@oh-my-pi/pi-coding-agent 15.5.6 → 15.5.8

package/src/session/agent-session.ts

@@ -18,6 +18,7 @@ import * as fs from "node:fs";
 import * as path from "node:path";
 import { scheduler } from "node:timers/promises";
 import { isPromise } from "node:util/types";
+import type { InMemorySnapshotStore } from "@oh-my-pi/hashline";
 import {
 	type AfterToolCallContext,
 	type AfterToolCallResult,
@@ -104,6 +105,8 @@ import { onAppendOnlyModeChanged } from "../config/settings";
 import { RawSseDebugBuffer } from "../debug/raw-sse-buffer";
 import { loadCapability } from "../discovery";
 import { expandApplyPatchToEntries, normalizeDiff, normalizeToLF, ParseError, previewPatch, stripBom } from "../edit";
+import { getFileSnapshotStore } from "../edit/file-snapshot-store";
+import { namespaceSessionId as namespacePythonSessionId } from "../eval/py";
 import {
 	disposeKernelSessionsByOwner,
 	executePython as executePythonCommand,
@@ -209,7 +212,11 @@ import { YieldQueue } from "./yield-queue";
 /** Session-specific events that extend the core AgentEvent */
 export type AgentSessionEvent =
 	| AgentEvent
-	| { type: "auto_compaction_start"; reason: "threshold" | "overflow" | "idle"; action: "context-full" | "handoff" }
+	| {
+			type: "auto_compaction_start";
+			reason: "threshold" | "overflow" | "idle" | "incomplete";
+			action: "context-full" | "handoff";
+	  }
 	| {
 			type: "auto_compaction_end";
 			action: "context-full" | "handoff";
@@ -738,6 +745,7 @@ export class AgentSession {
 	readonly sessionManager: SessionManager;
 	readonly settings: Settings;
 	readonly yieldQueue: YieldQueue;
+	fileSnapshotStore?: InMemorySnapshotStore;
 
 	#powerAssertion: MacOSPowerAssertion | undefined;
 
@@ -3631,9 +3639,17 @@ export class AgentSession {
 		const sessionOnResponse = this.#onResponse;
 		const sessionMetadata = this.agent.metadataForProvider(provider);
 		const sessionOnSseEvent = this.#onSseEvent;
-		if (!sessionOnPayload && !sessionOnResponse && !sessionMetadata && !sessionOnSseEvent) return options;
-
-		const preparedOptions: SimpleStreamOptions = { ...options };
+		const openrouterRoutingPreset =
+			provider === "openrouter" ? this.settings.get("providers.openrouterVariant") : "default";
+		const openrouterVariant =
+			openrouterRoutingPreset !== "default" && options.openrouterVariant === undefined
+				? openrouterRoutingPreset
+				: undefined;
+		if (!sessionOnPayload && !sessionOnResponse && !sessionMetadata && !sessionOnSseEvent && !openrouterVariant)
+			return options;
+
+		const preparedOptions: SimpleStreamOptions =
+			openrouterVariant === undefined ? { ...options } : { ...options, openrouterVariant };
 
 		// Stamp session metadata (e.g. user_id={session_id}) onto direct-call requests so
 		// they share the same session bucket as Agent.prompt-routed requests on Anthropic
@@ -3758,6 +3774,10 @@ export class AgentSession {
 		this.#planReferencePath = path;
 	}
 
+	getPlanReferencePath(): string {
+		return this.#planReferencePath;
+	}
+
 	get clientBridge(): ClientBridge | undefined {
 		return this.#clientBridge;
 	}
@@ -4144,6 +4164,7 @@ export class AgentSession {
 				const fileMentionMessages = await generateFileMentionMessages(fileMentions, this.sessionManager.getCwd(), {
 					autoResizeImages: this.settings.get("images.autoResize"),
 					useHashLines: resolveFileDisplayMode(this).hashLines,
+					snapshotStore: getFileSnapshotStore(this),
 				});
 				messages.push(...fileMentionMessages);
 			}
@@ -5575,6 +5596,11 @@ export class AgentSession {
 					initiatorOverride: "agent",
 					metadata: this.agent.metadataForProvider(model.provider),
 					telemetry: resolveTelemetry(this.agent.telemetry, this.sessionId),
+					// Honor the user's /model thinking selection on the handoff
+					// path. Clamped per-model inside generateHandoff via
+					// resolveCompactionEffort so unsupported-effort models don't
+					// trip requireSupportedEffort.
+					thinkingLevel: this.thinkingLevel,
 				},
 				handoffSignal,
 			);
@@ -5645,10 +5671,14 @@ export class AgentSession {
 	 * Check if context maintenance or promotion is needed and run it.
 	 * Called after agent_end and before prompt submission.
 	 *
-	 * Three cases (in order):
-	 * 1. Overflow + promotion: promote to larger model, retry without maintenance
-	 * 2. Overflow + no promotion target: run context maintenance, auto-retry on same model
-	 * 3. Threshold: Context over threshold, run context maintenance (no auto-retry)
+	 * Four cases (in order):
+	 * 1. Input overflow + promotion: promote to larger model, retry without maintenance.
+	 * 2. Input overflow + no promotion target: run context maintenance, auto-retry on same model.
+	 * 3. Output incomplete (stopReason === "length", e.g. `response.incomplete`): the
+	 *    model burned its output budget without producing an actionable deliverable
+	 *    (reasoning-only or truncated). Drop the dead turn, try promotion, otherwise
+	 *    run compaction/handoff and retry.
+	 * 4. Threshold: context over threshold, run context maintenance (no auto-retry).
 	 *
 	 * @param assistantMessage The assistant message to check
 	 * @param skipAbortedCheck If false, include aborted messages (for pre-prompt check). Default: true
@@ -5707,10 +5737,49 @@ export class AgentSession {
 			}
 			return false;
 		}
+
+		// Case 3: Output-side incomplete — `response.incomplete` from OpenAI Responses
+		// (and Codex) maps to stopReason === "length". The model burned its
+		// `max_output_tokens` budget on reasoning/text and emitted no actionable
+		// deliverable. Same recovery class as overflow: promotion if available,
+		// otherwise compaction/handoff. Unlike overflow, the *input* is fine, so we
+		// allow the handoff strategy to actually run.
+		if (sameModel && !errorIsFromBeforeCompaction && assistantMessage.stopReason === "length") {
+			const messages = this.agent.state.messages;
+			if (messages.length > 0 && messages[messages.length - 1].role === "assistant") {
+				this.agent.replaceMessages(messages.slice(0, -1));
+			}
+
+			const promoted = await this.#tryContextPromotion(assistantMessage);
+			if (promoted) {
+				logger.debug("Context promotion triggered by response.incomplete (length stop)", {
+					from: `${assistantMessage.provider}/${assistantMessage.model}`,
+				});
+				this.#scheduleAgentContinue({ delayMs: 100, generation });
+				return false;
+			}
+
+			const incompleteCompactionSettings = this.settings.getGroup("compaction");
+			if (incompleteCompactionSettings.enabled && incompleteCompactionSettings.strategy !== "off") {
+				logger.debug("Compaction triggered by response.incomplete (length stop, no promotion target)", {
+					model: `${assistantMessage.provider}/${assistantMessage.model}`,
+					strategy: incompleteCompactionSettings.strategy,
+				});
+				await this.#runAutoCompaction("incomplete", true, false, allowDefer);
+			} else {
+				// Neither promotion nor compaction is available — surface the dead-end so
+				// the user understands why the turn yielded with nothing.
+				logger.warn("response.incomplete with no recovery path (promotion + compaction both unavailable)", {
+					model: `${assistantMessage.provider}/${assistantMessage.model}`,
+				});
+			}
+			return false;
+		}
+
 		const compactionSettings = this.settings.getGroup("compaction");
 		if (!compactionSettings.enabled || compactionSettings.strategy === "off") return false;
 
-		// Case 2: Threshold - turn succeeded but context is getting large
+		// Case 4: Threshold - turn succeeded but context is getting large
 		// Skip if this was an error (non-overflow errors don't have usage data)
 		if (assistantMessage.stopReason === "error") return false;
 		const pruneResult = await this.#pruneToolOutputs();
@@ -6345,6 +6414,11 @@ export class AgentSession {
 					metadata: this.agent.metadataForProvider(candidate.provider),
 					convertToLlm,
 					telemetry,
+					// Honor the user's /model thinking selection (incl. `off`) on
+					// the manual `/compact` path. Clamped per-model inside compact()
+					// via resolveCompactionEffort so unsupported-effort models
+					// (xai-oauth/grok-build) don't trip requireSupportedEffort.
+					thinkingLevel: this.thinkingLevel,
 				});
 			} catch (error) {
 				if (!this.#isCompactionAuthFailure(error)) {
@@ -6428,7 +6502,7 @@ export class AgentSession {
 	 * @returns true when a deferred handoff was scheduled. Inline runs always return false.
 	 */
 	async #runAutoCompaction(
-		reason: "overflow" | "threshold" | "idle",
+		reason: "overflow" | "threshold" | "idle" | "incomplete",
 		willRetry: boolean,
 		deferred = false,
 		allowDefer = true,
@@ -6437,10 +6511,14 @@ export class AgentSession {
 		if (compactionSettings.strategy === "off") return false;
 		if (reason !== "idle" && !compactionSettings.enabled) return false;
 		const generation = this.#promptGeneration;
+		// "overflow" and "incomplete" force inline execution because they are recovery
+		// paths the caller wants resolved before scheduling the next turn. "idle" is
+		// triggered by the idle loop and does its own scheduling.
 		if (
 			!deferred &&
 			allowDefer &&
 			reason !== "overflow" &&
+			reason !== "incomplete" &&
 			reason !== "idle" &&
 			compactionSettings.strategy === "handoff"
 		) {
@@ -6455,6 +6533,9 @@ export class AgentSession {
 			return true;
 		}
 
+		// "overflow" forces context-full because the input itself is broken — a handoff
+		// LLM call would hit the same overflow. "incomplete" is an output-side problem,
+		// so a handoff request on the existing context is still viable.
 		let action: "context-full" | "handoff" =
 			compactionSettings.strategy === "handoff" && reason !== "overflow" ? "handoff" : "context-full";
 		await this.#emitSessionEvent({ type: "auto_compaction_start", reason, action });
@@ -6617,6 +6698,11 @@ export class AgentSession {
 								initiatorOverride: "agent",
 								convertToLlm,
 								telemetry,
+								// Honor the user's /model thinking selection on the
+								// auto-compaction path — the most-fired compaction
+								// site. Clamped per-model inside compact() via
+								// resolveCompactionEffort.
+								thinkingLevel: this.thinkingLevel,
 							});
 							break;
 						} catch (error) {
@@ -6750,8 +6836,18 @@ export class AgentSession {
 			if (willRetry) {
 				const messages = this.agent.state.messages;
 				const lastMsg = messages[messages.length - 1];
-				if (lastMsg?.role === "assistant" && (lastMsg as AssistantMessage).stopReason === "error") {
-					this.agent.replaceMessages(messages.slice(0, -1));
+				if (lastMsg?.role === "assistant") {
+					const lastAssistant = lastMsg as AssistantMessage;
+					// Drop the prior turn before retry when it carries no actionable deliverable:
+					// - "error": failure was kept in history but must not re-enter the next turn's prompt.
+					// - reason === "incomplete" && stopReason === "length": truncated output (typically
+					//   reasoning-only) — re-running it produces the same dead-end.
+					const shouldDrop =
+						lastAssistant.stopReason === "error" ||
+						(reason === "incomplete" && lastAssistant.stopReason === "length");
+					if (shouldDrop) {
+						this.agent.replaceMessages(messages.slice(0, -1));
+					}
 				}
 
 				this.#scheduleAgentContinue({ delayMs: 100, generation });
@@ -6785,7 +6881,9 @@ export class AgentSession {
 				errorMessage:
 					reason === "overflow"
 						? `Context overflow recovery failed: ${errorMessage}`
-						: `Auto-compaction failed: ${errorMessage}`,
+						: reason === "incomplete"
+							? `Incomplete response recovery failed: ${errorMessage}`
+							: `Auto-compaction failed: ${errorMessage}`,
 			});
 		} finally {
 			if (this.#autoCompactionAbortController === autoCompactionAbortController) {
@@ -7494,7 +7592,7 @@ export class AgentSession {
 				});
 			const result = await executePythonCommand(code, {
 				cwd,
-				sessionId,
+				sessionId: namespacePythonSessionId(sessionId),
 				kernelOwnerId: this.#evalKernelOwnerId,
 				kernelMode: this.settings.get("python.kernelMode"),
 				onChunk,

package/src/system-prompt.ts

@@ -11,6 +11,7 @@ import { systemPromptCapability } from "./capability/system-prompt";
 import type { SkillsSettings } from "./config/settings";
 import { type ContextFile, loadCapability, type SystemPrompt as SystemPromptFile } from "./discovery";
 import { loadSkills, type Skill } from "./extensibility/skills";
+import { hasObsidian } from "./internal-urls/vault-protocol";
 import customSystemPromptTemplate from "./prompts/system/custom-system-prompt.md" with { type: "text" };
 import projectPromptTemplate from "./prompts/system/project-prompt.md" with { type: "text" };
 import systemPromptTemplate from "./prompts/system/system-prompt.md" with { type: "text" };
@@ -569,6 +570,7 @@ export async function buildSystemPrompt(options: BuildSystemPromptOptions = {}):
 		mcpDiscoveryServerSummaries,
 		eagerTasks,
 		secretsEnabled,
+		hasObsidian: hasObsidian(),
 	};
 	const rendered = prompt.render(resolvedCustomPrompt ? customSystemPromptTemplate : systemPromptTemplate, data);
 	const systemPrompt = [rendered];

package/src/tools/ast-edit.ts

@@ -1,11 +1,13 @@
 import * as path from "node:path";
-import { computeFileHash, formatHashlineHeader } from "@oh-my-pi/hashline";
+import { formatHashlineHeader } from "@oh-my-pi/hashline";
 import type { AgentTool, AgentToolContext, AgentToolResult, AgentToolUpdateCallback } from "@oh-my-pi/pi-agent-core";
 import { type AstReplaceChange, type AstReplaceFileChange, astEdit } from "@oh-my-pi/pi-natives";
 import type { Component } from "@oh-my-pi/pi-tui";
 import { Text } from "@oh-my-pi/pi-tui";
 import { $envpos, prompt, untilAborted } from "@oh-my-pi/pi-utils";
 import * as z from "zod/v4";
+import { getFileSnapshotStore } from "../edit/file-snapshot-store";
+import { normalizeToLF } from "../edit/normalize";
 import type { RenderResultOptions } from "../extensibility/custom-tools/types";
 import type { Theme } from "../modes/theme/theme";
 import astEditDescription from "../prompts/tools/ast-edit.md" with { type: "text" };
@@ -281,14 +283,15 @@ export class AstEditTool implements AgentTool<typeof astEditSchema, AstEditToolD
 			}
 
 			const useHashLines = resolveFileDisplayMode(this.session).hashLines;
-			const hashContexts = new Map<string, { fileHash: string }>();
+			const hashContexts = new Map<string, { tag: string }>();
 			if (useHashLines) {
+				const snapshotStore = getFileSnapshotStore(this.session);
 				for (const relativePath of fileList) {
 					const absolutePath = path.resolve(this.session.cwd, relativePath);
 					try {
-						const fullText = await Bun.file(absolutePath).text();
-						const fileHash = computeFileHash(fullText);
-						hashContexts.set(relativePath, { fileHash });
+						const fullText = normalizeToLF(await Bun.file(absolutePath).text());
+						const tag = snapshotStore.recordContiguous(absolutePath, 1, fullText.split("\n"), { fullText });
+						hashContexts.set(relativePath, { tag });
 					} catch {
 						// Best-effort: if a file disappears between ast-edit and rendering, emit plain line output.
 					}
@@ -326,7 +329,7 @@ export class AstEditTool implements AgentTool<typeof astEditSchema, AstEditToolD
 					const rendered = renderChangesForFile(relativePath);
 					const count = fileReplacementCounts.get(relativePath) ?? 0;
 					const hashContext = hashContexts.get(relativePath);
-					const hashSuffix = hashContext ? `#${hashContext.fileHash}` : "";
+					const hashSuffix = hashContext ? `#${hashContext.tag}` : "";
 					return {
 						headerSuffix: `${hashSuffix} (${formatCount("replacement", count)})`,
 						modelLines: rendered.model,
@@ -346,7 +349,7 @@ export class AstEditTool implements AgentTool<typeof astEditSchema, AstEditToolD
 					}
 					const hashContext = hashContexts.get(relativePath);
 					if (hashContext) {
-						outputLines.push(formatHashlineHeader(relativePath, hashContext.fileHash));
+						outputLines.push(formatHashlineHeader(relativePath, hashContext.tag));
 					}
 					outputLines.push(...rendered.model);
 					displayLines.push(...rendered.display);

package/src/tools/ast-grep.ts

@@ -1,5 +1,7 @@
+import { constants } from "node:fs";
+import { access } from "node:fs/promises";
 import * as path from "node:path";
-import { computeFileHash, formatHashlineHeader } from "@oh-my-pi/hashline";
+import { formatHashlineHeader } from "@oh-my-pi/hashline";
 import type { AgentTool, AgentToolContext, AgentToolResult, AgentToolUpdateCallback } from "@oh-my-pi/pi-agent-core";
 import { type AstFindMatch, astGrep } from "@oh-my-pi/pi-natives";
 import type { Component } from "@oh-my-pi/pi-tui";
@@ -219,14 +221,14 @@ export class AstGrepTool implements AgentTool<typeof astGrepSchema, AstGrepToolD
 			}
 
 			const useHashLines = resolveFileDisplayMode(this.session).hashLines;
-			const hashContexts = new Map<string, { absolutePath: string; fileHash: string }>();
+			const hashContexts = new Map<string, { absolutePath: string; tag?: string }>();
+			const snapshotStore = useHashLines ? getFileSnapshotStore(this.session) : undefined;
 			if (useHashLines) {
 				for (const relativePath of fileList) {
 					const absolutePath = path.resolve(this.session.cwd, relativePath);
 					try {
-						const fullText = await Bun.file(absolutePath).text();
-						const fileHash = computeFileHash(fullText);
-						hashContexts.set(relativePath, { absolutePath, fileHash });
+						await access(absolutePath, constants.R_OK);
+						hashContexts.set(relativePath, { absolutePath });
 					} catch {
 						// Best-effort: if a file disappears between ast-grep and rendering, emit plain line output.
 					}
@@ -268,9 +270,8 @@ export class AstGrepTool implements AgentTool<typeof astGrepSchema, AstGrepToolD
 					fileMatchCounts.set(relativePath, (fileMatchCounts.get(relativePath) ?? 0) + 1);
 				}
 				if (hashContext && cacheEntries.length > 0) {
-					getFileSnapshotStore(this.session).recordSparse(hashContext.absolutePath, cacheEntries, {
-						fileHash: hashContext.fileHash,
-					});
+					const tag = snapshotStore?.recordSparse(hashContext.absolutePath, cacheEntries);
+					if (tag) hashContext.tag = tag;
 				}
 				return { model: modelOut, display: displayOut };
 			};
@@ -282,7 +283,7 @@ export class AstGrepTool implements AgentTool<typeof astGrepSchema, AstGrepToolD
 					return {
 						modelLines: rendered.model,
 						displayLines: rendered.display,
-						headerSuffix: hashContext ? `#${hashContext.fileHash}` : "",
+						headerSuffix: hashContext?.tag ? `#${hashContext.tag}` : "",
 						skip: rendered.model.length === 0,
 					};
 				});
@@ -297,8 +298,8 @@ export class AstGrepTool implements AgentTool<typeof astGrepSchema, AstGrepToolD
 						displayLines.push("");
 					}
 					const hashContext = hashContexts.get(relativePath);
-					if (hashContext) {
-						outputLines.push(formatHashlineHeader(relativePath, hashContext.fileHash));
+					if (hashContext?.tag) {
+						outputLines.push(formatHashlineHeader(relativePath, hashContext.tag));
 					}
 					outputLines.push(...rendered.model);
 					displayLines.push(...rendered.display);

package/src/tools/eval.ts

@@ -14,6 +14,7 @@ import { getMarkdownTheme, type Theme } from "../modes/theme/theme";
 import evalDescription from "../prompts/tools/eval.md" with { type: "text" };
 import { DEFAULT_MAX_BYTES, OutputSink, type OutputSummary, TailBuffer } from "../session/streaming-output";
 import { getTreeBranch, getTreeContinuePrefix, renderCodeCell } from "../tui";
+import { formatDimensionNote, resizeImage } from "../utils/image-resize";
 import { resolveEvalBackends, type ToolSession } from ".";
 import { truncateForPrompt } from "./approval";
 import {
@@ -403,6 +404,7 @@ export class EvalTool implements AgentTool<typeof evalSchema> {
 
 					const cellStatusEvents: EvalStatusEvent[] = [];
 					const cellDisplayOutputs: EvalDisplayOutput[] = [];
+					const cellImageNotes: string[] = [];
 					let cellHasMarkdown = false;
 					for (const output of result.displayOutputs) {
 						if (output.type === "json") {
@@ -410,8 +412,26 @@ export class EvalTool implements AgentTool<typeof evalSchema> {
 							cellDisplayOutputs.push(output);
 						}
 						if (output.type === "image") {
-							images.push({ type: "image", data: output.data, mimeType: output.mimeType });
-							cellDisplayOutputs.push(output);
+							const resized = await resizeImage({
+								type: "image",
+								data: output.data,
+								mimeType: output.mimeType,
+							});
+							const image: ImageContent = {
+								type: "image",
+								data: resized.data,
+								mimeType: resized.mimeType,
+							};
+							images.push(image);
+							cellDisplayOutputs.push({
+								type: "image",
+								data: image.data,
+								mimeType: image.mimeType,
+							});
+							const dimensionNote = formatDimensionNote(resized);
+							if (dimensionNote) {
+								cellImageNotes.push(`display image ${cellImageNotes.length + 1}: ${dimensionNote}`);
+							}
 						}
 						if (output.type === "status") {
 							statusEvents.push(output.event);
@@ -423,9 +443,14 @@ export class EvalTool implements AgentTool<typeof evalSchema> {
 					}
 
 					const stdoutTrimmed = result.output.trim();
+					const imageText = cellImageNotes.join("\n");
 					const displayText = formatDisplayOutputsForText(cellDisplayOutputs);
+					const visibleDisplayText =
+						displayText && imageText ? `${displayText}\n\n${imageText}` : displayText || imageText;
 					const cellOutput =
-						stdoutTrimmed && displayText ? `${stdoutTrimmed}\n\n${displayText}` : stdoutTrimmed || displayText;
+						stdoutTrimmed && visibleDisplayText
+							? `${stdoutTrimmed}\n\n${visibleDisplayText}`
+							: stdoutTrimmed || visibleDisplayText;
 					cellResult.output = cellOutput;
 					cellResult.exitCode = result.exitCode;
 					cellResult.durationMs = durationMs;

package/src/tools/fetch.ts

@@ -562,9 +562,22 @@ function parseFeedToMarkdown(content: string, maxItems = 10): string {
 }
 
 /**
- * Render HTML to markdown using Parallel, jina, trafilatura, lynx (in order of preference)
+ * Cap on any single remote reader-mode request (Parallel, Jina) so a stalled
+ * remote endpoint cannot consume the whole reader-mode budget and starve the
+ * local fallback renderers (trafilatura, lynx, native). See #1449.
  */
-async function renderHtmlToText(
+const REMOTE_READER_MAX_MS = 10_000;
+
+/**
+ * Render HTML to markdown using Parallel, jina, trafilatura, lynx, then the
+ * in-process native converter. The overall `timeout` budget bounds the call,
+ * but remote reader requests are additionally capped at `REMOTE_READER_MAX_MS`
+ * so that a hung remote endpoint cannot prevent local fallbacks from running.
+ * Only a real `userSignal` cancellation aborts the chain — remote per-attempt
+ * timeouts and the overall reader-mode timeout still allow later renderers
+ * (especially the purely-local native converter) to be tried.
+ */
+export async function renderHtmlToText(
 	url: string,
 	html: string,
 	timeout: number,
@@ -572,14 +585,15 @@ async function renderHtmlToText(
 	userSignal: AbortSignal | undefined,
 	storage: AgentStorage | null,
 ): Promise<{ content: string; ok: boolean; method: string }> {
-	const signal = ptree.combineSignals(userSignal, timeout * 1000);
+	const overallSignal = ptree.combineSignals(userSignal, timeout * 1000);
 	const execOptions = {
 		mode: "group" as const,
 		allowNonZero: true,
 		allowAbort: true,
 		stderr: "full" as const,
-		signal,
+		signal: overallSignal,
 	};
+	const remoteBudgetMs = Math.min(timeout * 1000, REMOTE_READER_MAX_MS);
 
 	// Try Parallel extract first when credentials are configured
 	if (settings.get("providers.parallelFetch") && findParallelApiKey(storage)) {
@@ -590,7 +604,7 @@ async function renderHtmlToText(
 					objective: "Extract the main content",
 					excerpts: true,
 					fullContent: false,
-					signal,
+					signal: ptree.combineSignals(userSignal, remoteBudgetMs),
 				},
 				storage,
 			);
@@ -602,17 +616,18 @@ async function renderHtmlToText(
 				}
 			}
 		} catch {
-			// Parallel extract failed, continue to next method
-			signal?.throwIfAborted();
+			// Parallel extract failed or stalled; honour real cancellation only.
+			userSignal?.throwIfAborted();
 		}
 	}
 
-	// Try jina first (reader API)
+	// Try jina reader API with its own sub-budget so a stall cannot starve
+	// later fallbacks (#1449).
 	try {
 		const jinaUrl = `https://r.jina.ai/${url}`;
 		const response = await fetch(jinaUrl, {
 			headers: { Accept: "text/markdown" },
-			signal,
+			signal: ptree.combineSignals(userSignal, remoteBudgetMs),
 		});
 		if (response.ok) {
 			const content = await response.text();
@@ -621,37 +636,50 @@ async function renderHtmlToText(
 			}
 		}
 	} catch {
-		// Jina failed, continue to next method
-		signal?.throwIfAborted();
+		// Jina failed or stalled; honour real cancellation only.
+		userSignal?.throwIfAborted();
 	}
 
 	// Try trafilatura (auto-install via uv/pip)
-	const trafilatura = await ensureTool("trafilatura", { signal, silent: true });
-	if (trafilatura) {
-		const result = await ptree.exec([trafilatura, "-u", url, "--output-format", "markdown"], execOptions);
-		if (result.ok && result.stdout.trim().length > 100) {
-			return { content: result.stdout, ok: true, method: "trafilatura" };
+	try {
+		const trafilatura = await ensureTool("trafilatura", { signal: overallSignal, silent: true });
+		if (trafilatura) {
+			const result = await ptree.exec([trafilatura, "-u", url, "--output-format", "markdown"], execOptions);
+			if (result.ok && result.stdout.trim().length > 100) {
+				return { content: result.stdout, ok: true, method: "trafilatura" };
+			}
 		}
+	} catch {
+		// trafilatura unavailable or stalled; continue to next method.
+		userSignal?.throwIfAborted();
 	}
 
 	// Try lynx (can't auto-install, system package)
-	const lynx = hasCommand("lynx");
-	if (lynx) {
-		const result = await ptree.exec(["lynx", "-dump", "-nolist", "-width", "250", url], execOptions);
-		if (result.ok) {
-			return { content: result.stdout, ok: true, method: "lynx" };
+	try {
+		const lynx = hasCommand("lynx");
+		if (lynx) {
+			const result = await ptree.exec(["lynx", "-dump", "-nolist", "-width", "250", url], execOptions);
+			if (result.ok) {
+				return { content: result.stdout, ok: true, method: "lynx" };
+			}
 		}
+	} catch {
+		// lynx failed or stalled; continue to native converter.
+		userSignal?.throwIfAborted();
 	}
 
-	// Fall back to native converter (fastest, no network/subprocess)
+	// Fall back to native converter (purely local, no network/subprocess).
+	// Always attempted: even if remote renderers and subprocesses were aborted
+	// by the overall reader-mode timeout, this still works on already-loaded
+	// HTML (#1449).
 	try {
 		const content = await htmlToMarkdown(html, { cleanContent: true });
 		if (content.trim().length > 100 && !isLowQualityOutput(content)) {
 			return { content, ok: true, method: "native" };
 		}
 	} catch {
-		// Native converter failed, continue to next method
-		signal?.throwIfAborted();
+		// Native converter failed; nothing else to try.
+		userSignal?.throwIfAborted();
 	}
 	return { content: "", ok: false, method: "none" };
 }