@hiper2d/ai-agents 0.1.0 → 0.1.2

package/README.md

@@ -9,6 +9,9 @@ Extracted from the [AI Werewolf](https://aiwerewolf.net) game so its model layer
 shared across apps. Text agents today; TTS, STT and image-generation agents are planned as
 subpath exports.
 
+Migrating a project that has its own agent implementations? Hand [`MIGRATION.md`](./MIGRATION.md)
+to your AI coding assistant — it is written for that job.
+
 ## Install
 
 ```bash

package/dist/index.js

@@ -934,10 +934,10 @@ var LLM_CONSTANTS = {
 };
 var DEFAULT_MAX_OUTPUT_TOKENS = 8192;
 var SupportedAiModels = {
-  // Claude Fable - frontier reasoning model. Thinking is always on (no non-thinking variant).
+  // Claude Fable 5.1 (2026-09-02, was Fable 5) - frontier reasoning model. Thinking is always on (no non-thinking variant).
   [LLM_CONSTANTS.CLAUDE_FABLE]: {
-    displayName: "Claude Fable 5",
-    modelApiName: "claude-fable-5",
+    displayName: "Claude Fable 5.1",
+    modelApiName: "claude-fable-5-1",
     apiKeyName: API_KEY_CONSTANTS.ANTHROPIC,
     hasThinking: true,
     reasoningEffort: "high",
@@ -1036,10 +1036,11 @@ var SupportedAiModels = {
     tags: ["expensive"]
   },
   [LLM_CONSTANTS.GEMINI_FLASH]: {
-    // Repointed from gemini-3.6-flash 2026-08-13 (stable picker id, same pattern as gpt).
-    // 3.7 rejects thinkingLevel 'minimal' (low|medium|high only), unlike 3.5/3.6.
-    displayName: "Gemini 3.7 Flash",
-    modelApiName: "gemini-3.7-flash",
+    // Repointed 3.6 → 3.7 (2026-08-13) → 3.8 (2026-09-02); stable picker id, same pattern as gpt.
+    // 3.7 rejected thinkingLevel 'minimal' (low|medium|high only), unlike 3.5/3.6 — 3.8 untested
+    // on 'minimal', so keep the pin at 'medium' or above.
+    displayName: "Gemini 3.8 Flash",
+    modelApiName: "gemini-3.8-flash",
     apiKeyName: API_KEY_CONSTANTS.GOOGLE,
     hasThinking: true,
     reasoningEffort: "medium",
@@ -1335,10 +1336,11 @@ var MODEL_PRICING = {
   },
   // Anthropic models
   [SupportedAiModels[LLM_CONSTANTS.CLAUDE_FABLE].modelApiName]: {
-    // Full 1M context window at standard pricing (no extended-context premium)
+    // Full 1M context window at standard pricing (no extended-context premium).
+    // Fable 5.1 (2026-09-02): same $10/$50 as Fable 5, but cache reads dropped to $0.25/MTok.
     inputPrice: 10,
     outputPrice: 50,
-    cacheHitPrice: 1
+    cacheHitPrice: 0.25
   },
   [SupportedAiModels[LLM_CONSTANTS.CLAUDE_OPUS].modelApiName]: {
     inputPrice: 5,
@@ -1366,9 +1368,9 @@ var MODEL_PRICING = {
     extendedContextThresholdTokens: 2e5
   },
   [SupportedAiModels[LLM_CONSTANTS.GEMINI_FLASH].modelApiName]: {
-    // Launch pricing through 2026-12-31; doubles to $1.50/$7.50/$0.15 on 2027-01-01
-    // (ai.google.dev pricing page, fetched 2026-08-13) — ACTION NEEDED then: update these
-    // rates.
+    // 3.8 Flash (2026-09-02) launched at the same rates as 3.7. 3.7's launch pricing was
+    // scheduled to double to $1.50/$7.50/$0.15 on 2027-01-01 (ai.google.dev pricing page,
+    // fetched 2026-08-13) — ACTION NEEDED then: re-check whether 3.8 follows and update.
     // Cache storage cost ($0.50 / 1M tokens per hour) is not tracked here — the
     // schema only models per-token call costs, not time-based storage.
     inputPrice: 0.75,