fix(context-window-monitor): use model-specific context limits

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

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

src/hooks/context-window-monitor.model-context-limits.test.ts

@@ -0,0 +1,90 @@
+/// <reference types="bun-types" />
+
+import { describe, expect, it } from "bun:test"
+import { createContextWindowMonitorHook } from "./context-window-monitor"
+
+function createOutput() {
+  return { title: "", output: "original", metadata: null }
+}
+
+describe("context-window-monitor modelContextLimitsCache", () => {
+  it("does not append reminder below cached non-anthropic threshold", async () => {
+    // given
+    const modelContextLimitsCache = new Map<string, number>()
+    modelContextLimitsCache.set("opencode/kimi-k2.5-free", 262144)
+
+    const hook = createContextWindowMonitorHook({} as never, {
+      anthropicContext1MEnabled: false,
+      modelContextLimitsCache,
+    })
+    const sessionID = "ses_non_anthropic_below_threshold"
+
+    await hook.event({
+      event: {
+        type: "message.updated",
+        properties: {
+          info: {
+            role: "assistant",
+            sessionID,
+            providerID: "opencode",
+            modelID: "kimi-k2.5-free",
+            finish: true,
+            tokens: {
+              input: 150000,
+              output: 0,
+              reasoning: 0,
+              cache: { read: 10000, write: 0 },
+            },
+          },
+        },
+      },
+    })
+
+    // when
+    const output = createOutput()
+    await hook["tool.execute.after"]({ tool: "bash", sessionID, callID: "call_1" }, output)
+
+    // then
+    expect(output.output).toBe("original")
+  })
+
+  it("appends reminder above cached non-anthropic threshold", async () => {
+    // given
+    const modelContextLimitsCache = new Map<string, number>()
+    modelContextLimitsCache.set("opencode/kimi-k2.5-free", 262144)
+
+    const hook = createContextWindowMonitorHook({} as never, {
+      anthropicContext1MEnabled: false,
+      modelContextLimitsCache,
+    })
+    const sessionID = "ses_non_anthropic_above_threshold"
+
+    await hook.event({
+      event: {
+        type: "message.updated",
+        properties: {
+          info: {
+            role: "assistant",
+            sessionID,
+            providerID: "opencode",
+            modelID: "kimi-k2.5-free",
+            finish: true,
+            tokens: {
+              input: 180000,
+              output: 0,
+              reasoning: 0,
+              cache: { read: 10000, write: 0 },
+            },
+          },
+        },
+      },
+    })
+
+    // when
+    const output = createOutput()
+    await hook["tool.execute.after"]({ tool: "bash", sessionID, callID: "call_1" }, output)
+
+    // then
+    expect(output.output).toContain("context remaining")
+  })
+})

src/hooks/context-window-monitor.ts

@@ -7,6 +7,7 @@ const CONTEXT_WARNING_THRESHOLD = 0.70
 
 type ModelCacheStateLike = {
   anthropicContext1MEnabled: boolean
+  modelContextLimitsCache?: Map<string, number>
 }
 
 function getAnthropicActualLimit(modelCacheState?: ModelCacheStateLike): number {
@@ -32,6 +33,7 @@ interface TokenInfo {
 
 interface CachedTokenState {
   providerID: string
+  modelID: string
   tokens: TokenInfo
 }
 
@@ -57,13 +59,19 @@ export function createContextWindowMonitorHook(
     const cached = tokenCache.get(sessionID)
     if (!cached) return
 
-    if (!isAnthropicProvider(cached.providerID)) return
+    const cachedLimit = modelCacheState?.modelContextLimitsCache?.get(
+      `${cached.providerID}/${cached.modelID}`
+    )
+    const actualLimit =
+      cachedLimit ??
+      (isAnthropicProvider(cached.providerID) ? getAnthropicActualLimit(modelCacheState) : null)
+
+    if (!actualLimit) return
 
     const lastTokens = cached.tokens
     const totalInputTokens = (lastTokens?.input ?? 0) + (lastTokens?.cache?.read ?? 0)
 
-    const actualUsagePercentage =
-      totalInputTokens / getAnthropicActualLimit(modelCacheState)
+    const actualUsagePercentage = totalInputTokens / actualLimit
 
     if (actualUsagePercentage < CONTEXT_WARNING_THRESHOLD) return
 
@@ -95,6 +103,7 @@ export function createContextWindowMonitorHook(
         role?: string
         sessionID?: string
         providerID?: string
+        modelID?: string
         finish?: boolean
         tokens?: TokenInfo
       } | undefined
@@ -104,6 +113,7 @@ export function createContextWindowMonitorHook(
 
       tokenCache.set(info.sessionID, {
         providerID: info.providerID,
+        modelID: info.modelID ?? "",
         tokens: info.tokens,
       })
     }