pi-provider-litellm 2.0.2 → 2.0.3

package/dist/discover.js

@@ -85,6 +85,8 @@ function catalogLookupIds(id) {
     const match = /^(?:claude-)?(opus|sonnet|haiku)-(\d+)-(\d+)$/.exec(anthropicAlias);
     if (match)
         lookupIds.add(`claude-${match[1]}-${match[2]}-${match[3]}`);
+    if (anthropicAlias === "fable-5")
+        lookupIds.add("claude-fable-5");
     return [...lookupIds];
 }
 function findCatalogModelInProvider(provider, lookupIds) {

package/dist/index.js

@@ -3,7 +3,7 @@ import { readFile } from "node:fs/promises";
 import { join } from "node:path";
 import { getAgentDir } from "@earendil-works/pi-coding-agent";
 import { setupLiteLLMCostTracking } from "./cost.js";
-import { discoverModels, isGpt55Model, normalizeBaseUrl, shouldSuppressReasoningContent } from "./discover.js";
+import { discoverModels, isGpt55Model, isMoonshotModel, normalizeBaseUrl, shouldSuppressReasoningContent, } from "./discover.js";
 import { getGcloudToken, getGcloudTokenCacheKey, getGcloudTokenCommand, isGcloudTokenAuthEnabled, } from "./gcloud-token.js";
 import { getSessionIdFromFile } from "./litellm.js";
 import { createMcpToolDefinitions } from "./mcp-tools.js";
@@ -579,6 +579,56 @@ function createProviderAuth(definition) {
 function isReasoningItem(item) {
     return typeof item === "object" && item !== null && item.type === "reasoning";
 }
+// Returns the original array reference when no normalization is needed.
+function normalizeStrictToolMessages(messages) {
+    const normalized = [];
+    let imageParts = [];
+    let changed = false;
+    const flushImages = () => {
+        if (imageParts.length === 0)
+            return;
+        normalized.push({
+            role: "user",
+            content: [{ type: "text", text: "Attached image(s) from tool result:" }, ...imageParts],
+        });
+        imageParts = [];
+    };
+    for (const message of messages) {
+        if (message.role !== "tool")
+            flushImages();
+        if (message.role === "assistant" &&
+            Array.isArray(message.tool_calls) &&
+            message.tool_calls.length > 0 &&
+            message.content == null) {
+            normalized.push({ ...message, content: "" });
+            changed = true;
+        }
+        else if (message.role === "tool" && Array.isArray(message.content)) {
+            const previousImageCount = imageParts.length;
+            const content = message.content
+                .map((part) => {
+                if (typeof part === "string")
+                    return part;
+                if (part?.type === "text")
+                    return part.text || "";
+                if (part?.type === "image_url")
+                    imageParts.push(part);
+                return "";
+            })
+                .join("");
+            normalized.push({
+                ...message,
+                content: content || (imageParts.length > previousImageCount ? "(see attached image)" : "(no tool output)"),
+            });
+            changed = true;
+        }
+        else {
+            normalized.push(message);
+        }
+    }
+    flushImages();
+    return changed ? normalized : messages;
+}
 // Reasoning fields LiteLLM forwards to chat-completions providers. The Moonshot
 // path defaults them off; the gpt-5.5 tool path strips them entirely.
 const REASONING_SUPPRESSION_DEFAULTS = {
@@ -587,6 +637,7 @@ const REASONING_SUPPRESSION_DEFAULTS = {
     merge_reasoning_content_in_choices: true,
     thinking: { type: "disabled" },
 };
+const GEMINI_MODEL_PATTERN = /(?:^|[-_/.:])gemini(?=$|[-_/.:])/i;
 function prepareLiteLLMRequestPayload(payload, modelId, api, sessionId) {
     let next;
     const update = (key, value) => {
@@ -632,6 +683,39 @@ function prepareLiteLLMRequestPayload(payload, modelId, api, sessionId) {
             next.input = input.filter((item) => !isReasoningItem(item));
         }
     }
+    // Moonshot/Kimi applies strict OpenAI schema validation: assistant tool calls
+    // must carry string content, and tool results must be plain text.
+    if (modelId && isMoonshotModel(modelId)) {
+        const messages = (next ?? payload).messages;
+        if (Array.isArray(messages)) {
+            const normalized = normalizeStrictToolMessages(messages);
+            if (normalized !== messages) {
+                next ??= { ...payload };
+                next.messages = normalized;
+            }
+        }
+    }
+    if (modelId && GEMINI_MODEL_PATTERN.test(modelId)) {
+        const currentPayload = next ?? payload;
+        if (typeof currentPayload.reasoning_effort === "string") {
+            const lower = currentPayload.reasoning_effort.toLowerCase();
+            if (currentPayload.reasoning_effort !== lower) {
+                next ??= { ...payload };
+                next.reasoning_effort = lower;
+            }
+        }
+        const reasoningPayload = next ?? payload;
+        if (isPlainObject(reasoningPayload.reasoning) && typeof reasoningPayload.reasoning.effort === "string") {
+            const lower = reasoningPayload.reasoning.effort.toLowerCase();
+            if (reasoningPayload.reasoning.effort !== lower) {
+                next ??= { ...payload };
+                next.reasoning = {
+                    ...next.reasoning,
+                    effort: lower,
+                };
+            }
+        }
+    }
     if (sessionId) {
         next ??= { ...payload };
         next.litellm_session_id = sessionId;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-provider-litellm",
-  "version": "2.0.2",
+  "version": "2.0.3",
   "description": "LiteLLM proxy provider extension for Pi",
   "type": "module",
   "license": "MIT",