@steipete/summarize 0.11.1 → 0.12.0

package/dist/esm/run/flows/url/summary.js

@@ -1,174 +1,26 @@
-import { isTwitterStatusUrl, isYouTubeUrl } from "@steipete/summarize-core/content/url";
-import { countTokens } from "gpt-tokenizer";
 import { render as renderMarkdownAnsi } from "markdansi";
-import { buildLanguageKey, buildLengthKey, buildPromptHash, buildSummaryCacheKey, hashString, normalizeContentForHash, } from "../../../cache.js";
-import { formatOutputLanguageForJson } from "../../../language.js";
-import { parseGatewayStyleModelId } from "../../../llm/model-id.js";
-import { buildAutoModelAttempts } from "../../../model-auto.js";
-import { buildLinkSummaryPrompt, SUMMARY_LENGTH_TARGET_CHARACTERS, SUMMARY_SYSTEM_PROMPT, } from "../../../prompts/index.js";
-import { readLastSuccessfulCliProvider, writeLastSuccessfulCliProvider, } from "../../cli-fallback-state.js";
-import { parseCliUserModelId } from "../../env.js";
-import { buildExtractFinishLabel, buildLengthPartsForFinishLine, writeFinishLine, } from "../../finish-line.js";
-import { resolveTargetCharacters } from "../../format.js";
+import { buildExtractFinishLabel, writeFinishLine } from "../../finish-line.js";
 import { writeVerbose } from "../../logging.js";
 import { prepareMarkdownForTerminal } from "../../markdown.js";
-import { runModelAttempts } from "../../model-attempts.js";
-import { buildOpenRouterNoAllowedProvidersMessage } from "../../openrouter.js";
 import { isRichTty, markdownRenderWidth, supportsColor } from "../../terminal.js";
-import { coerceSummaryWithSlides, interleaveSlidesIntoTranscript, normalizeSummarySlideHeadings, } from "./slides-text.js";
-const MAX_SLIDE_TRANSCRIPT_CHARS_BY_PRESET = {
-    short: 2500,
-    medium: 5000,
-    long: 9000,
-    xl: 15000,
-    xxl: 24000,
-};
-const SLIDE_TRANSCRIPT_DEFAULT_EDGE_SECONDS = 30;
-const SLIDE_TRANSCRIPT_LEEWAY_SECONDS = 10;
-function parseTimestampSeconds(value) {
-    const parts = value.split(":").map((item) => Number(item));
-    if (parts.some((item) => !Number.isFinite(item)))
-        return null;
-    if (parts.length === 2) {
-        const [minutes, seconds] = parts;
-        return minutes * 60 + seconds;
-    }
-    if (parts.length === 3) {
-        const [hours, minutes, seconds] = parts;
-        return hours * 3600 + minutes * 60 + seconds;
-    }
-    return null;
-}
-function parseTranscriptTimedText(input) {
-    if (!input)
-        return [];
-    const segments = [];
-    for (const line of input.split("\n")) {
-        const trimmed = line.trim();
-        if (!trimmed.startsWith("["))
-            continue;
-        const match = trimmed.match(/^\[(\d{1,2}:\d{2}(?::\d{2})?)\]\s*(.*)$/);
-        if (!match)
-            continue;
-        const seconds = parseTimestampSeconds(match[1]);
-        if (seconds == null)
-            continue;
-        const text = (match[2] ?? "").trim();
-        if (!text)
-            continue;
-        segments.push({ startSeconds: seconds, text });
-    }
-    segments.sort((a, b) => a.startSeconds - b.startSeconds);
-    return segments;
-}
-function formatTimestamp(seconds) {
-    const clamped = Math.max(0, Math.floor(seconds));
-    const hours = Math.floor(clamped / 3600);
-    const minutes = Math.floor((clamped % 3600) / 60);
-    const secs = clamped % 60;
-    const mm = String(minutes).padStart(2, "0");
-    const ss = String(secs).padStart(2, "0");
-    if (hours <= 0)
-        return `${minutes}:${ss}`;
-    const hh = String(hours).padStart(2, "0");
-    return `${hh}:${mm}:${ss}`;
-}
-function truncateTranscript(value, limit) {
-    if (value.length <= limit)
-        return value;
-    const truncated = value.slice(0, limit).trimEnd();
-    const clean = truncated.replace(/\s+\S*$/, "").trim();
-    const result = clean.length > 0 ? clean : truncated.trim();
-    return result.length > 0 ? `${result}…` : "";
-}
-function buildSlidesPromptText({ slides, transcriptTimedText, preset, }) {
-    if (!slides || slides.slides.length === 0)
-        return null;
-    const segments = parseTranscriptTimedText(transcriptTimedText);
-    const slidesWithTimestamps = slides.slides
-        .filter((slide) => Number.isFinite(slide.timestamp))
-        .map((slide) => ({ index: slide.index, timestamp: Math.max(0, Math.floor(slide.timestamp)) }))
-        .sort((a, b) => a.timestamp - b.timestamp);
-    if (slidesWithTimestamps.length === 0)
-        return null;
-    const totalBudget = Number(MAX_SLIDE_TRANSCRIPT_CHARS_BY_PRESET[preset]);
-    const perSlideBudget = Math.max(120, Math.floor(totalBudget / Math.max(1, slidesWithTimestamps.length)));
-    let remaining = totalBudget;
-    const blocks = [];
-    for (let i = 0; i < slidesWithTimestamps.length; i += 1) {
-        const slide = slidesWithTimestamps[i];
-        if (!slide)
-            continue;
-        const prev = slidesWithTimestamps[i - 1];
-        const next = slidesWithTimestamps[i + 1];
-        const startBase = prev ? Math.floor((prev.timestamp + slide.timestamp) / 2) : slide.timestamp;
-        const endBase = next ? Math.ceil((slide.timestamp + next.timestamp) / 2) : slide.timestamp;
-        const start = Math.max(0, (prev ? startBase : slide.timestamp - SLIDE_TRANSCRIPT_DEFAULT_EDGE_SECONDS) -
-            SLIDE_TRANSCRIPT_LEEWAY_SECONDS);
-        const end = (next ? endBase : slide.timestamp + SLIDE_TRANSCRIPT_DEFAULT_EDGE_SECONDS) +
-            SLIDE_TRANSCRIPT_LEEWAY_SECONDS;
-        const excerptParts = [];
-        for (const segment of segments) {
-            if (segment.startSeconds < start)
-                continue;
-            if (segment.startSeconds > end)
-                break;
-            excerptParts.push(segment.text);
-        }
-        const excerptRaw = excerptParts.join(" ").trim().replace(/\s+/g, " ");
-        const excerptBudget = remaining > 0 ? Math.min(perSlideBudget, remaining) : 0;
-        const excerpt = excerptRaw && excerptBudget > 0 ? truncateTranscript(excerptRaw, excerptBudget) : "";
-        const label = `[slide:${slide.index}] [${formatTimestamp(start)}–${formatTimestamp(end)}]`;
-        const block = excerpt ? `${label}\n${excerpt}` : label;
-        blocks.push(block);
-        remaining = Math.max(0, remaining - block.length);
-    }
-    return blocks.length > 0 ? blocks.join("\n\n") : null;
-}
+import { coerceSummaryWithSlides, interleaveSlidesIntoTranscript, } from "./slides-text.js";
+import { buildFinishExtras, pickModelForFinishLine, } from "./summary-finish.js";
+import { buildUrlJsonEnv, buildUrlJsonInput } from "./summary-json.js";
+import { buildUrlPrompt as buildSummaryPrompt, } from "./summary-prompt.js";
+import { resolveUrlSummaryExecution } from "./summary-resolution.js";
+import { buildSummaryTimestampLimitInstruction } from "./summary-timestamps.js";
 export function buildUrlPrompt({ extracted, outputLanguage, lengthArg, promptOverride, lengthInstruction, languageInstruction, slides, }) {
-    const isYouTube = extracted.siteName === "YouTube";
-    const preset = lengthArg.kind === "preset" ? lengthArg.preset : "medium";
-    const slidesText = buildSlidesPromptText({
-        slides,
-        transcriptTimedText: extracted.transcriptTimedText,
-        preset,
-    });
-    return buildLinkSummaryPrompt({
-        url: extracted.url,
-        title: extracted.title,
-        siteName: extracted.siteName,
-        description: extracted.description,
-        content: extracted.content,
-        truncated: extracted.truncated,
-        hasTranscript: isYouTube ||
-            (extracted.transcriptSource !== null && extracted.transcriptSource !== "unavailable"),
-        hasTranscriptTimestamps: Boolean(extracted.transcriptTimedText),
-        slides: slidesText ? { count: slides?.slides.length ?? 0, text: slidesText } : null,
-        summaryLength: lengthArg.kind === "preset" ? lengthArg.preset : { maxCharacters: lengthArg.maxCharacters },
+    return buildSummaryPrompt({
+        extracted,
         outputLanguage,
-        shares: [],
-        promptOverride: promptOverride ?? null,
-        lengthInstruction: lengthInstruction ?? null,
-        languageInstruction: languageInstruction ?? null,
+        lengthArg,
+        promptOverride,
+        lengthInstruction,
+        languageInstruction,
+        slides,
+        buildSummaryTimestampLimitInstruction,
     });
 }
-function shouldBypassShortContentSummary({ extracted, lengthArg, forceSummary, maxOutputTokensArg, json, }) {
-    if (forceSummary)
-        return false;
-    if (!extracted.content || extracted.content.length === 0)
-        return false;
-    const targetCharacters = resolveTargetCharacters(lengthArg, SUMMARY_LENGTH_TARGET_CHARACTERS);
-    if (!Number.isFinite(targetCharacters) || targetCharacters <= 0)
-        return false;
-    if (extracted.content.length > targetCharacters)
-        return false;
-    if (!json && typeof maxOutputTokensArg === "number") {
-        const tokenCount = countTokens(extracted.content);
-        if (tokenCount > maxOutputTokensArg)
-            return false;
-    }
-    return true;
-}
 async function outputSummaryFromExtractedContent({ ctx, url, extracted, extractionUi, prompt, effectiveMarkdownMode, transcriptionCostLabel, slides, footerLabel, verboseMessage, }) {
     const { io, flags, model, hooks } = ctx;
     hooks.clearProgressForStdout();
@@ -177,30 +29,14 @@ async function outputSummaryFromExtractedContent({ ctx, url, extracted, extracti
         const finishReport = flags.shouldComputeReport ? await hooks.buildReport() : null;
         const payload = {
             input: {
-                kind: "url",
-                url,
-                timeoutMs: flags.timeoutMs,
-                youtube: flags.youtubeMode,
-                firecrawl: flags.firecrawlMode,
-                format: flags.format,
-                markdown: effectiveMarkdownMode,
-                timestamps: flags.transcriptTimestamps,
-                length: flags.lengthArg.kind === "preset"
-                    ? { kind: "preset", preset: flags.lengthArg.preset }
-                    : { kind: "chars", maxCharacters: flags.lengthArg.maxCharacters },
-                maxOutputTokens: flags.maxOutputTokensArg,
-                model: model.requestedModelLabel,
-                language: formatOutputLanguageForJson(flags.outputLanguage),
-            },
-            env: {
-                hasXaiKey: Boolean(model.apiStatus.xaiApiKey),
-                hasOpenAIKey: Boolean(model.apiStatus.apiKey),
-                hasOpenRouterKey: Boolean(model.apiStatus.openrouterApiKey),
-                hasApifyToken: Boolean(model.apiStatus.apifyToken),
-                hasFirecrawlKey: model.apiStatus.firecrawlConfigured,
-                hasGoogleKey: model.apiStatus.googleConfigured,
-                hasAnthropicKey: model.apiStatus.anthropicConfigured,
+                ...buildUrlJsonInput({
+                    flags,
+                    url,
+                    effectiveMarkdownMode,
+                    modelLabel: model.requestedModelLabel,
+                }),
             },
+            env: buildUrlJsonEnv(model.apiStatus),
             extracted,
             slides,
             prompt,
@@ -243,37 +79,6 @@ async function outputSummaryFromExtractedContent({ ctx, url, extracted, extracti
         writeVerbose(io.stderr, flags.verbose, verboseMessage, flags.verboseColor, io.envForRun);
     }
 }
-const buildFinishExtras = ({ extracted, metricsDetailed, transcriptionCostLabel, }) => {
-    const parts = [
-        ...(buildLengthPartsForFinishLine(extracted, metricsDetailed) ?? []),
-        ...(transcriptionCostLabel ? [transcriptionCostLabel] : []),
-    ];
-    return parts.length > 0 ? parts : null;
-};
-const pickModelForFinishLine = (llmCalls, fallback) => {
-    const findLastModel = (purpose) => {
-        for (let i = llmCalls.length - 1; i >= 0; i -= 1) {
-            const call = llmCalls[i];
-            if (call && call.purpose === purpose)
-                return call.model;
-        }
-        return null;
-    };
-    return (findLastModel("summary") ??
-        findLastModel("markdown") ??
-        (llmCalls.length > 0 ? (llmCalls[llmCalls.length - 1]?.model ?? null) : null) ??
-        fallback);
-};
-const buildModelMetaFromAttempt = (attempt) => {
-    if (attempt.transport === "cli") {
-        return { provider: "cli", canonical: attempt.userModelId };
-    }
-    const parsed = parseGatewayStyleModelId(attempt.llmModelId ?? attempt.userModelId);
-    const canonical = attempt.userModelId.toLowerCase().startsWith("openrouter/")
-        ? attempt.userModelId
-        : parsed.canonical;
-    return { provider: parsed.provider, canonical };
-};
 export async function outputExtractedUrl({ ctx, url, extracted, extractionUi, prompt, effectiveMarkdownMode, transcriptionCostLabel, slides, slidesOutput, }) {
     const { io, flags, model, hooks } = ctx;
     hooks.clearProgressForStdout();
@@ -288,30 +93,14 @@ export async function outputExtractedUrl({ ctx, url, extracted, extractionUi, pr
         const finishReport = flags.shouldComputeReport ? await hooks.buildReport() : null;
         const payload = {
             input: {
-                kind: "url",
-                url,
-                timeoutMs: flags.timeoutMs,
-                youtube: flags.youtubeMode,
-                firecrawl: flags.firecrawlMode,
-                format: flags.format,
-                markdown: effectiveMarkdownMode,
-                timestamps: flags.transcriptTimestamps,
-                length: flags.lengthArg.kind === "preset"
-                    ? { kind: "preset", preset: flags.lengthArg.preset }
-                    : { kind: "chars", maxCharacters: flags.lengthArg.maxCharacters },
-                maxOutputTokens: flags.maxOutputTokensArg,
-                model: model.requestedModelLabel,
-                language: formatOutputLanguageForJson(flags.outputLanguage),
-            },
-            env: {
-                hasXaiKey: Boolean(model.apiStatus.xaiApiKey),
-                hasOpenAIKey: Boolean(model.apiStatus.apiKey),
-                hasOpenRouterKey: Boolean(model.apiStatus.openrouterApiKey),
-                hasApifyToken: Boolean(model.apiStatus.apifyToken),
-                hasFirecrawlKey: model.apiStatus.firecrawlConfigured,
-                hasGoogleKey: model.apiStatus.googleConfigured,
-                hasAnthropicKey: model.apiStatus.anthropicConfigured,
+                ...buildUrlJsonInput({
+                    flags,
+                    url,
+                    effectiveMarkdownMode,
+                    modelLabel: model.requestedModelLabel,
+                }),
             },
+            env: buildUrlJsonEnv(model.apiStatus),
             extracted,
             slides,
             prompt,
@@ -435,251 +224,16 @@ export async function outputExtractedUrl({ ctx, url, extracted, extractionUi, pr
 }
 export async function summarizeExtractedUrl({ ctx, url, extracted, extractionUi, prompt, effectiveMarkdownMode, transcriptionCostLabel, onModelChosen, slides, slidesOutput, }) {
     const { io, flags, model, cache: cacheState, hooks } = ctx;
-    const lastSuccessfulCliProvider = model.isFallbackModel
-        ? await readLastSuccessfulCliProvider(io.envForRun)
-        : null;
-    const promptPayload = { system: SUMMARY_SYSTEM_PROMPT, userText: prompt };
-    const promptTokens = countTokens(promptPayload.userText);
-    const kindForAuto = extracted.siteName === "YouTube" ? "youtube" : "website";
-    const attempts = await (async () => {
-        if (model.isFallbackModel) {
-            const catalog = await model.getLiteLlmCatalog();
-            const list = buildAutoModelAttempts({
-                kind: kindForAuto,
-                promptTokens,
-                desiredOutputTokens: model.desiredOutputTokens,
-                requiresVideoUnderstanding: false,
-                env: model.envForAuto,
-                config: model.configForModelSelection,
-                catalog,
-                openrouterProvidersFromEnv: null,
-                cliAvailability: model.cliAvailability,
-                isImplicitAutoSelection: model.isImplicitAutoSelection,
-                allowAutoCliFallback: model.allowAutoCliFallback,
-                lastSuccessfulCliProvider,
-            });
-            if (flags.verbose) {
-                for (const attempt of list.slice(0, 8)) {
-                    writeVerbose(io.stderr, flags.verbose, `auto candidate ${attempt.debug}`, flags.verboseColor, io.envForRun);
-                }
-            }
-            return list.map((attempt) => {
-                if (attempt.transport !== "cli")
-                    return model.summaryEngine.applyZaiOverrides(attempt);
-                const parsed = parseCliUserModelId(attempt.userModelId);
-                return { ...attempt, cliProvider: parsed.provider, cliModel: parsed.model };
-            });
-        }
-        /* v8 ignore next */
-        if (!model.fixedModelSpec) {
-            throw new Error("Internal error: missing fixed model spec");
-        }
-        if (model.fixedModelSpec.transport === "cli") {
-            return [
-                {
-                    transport: "cli",
-                    userModelId: model.fixedModelSpec.userModelId,
-                    llmModelId: null,
-                    cliProvider: model.fixedModelSpec.cliProvider,
-                    cliModel: model.fixedModelSpec.cliModel,
-                    openrouterProviders: null,
-                    forceOpenRouter: false,
-                    requiredEnv: model.fixedModelSpec.requiredEnv,
-                },
-            ];
-        }
-        const openaiOverrides = model.fixedModelSpec.requiredEnv === "Z_AI_API_KEY"
-            ? {
-                openaiApiKeyOverride: model.apiStatus.zaiApiKey,
-                openaiBaseUrlOverride: model.apiStatus.zaiBaseUrl,
-                forceChatCompletions: true,
-            }
-            : {};
-        return [
-            {
-                transport: model.fixedModelSpec.transport === "openrouter" ? "openrouter" : "native",
-                userModelId: model.fixedModelSpec.userModelId,
-                llmModelId: model.fixedModelSpec.llmModelId,
-                openrouterProviders: model.fixedModelSpec.openrouterProviders,
-                forceOpenRouter: model.fixedModelSpec.forceOpenRouter,
-                requiredEnv: model.fixedModelSpec.requiredEnv,
-                ...openaiOverrides,
-            },
-        ];
-    })();
-    const cacheStore = cacheState.mode === "default" && !flags.summaryCacheBypass ? cacheState.store : null;
-    const contentHash = cacheStore ? hashString(normalizeContentForHash(extracted.content)) : null;
-    const promptHash = cacheStore ? buildPromptHash(prompt) : null;
-    const lengthKey = buildLengthKey(flags.lengthArg);
-    const languageKey = buildLanguageKey(flags.outputLanguage);
-    const autoSelectionCacheModel = model.isFallbackModel
-        ? `selection:${model.requestedModelInput.toLowerCase()}`
-        : null;
-    let summaryResult = null;
-    let usedAttempt = null;
-    let summaryFromCache = false;
-    let cacheChecked = false;
-    const isTweet = extracted.siteName?.toLowerCase() === "x" || isTwitterStatusUrl(extracted.url);
-    const isYouTube = extracted.siteName === "YouTube" || isYouTubeUrl(url);
-    const hasMedia = Boolean(extracted.video) ||
-        (extracted.transcriptSource != null && extracted.transcriptSource !== "unavailable") ||
-        (typeof extracted.mediaDurationSeconds === "number" && extracted.mediaDurationSeconds > 0) ||
-        extracted.isVideoOnly === true;
-    const autoBypass = ctx.model.isFallbackModel && !ctx.model.isNamedModelSelection;
-    const canBypassShortContent = (autoBypass || isTweet) &&
-        !flags.slides &&
-        !hasMedia &&
-        flags.streamMode !== "on" &&
-        !isYouTube &&
-        shouldBypassShortContentSummary({
-            extracted,
-            lengthArg: flags.lengthArg,
-            forceSummary: flags.forceSummary,
-            maxOutputTokensArg: flags.maxOutputTokensArg,
-            json: flags.json,
-        });
-    if (canBypassShortContent) {
-        await outputSummaryFromExtractedContent({
-            ctx,
-            url,
-            extracted,
-            extractionUi,
-            prompt,
-            effectiveMarkdownMode,
-            transcriptionCostLabel,
-            slides,
-            footerLabel: "short content",
-            verboseMessage: "short content: skipping summary",
-        });
-        return;
-    }
-    if (cacheStore && contentHash && promptHash) {
-        cacheChecked = true;
-        if (autoSelectionCacheModel) {
-            const key = buildSummaryCacheKey({
-                contentHash,
-                promptHash,
-                model: autoSelectionCacheModel,
-                lengthKey,
-                languageKey,
-            });
-            const cached = cacheStore.getJson("summary", key);
-            const cachedSummary = cached && typeof cached.summary === "string" ? cached.summary.trim() : null;
-            const cachedModelId = cached && typeof cached.model === "string" ? cached.model.trim() : null;
-            if (cachedSummary) {
-                const cachedAttempt = cachedModelId
-                    ? (attempts.find((attempt) => attempt.userModelId === cachedModelId) ?? null)
-                    : null;
-                const fallbackAttempt = attempts.find((attempt) => model.summaryEngine.envHasKeyFor(attempt.requiredEnv)) ??
-                    attempts[0] ??
-                    null;
-                const matchedAttempt = cachedAttempt && model.summaryEngine.envHasKeyFor(cachedAttempt.requiredEnv)
-                    ? cachedAttempt
-                    : fallbackAttempt;
-                if (matchedAttempt) {
-                    writeVerbose(io.stderr, flags.verbose, "cache hit summary (auto selection)", flags.verboseColor, io.envForRun);
-                    onModelChosen?.(cachedModelId || matchedAttempt.userModelId);
-                    summaryResult = {
-                        summary: cachedSummary,
-                        summaryAlreadyPrinted: false,
-                        modelMeta: buildModelMetaFromAttempt(matchedAttempt),
-                        maxOutputTokensForCall: null,
-                    };
-                    usedAttempt = matchedAttempt;
-                    summaryFromCache = true;
-                }
-            }
-        }
-        if (!summaryFromCache) {
-            for (const attempt of attempts) {
-                if (!model.summaryEngine.envHasKeyFor(attempt.requiredEnv))
-                    continue;
-                const key = buildSummaryCacheKey({
-                    contentHash,
-                    promptHash,
-                    model: attempt.userModelId,
-                    lengthKey,
-                    languageKey,
-                });
-                const cached = cacheStore.getText("summary", key);
-                if (!cached)
-                    continue;
-                writeVerbose(io.stderr, flags.verbose, "cache hit summary", flags.verboseColor, io.envForRun);
-                onModelChosen?.(attempt.userModelId);
-                summaryResult = {
-                    summary: cached,
-                    summaryAlreadyPrinted: false,
-                    modelMeta: buildModelMetaFromAttempt(attempt),
-                    maxOutputTokensForCall: null,
-                };
-                usedAttempt = attempt;
-                summaryFromCache = true;
-                break;
-            }
-        }
-    }
-    if (cacheChecked && !summaryFromCache) {
-        writeVerbose(io.stderr, flags.verbose, "cache miss summary", flags.verboseColor, io.envForRun);
-    }
-    ctx.hooks.onSummaryCached?.(summaryFromCache);
-    let lastError = null;
-    let missingRequiredEnvs = new Set();
-    let sawOpenRouterNoAllowedProviders = false;
-    if (!summaryResult || !usedAttempt) {
-        const attemptOutcome = await runModelAttempts({
-            attempts,
-            isFallbackModel: model.isFallbackModel,
-            isNamedModelSelection: model.isNamedModelSelection,
-            envHasKeyFor: model.summaryEngine.envHasKeyFor,
-            formatMissingModelError: model.summaryEngine.formatMissingModelError,
-            onAutoSkip: (attempt) => {
-                writeVerbose(io.stderr, flags.verbose, `auto skip ${attempt.userModelId}: missing ${attempt.requiredEnv}`, flags.verboseColor, io.envForRun);
-            },
-            onAutoFailure: (attempt, error) => {
-                writeVerbose(io.stderr, flags.verbose, `auto failed ${attempt.userModelId}: ${error instanceof Error ? error.message : String(error)}`, flags.verboseColor, io.envForRun);
-            },
-            onFixedModelError: (_attempt, error) => {
-                throw error;
-            },
-            runAttempt: (attempt) => model.summaryEngine.runSummaryAttempt({
-                attempt,
-                prompt: promptPayload,
-                allowStreaming: flags.streamingEnabled,
-                onModelChosen: onModelChosen ?? null,
-                streamHandler: slidesOutput?.streamHandler ?? null,
-            }),
-        });
-        summaryResult = attemptOutcome.result;
-        usedAttempt = attemptOutcome.usedAttempt;
-        lastError = attemptOutcome.lastError;
-        missingRequiredEnvs = attemptOutcome.missingRequiredEnvs;
-        sawOpenRouterNoAllowedProviders = attemptOutcome.sawOpenRouterNoAllowedProviders;
-    }
-    if (!summaryResult || !usedAttempt) {
-        // Auto mode: surface raw extracted content when no model can run.
-        const withFreeTip = (message) => {
-            if (!model.isNamedModelSelection || !model.wantsFreeNamedModel)
-                return message;
-            return (`${message}\n` +
-                `Tip: run "summarize refresh-free" to refresh the free model candidates (writes ~/.summarize/config.json).`);
-        };
-        if (model.isNamedModelSelection) {
-            if (lastError === null && missingRequiredEnvs.size > 0) {
-                throw new Error(withFreeTip(`Missing ${Array.from(missingRequiredEnvs).sort().join(", ")} for --model ${model.requestedModelInput}.`));
-            }
-            if (lastError instanceof Error) {
-                if (sawOpenRouterNoAllowedProviders) {
-                    const message = await buildOpenRouterNoAllowedProvidersMessage({
-                        attempts,
-                        fetchImpl: io.fetch,
-                        timeoutMs: flags.timeoutMs,
-                    });
-                    throw new Error(withFreeTip(message), { cause: lastError });
-                }
-                throw new Error(withFreeTip(lastError.message), { cause: lastError });
-            }
-            throw new Error(withFreeTip(`No model available for --model ${model.requestedModelInput}`));
-        }
+    const resolution = await resolveUrlSummaryExecution({
+        ctx,
+        url,
+        extracted,
+        prompt,
+        onModelChosen,
+        slides,
+        slidesOutput,
+    });
+    if (resolution.kind === "use-extracted") {
         await outputSummaryFromExtractedContent({
             ctx,
             url,
@@ -689,72 +243,24 @@ export async function summarizeExtractedUrl({ ctx, url, extracted, extractionUi,
             effectiveMarkdownMode,
             transcriptionCostLabel,
             slides,
-            footerLabel: "no model",
-            verboseMessage: lastError instanceof Error ? `auto failed all models: ${lastError.message}` : null,
+            footerLabel: resolution.footerLabel,
+            verboseMessage: resolution.verboseMessage,
         });
         return;
     }
-    if (!summaryFromCache && cacheStore && contentHash && promptHash) {
-        const perModelKey = buildSummaryCacheKey({
-            contentHash,
-            promptHash,
-            model: usedAttempt.userModelId,
-            lengthKey,
-            languageKey,
-        });
-        cacheStore.setText("summary", perModelKey, summaryResult.summary, cacheState.ttlMs);
-        writeVerbose(io.stderr, flags.verbose, "cache write summary", flags.verboseColor, io.envForRun);
-        if (autoSelectionCacheModel) {
-            const selectionKey = buildSummaryCacheKey({
-                contentHash,
-                promptHash,
-                model: autoSelectionCacheModel,
-                lengthKey,
-                languageKey,
-            });
-            cacheStore.setJson("summary", selectionKey, { summary: summaryResult.summary, model: usedAttempt.userModelId }, cacheState.ttlMs);
-            writeVerbose(io.stderr, flags.verbose, "cache write summary (auto selection)", flags.verboseColor, io.envForRun);
-        }
-    }
-    if (!summaryFromCache &&
-        model.isFallbackModel &&
-        usedAttempt.transport === "cli" &&
-        usedAttempt.cliProvider) {
-        await writeLastSuccessfulCliProvider({
-            env: io.envForRun,
-            provider: usedAttempt.cliProvider,
-        });
-    }
-    const { summary, summaryAlreadyPrinted, modelMeta, maxOutputTokensForCall } = summaryResult;
-    const normalizedSummary = slides && slides.slides.length > 0 ? normalizeSummarySlideHeadings(summary) : summary;
+    const { normalizedSummary, summaryAlreadyPrinted, summaryFromCache, usedAttempt, modelMeta, maxOutputTokensForCall, } = resolution;
     if (flags.json) {
         const finishReport = flags.shouldComputeReport ? await hooks.buildReport() : null;
         const payload = {
             input: {
-                kind: "url",
-                url,
-                timeoutMs: flags.timeoutMs,
-                youtube: flags.youtubeMode,
-                firecrawl: flags.firecrawlMode,
-                format: flags.format,
-                markdown: effectiveMarkdownMode,
-                timestamps: flags.transcriptTimestamps,
-                length: flags.lengthArg.kind === "preset"
-                    ? { kind: "preset", preset: flags.lengthArg.preset }
-                    : { kind: "chars", maxCharacters: flags.lengthArg.maxCharacters },
-                maxOutputTokens: flags.maxOutputTokensArg,
-                model: model.requestedModelLabel,
-                language: formatOutputLanguageForJson(flags.outputLanguage),
-            },
-            env: {
-                hasXaiKey: Boolean(model.apiStatus.xaiApiKey),
-                hasOpenAIKey: Boolean(model.apiStatus.apiKey),
-                hasOpenRouterKey: Boolean(model.apiStatus.openrouterApiKey),
-                hasApifyToken: Boolean(model.apiStatus.apifyToken),
-                hasFirecrawlKey: model.apiStatus.firecrawlConfigured,
-                hasGoogleKey: model.apiStatus.googleConfigured,
-                hasAnthropicKey: model.apiStatus.anthropicConfigured,
+                ...buildUrlJsonInput({
+                    flags,
+                    url,
+                    effectiveMarkdownMode,
+                    modelLabel: model.requestedModelLabel,
+                }),
             },
+            env: buildUrlJsonEnv(model.apiStatus),
             extracted,
             slides,
             prompt,