fix(config): clear stale context limit cache on provider updates

Rebuilding provider model limits prevents removed entries from leaking into later compaction decisions after config changes.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

src/hooks/preemptive-compaction.context-limit-cache.test.ts

@@ -0,0 +1,94 @@
+import { describe, expect, it, mock } from "bun:test"
+
+import { applyProviderConfig } from "../plugin-handlers/provider-config-handler"
+import { createModelCacheState } from "../plugin-state"
+
+const logMock = mock(() => {})
+
+mock.module("../shared/logger", () => ({
+  log: logMock,
+}))
+
+const { createPreemptiveCompactionHook } = await import("./preemptive-compaction")
+
+function createMockCtx() {
+  return {
+    client: {
+      session: {
+        messages: mock(() => Promise.resolve({ data: [] })),
+        summarize: mock(() => Promise.resolve({})),
+      },
+      tui: {
+        showToast: mock(() => Promise.resolve()),
+      },
+    },
+    directory: "/tmp/test",
+  }
+}
+
+describe("preemptive-compaction context-limit cache invalidation", () => {
+  it("skips compaction after provider config removes a cached model limit", async () => {
+    // given
+    const ctx = createMockCtx()
+    const modelCacheState = createModelCacheState()
+    const sessionID = "ses_removed_limit"
+
+    applyProviderConfig({
+      config: {
+        provider: {
+          opencode: {
+            models: {
+              "kimi-k2.5-free": {
+                limit: { context: 200000 },
+              },
+            },
+          },
+        },
+      },
+      modelCacheState,
+    })
+
+    const hook = createPreemptiveCompactionHook(ctx as never, {} as never, modelCacheState)
+
+    await hook.event({
+      event: {
+        type: "message.updated",
+        properties: {
+          info: {
+            role: "assistant",
+            sessionID,
+            providerID: "opencode",
+            modelID: "kimi-k2.5-free",
+            finish: true,
+            tokens: {
+              input: 170000,
+              output: 0,
+              reasoning: 0,
+              cache: { read: 0, write: 0 },
+            },
+          },
+        },
+      },
+    })
+
+    applyProviderConfig({
+      config: {
+        provider: {
+          opencode: {
+            models: {},
+          },
+        },
+      },
+      modelCacheState,
+    })
+
+    // when
+    await hook["tool.execute.after"](
+      { tool: "bash", sessionID, callID: "call_1" },
+      { title: "", output: "test", metadata: null },
+    )
+
+    // then
+    expect(ctx.client.session.summarize).not.toHaveBeenCalled()
+  })
+})

src/plugin-handlers/provider-config-handler.test.ts

@@ -6,6 +6,46 @@ import { createModelCacheState } from "../plugin-state"
 import { clearVisionCapableModelsCache, readVisionCapableModelsCache } from "../shared/vision-capable-models-cache"
 
 describe("applyProviderConfig", () => {
+  test("clears stale model context limits when provider config changes", () => {
+    // given
+    const modelCacheState = createModelCacheState()
+    applyProviderConfig({
+      config: {
+        provider: {
+          opencode: {
+            models: {
+              "kimi-k2.5-free": {
+                limit: { context: 262144 },
+              },
+            },
+          },
+        },
+      },
+      modelCacheState,
+    })
+
+    // when
+    applyProviderConfig({
+      config: {
+        provider: {
+          google: {
+            models: {
+              "gemini-2.5-pro": {
+                limit: { context: 1048576 },
+              },
+            },
+          },
+        },
+      },
+      modelCacheState,
+    })
+
+    // then
+    expect(Array.from(modelCacheState.modelContextLimitsCache.entries())).toEqual([
+      ["google/gemini-2.5-pro", 1048576],
+    ])
+  })
+
   test("caches vision-capable models from modalities and capabilities", () => {
     // given
     const modelCacheState = createModelCacheState()

src/plugin-handlers/provider-config-handler.ts

@@ -33,6 +33,9 @@ export function applyProviderConfig(params: {
   const providers = params.config.provider as
     | Record<string, ProviderConfig>
     | undefined;
+  const modelContextLimitsCache = params.modelCacheState.modelContextLimitsCache;
+
+  modelContextLimitsCache.clear()
 
   const anthropicBeta = providers?.anthropic?.options?.headers?.["anthropic-beta"];
   params.modelCacheState.anthropicContext1MEnabled =
@@ -61,7 +64,7 @@ export function applyProviderConfig(params: {
       const contextLimit = modelConfig?.limit?.context;
       if (!contextLimit) continue;
 
-      params.modelCacheState.modelContextLimitsCache.set(
+      modelContextLimitsCache.set(
         `${providerID}/${modelID}`,
         contextLimit,
       );