fix: tighten Anthropic provider matching and fix look-at test isolation

- Replace overly broad .includes('anthropic') with exact provider ID
  matching against known Anthropic providers (anthropic, google-vertex-
  anthropic, aws-bedrock-anthropic) in context-limit-resolver
- Add afterEach cleanup for vision-capable-models cache in look-at
  tool tests to prevent cross-test state leakage

src/shared/context-limit-resolver.ts

@@ -8,7 +8,8 @@ export type ContextLimitModelCacheState = {
 }
 
 function isAnthropicProvider(providerID: string): boolean {
-  return providerID.toLowerCase().includes("anthropic")
+  const normalized = providerID.toLowerCase()
+  return normalized === "anthropic" || normalized === "google-vertex-anthropic" || normalized === "aws-bedrock-anthropic"
 }
 
 function getAnthropicActualLimit(modelCacheState?: ContextLimitModelCacheState): number {

src/tools/look-at/tools.test.ts

@@ -1,9 +1,13 @@
-import { describe, expect, test, mock } from "bun:test"
+import { afterEach, describe, expect, test, mock } from "bun:test"
 import type { ToolContext } from "@opencode-ai/plugin/tool"
-import { setVisionCapableModelsCache } from "../../shared/vision-capable-models-cache"
+import { clearVisionCapableModelsCache, setVisionCapableModelsCache } from "../../shared/vision-capable-models-cache"
 import { normalizeArgs, validateArgs, createLookAt } from "./tools"
 
 describe("look-at tool", () => {
+  afterEach(() => {
+    clearVisionCapableModelsCache()
+  })
+
   describe("normalizeArgs", () => {
     // given LLM might use `path` instead of `file_path`
     // when called with path parameter