AZURE_API_*` aliases (endpoint/version/key). Resolve both, canonical names
+ * winning, and pass them explicitly so `--provider azure` works with whichever
+ * credential set is present. Deployment stays `AZURE_OPENAI_DEPLOYMENT` only;
+ * when unset the caller's `--model` becomes the deployment name downstream.
+ */
+export declare function resolveAzureResponsesEnv(env?: NodeJS.ProcessEnv): AzureResponsesEnv;
+export interface ClaudeCliResult {
+    /** Final assistant text (`result` field of the JSON output). */
+    text: string;
+    /** Session id, for `--resume` continuation in multi-turn loops. */
+    sessionId?: string;
+    /** Whether the CLI reported a non-error result. */
+    ok: boolean;
+    /** Output tokens (for cost/telemetry), if reported. */
+    costUsd?: number;
+}
+/** Quick availability probe — true if the `claude` binary is on PATH. */
+export declare function claudeAvailable(): Promise<boolean>;
+//# sourceMappingURL=claude-cli.d.ts.map

package/dist/eval/live/claude-cli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-cli.d.ts","sourceRoot":"","sources":["../../../src/eval/live/claude-cli.ts"],"names":[],"mappings":"AAKA;;;;;;;GAOG;AAEH,MAAM,WAAW,gBAAgB;IAC/B,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,oDAAoD;IACpD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kEAAkE;IAClE,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,2FAA2F;IAC3F,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,0DAA0D;IAC1D,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,wDAAwD;IACxD,cAAc,CAAC,EAAE,aAAa,GAAG,MAAM,GAAG,mBAAmB,GAAG,SAAS,GAAG,SAAS,GAAG,MAAM,CAAC;IAC/F,+DAA+D;IAC/D,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,wFAAwF;IACxF,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,6DAA6D;IAC7D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;;;OAMG;IACH,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB;;;;;;;;OAQG;IACH,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,gFAAgF;IAChF,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B;AAuBD,wBAAgB,oBAAoB,IAAI,MAAM,CAO7C;AAED,wBAAgB,iBAAiB,CAAC,OAAO,SAAyB,GAAG,MAAM,EAAE,CAO5E;AAED,wBAAgB,mBAAmB,CACjC,IAAI,EAAE,MAAM,CAAC,UAAU,EACvB,OAAO,SAAyB,GAC/B,MAAM,CAAC,UAAU,CAQnB;AAED;;;;;GAKG;AACH,eAAO,MAAM,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAIjD,CAAC;AAcF,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,MAAM,CAIpE;AAED;;;;GAIG;AACH,wBAAsB,SAAS,CAC7B,MAAM,EAAE,MAAM,EACd,IAAI,GAAE,gBAAqB,GAC1B,OAAO,CAAC,eAAe,CAAC,CAiC1B;AAED,sEAAsE;AACtE,MAAM,WAAW,iBAAiB;IAChC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED;;;;;;;;;GASG;AACH,wBAAgB,wBAAwB,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB,GAAG,iBAAiB,CAahG;AAuBD,MAAM,WAAW,eAAe;IAC9B,gEAAgE;IAChE,IAAI,EAAE,MAAM,CAAC;IACb,mEAAmE;IACnE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mDAAmD;IACnD,EAAE,EAAE,OAAO,CAAC;IACZ,uDAAuD;IACvD,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAiKD,yEAAyE;AACzE,wBAAsB,eAAe,IAAI,OAAO,CAAC,OAAO,CAAC,CAMxD"}