@diegopetrucci/pi-extensions 0.1.44 → 0.1.46

package/extensions/librarian/index.ts

@@ -629,20 +629,56 @@ function buildUserPrompt(query: string, repos: string[], owners: string[], maxSe
 	return `Task: locate and cite exact GitHub code locations that answer the query.\n\nQuery: ${query}\nRepository filters: ${repos.length ? repos.join(", ") : "(none)"}\nOwner filters: ${owners.length ? owners.join(", ") : "(none)"}\nMax search results per gh search call: ${maxSearchResults}\nLocal checkout cache: ${cache.mode === "enabled" ? `enabled at ${cache.root}` : "disabled"}\nCache decision: ${cache.decisionReason}\n\nRespond directly with concise, citation-heavy findings. Always pass --limit ${maxSearchResults} to gh search code unless a narrower command is clearly better.`;
 }
 
-function extractLastAssistantText(messages: unknown[]): string {
+type AssistantLikeMessage = {
+	role?: string;
+	content?: unknown;
+	stopReason?: unknown;
+	errorMessage?: unknown;
+};
+
+function getLastAssistantMessage(messages: unknown[]): AssistantLikeMessage | undefined {
 	for (let i = messages.length - 1; i >= 0; i--) {
-		const message = messages[i] as { role?: string; content?: unknown };
-		if (message?.role !== "assistant" || !Array.isArray(message.content)) continue;
-		const parts: string[] = [];
-		for (const part of message.content) {
-			if (part && typeof part === "object" && (part as { type?: string }).type === "text") {
-				const text = (part as { text?: unknown }).text;
-				if (typeof text === "string") parts.push(text);
-			}
+		const message = messages[i] as AssistantLikeMessage;
+		if (message?.role === "assistant") return message;
+	}
+	return undefined;
+}
+
+function extractLastAssistantText(messages: unknown[]): string {
+	const message = getLastAssistantMessage(messages);
+	if (!message || !Array.isArray(message.content)) return "";
+	const parts: string[] = [];
+	for (const part of message.content) {
+		if (part && typeof part === "object" && (part as { type?: string }).type === "text") {
+			const text = (part as { text?: unknown }).text;
+			if (typeof text === "string") parts.push(text);
 		}
-		if (parts.length) return parts.join("").trim();
 	}
-	return "";
+	return parts.join("").trim();
+}
+
+function describeNoAnswerReason(message: AssistantLikeMessage | undefined, turns: number): string {
+	if (!message) return "the internal subagent produced no assistant message";
+	if (message.stopReason === "error") {
+		const errorMessage = typeof message.errorMessage === "string" && message.errorMessage.trim()
+			? message.errorMessage.trim()
+			: "provider/model error";
+		return `the internal subagent stopped with an error: ${errorMessage}`;
+	}
+	if (message.stopReason === "aborted") return "the internal subagent was aborted before producing an answer";
+	if (turns >= MAX_TURNS) return `the internal subagent reached the ${MAX_TURNS}-turn budget without producing a final answer`;
+	const stopReason = typeof message.stopReason === "string" && message.stopReason.trim()
+		? ` (stopReason: ${message.stopReason.trim()})`
+		: "";
+	return `the internal subagent completed without final assistant text${stopReason}`;
+}
+
+function formatInternalFailure(details: LibrarianDetails, reason: string): string {
+	return [
+		`Internal librarian run failed: ${reason}.`,
+		`Diagnostics: status=${details.status}; model=${details.model.modelRef}; thinking=${details.model.thinkingLevel}; modelSelection=${details.model.autoSelected ? "auto" : "configured"}; turns=${details.turns}; toolCalls=${details.toolCalls.length}.`,
+		"No answer was produced. This is not a reliable \"no results found\" signal; retry later or configure a different Librarian model with /librarian-config.",
+	].join("\n");
 }
 
 function formatToolCall(call: ToolCall): string {
@@ -847,6 +883,13 @@ export default function librarianExtension(pi: ExtensionAPI) {
 		},
 	});
 
+	pi.on("tool_result", async (event) => {
+		if (event.toolName !== "librarian") return undefined;
+		const details = event.details as LibrarianDetails | undefined;
+		if (details?.status !== "error") return undefined;
+		return { isError: true };
+	});
+
 	pi.registerTool({
 		name: "librarian",
 		label: "Librarian",
@@ -1034,9 +1077,20 @@ export default function librarianExtension(pi: ExtensionAPI) {
 					await Promise.race([promptPromise, timeoutPromise]);
 				}
 
+				const lastAssistant = session ? getLastAssistantMessage(session.state.messages) : undefined;
 				const answer = session ? extractLastAssistantText(session.state.messages) : "";
-				lastContent = answer || (aborted ? "Aborted" : "(no output)");
-				details.status = aborted ? "aborted" : "done";
+				if (answer) {
+					lastContent = answer;
+					details.status = "done";
+				} else if (aborted) {
+					lastContent = "Aborted";
+					details.status = "aborted";
+				} else {
+					details.status = "error";
+					const reason = describeNoAnswerReason(lastAssistant, details.turns);
+					lastContent = formatInternalFailure(details, reason);
+					details.error = reason;
+				}
 				details.endedAt = Date.now();
 				emit(true);
 

package/extensions/oracle/README.md

@@ -27,7 +27,7 @@ By default, the extension:
 4. tries a provider-specific hardcoded priority list first
 5. falls back to a heuristic that favors stronger tiers like `opus`, `pro`, newer versions, and penalizes `mini`, `flash`, `haiku`, `spark`, etc.
 
-The hardcoded rankings now cover pi's built-in provider set, including Together; see the provider matrix for the current provider-by-provider top picks.
+The hardcoded rankings now cover pi's built-in provider set, including Claude Fable on Anthropic-compatible providers and Together; see the provider matrix for the current provider-by-provider top picks.
 
 If no reasoning model exists on the current provider, it falls back to the best available model on that provider.
 

package/extensions/oracle/index.ts

@@ -75,6 +75,7 @@ const ORACLE_CONFIG_FILE = "oracle.json";
 
 const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
 	"amazon-bedrock": [
+		"claude-fable-5",
 		"claude-opus-4-8",
 		"claude-opus-4-7",
 		"claude-opus-4-6",
@@ -92,6 +93,7 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
 		"zai.glm-5",
 	],
 	anthropic: [
+		"claude-fable-5",
 		"claude-opus-4-8",
 		"claude-opus-4.8",
 		"claude-opus-4-7",
@@ -132,6 +134,7 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
 	],
 	cerebras: ["gpt-oss-120b", "zai-glm-4.7", "llama3.1-8b"],
 	"cloudflare-ai-gateway": [
+		"claude-fable-5",
 		"claude-opus-4-7",
 		"claude-opus-4-6",
 		"claude-opus-4-5",
@@ -159,6 +162,7 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
 		"accounts/fireworks/models/gpt-oss-120b",
 	],
 	"github-copilot": [
+		"claude-fable-5",
 		"claude-opus-4.8",
 		"claude-opus-4.7",
 		"claude-opus-4.6",
@@ -257,6 +261,7 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
 		"big-pickle",
 	],
 	opencode: [
+		"claude-fable-5",
 		"gpt-5.5-pro",
 		"gpt-5.5",
 		"gpt-5.4-pro",
@@ -289,6 +294,8 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
 		"kimi-k2.5",
 	],
 	openrouter: [
+		"anthropic/claude-fable-5",
+		"~anthropic/claude-fable-latest",
 		"anthropic/claude-opus-4.8",
 		"anthropic/claude-opus-4.8-fast",
 		"anthropic/claude-opus-4.7",
@@ -327,6 +334,7 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
 		"MiniMaxAI/MiniMax-M2.5",
 	],
 	"vercel-ai-gateway": [
+		"anthropic/claude-fable-5",
 		"anthropic/claude-opus-4.8",
 		"anthropic/claude-opus-4.7",
 		"anthropic/claude-opus-4.6",
@@ -549,6 +557,26 @@ function extractTextFromContent(content: unknown): string {
 	return parts.join("\n\n").trim();
 }
 
+// Errors that typically resolve on retry: provider overload, rate limiting,
+// transient 5xx, gateway/network failures, and request timeouts. Keep this list
+// pattern-based so we recognize variants across providers without enumerating them.
+const TRANSIENT_ERROR_PATTERN =
+	/\b(overload(?:ed)?|rate[ _-]?limit(?:ed)?|too many requests|429|500|502|503|504|bad gateway|service unavailable|gateway timeout|temporarily unavailable|timed?[ _-]?out|timeout|econnreset|econnrefused|etimedout|enetunreach|socket hang up|fetch failed)\b/i;
+
+function isTransientErrorMessage(message: string): boolean {
+	return TRANSIENT_ERROR_PATTERN.test(message);
+}
+
+function formatOracleModelError(stopReason: "error" | "aborted", errorMessage: string | undefined): string {
+	const trimmed = errorMessage?.trim();
+	if (stopReason === "aborted") {
+		return trimmed ? `Oracle model turn aborted: ${trimmed}` : "Oracle model turn aborted.";
+	}
+	if (!trimmed) return "Oracle model error (no detail provided by provider).";
+	const base = `Oracle model error: ${trimmed}`;
+	return isTransientErrorMessage(trimmed) ? `${base} (transient; retry may succeed)` : base;
+}
+
 function parseVersionScore(text: string): number {
 	const matches = text.match(/\d+(?:\.\d+){0,2}/g) ?? [];
 	let best = 0;
@@ -809,6 +837,9 @@ async function runOracle(
 	let finalOutput = "";
 	let stderr = "";
 
+	let lastStopReason: string | undefined;
+	let lastErrorMessage: string | undefined;
+
 	const details: OracleDetails = {
 		...selection,
 		includeBash,
@@ -880,6 +911,11 @@ async function runOracle(
 				if (text) finalOutput = text;
 				currentText = "";
 
+				const stopReason = event.message.stopReason;
+				lastStopReason = typeof stopReason === "string" ? stopReason : undefined;
+				const errorMessageField = event.message.errorMessage;
+				lastErrorMessage = typeof errorMessageField === "string" ? errorMessageField : undefined;
+
 				const messageUsage = event.message.usage;
 				if (messageUsage) {
 					usage.turns += 1;
@@ -938,6 +974,19 @@ async function runOracle(
 		return { ok: false, error: "Oracle was aborted.", details };
 	}
 
+	// The pi subprocess in `--mode json` does not promote an errored assistant
+	// turn to a non-zero exit code or stderr; the error is only carried on the
+	// streamed assistant message via stopReason/errorMessage. Surface that here
+	// so callers can distinguish transient provider errors from a genuinely
+	// empty response.
+	if (lastStopReason === "error" || lastStopReason === "aborted") {
+		return {
+			ok: false,
+			error: formatOracleModelError(lastStopReason, lastErrorMessage),
+			details,
+		};
+	}
+
 	if (exitCode !== 0) {
 		return {
 			ok: false,

package/extensions/oracle/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@diegopetrucci/pi-oracle",
-  "version": "0.1.11",
+  "version": "0.1.13",
   "description": "An Amp-style oracle extension for pi that consults the strongest reasoning model on your current provider.",
   "keywords": [
     "pi-package",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@diegopetrucci/pi-extensions",
-  "version": "0.1.44",
+  "version": "0.1.46",
   "description": "A collection of pi extensions and skills for annotation UIs, context management, workflow audits, review-comment triage, notifications, brrr push alerts, safety guards, GitHub research, repo-local knowledge, todos, tool rendering, model/provider helpers, and Xiaohei-style article illustrations.",
   "keywords": [
     "pi-package",
@@ -13,6 +13,11 @@
     "type": "git",
     "url": "git+https://github.com/diegopetrucci/pi-extensions.git"
   },
+  "scripts": {
+    "ci": "npm run preflight:install-state && npm run typecheck",
+    "preflight:install-state": "node scripts/check-install-state.mjs",
+    "typecheck": "npx tsc --noEmit --module NodeNext --moduleResolution NodeNext --target ES2022 --skipLibCheck $(git ls-files '*.ts')"
+  },
   "files": [
     "extensions",
     "assets",