fix(openrouter): allow DeepSeek cache-ttl eligibility

CHANGELOG.md

@@ -72,6 +72,7 @@ Docs: https://docs.openclaw.ai
 ### Fixes
 
 - Dashboard/security: avoid writing tokenized Control UI URLs or SSH hints to runtime logs, keeping gateway bearer fragments out of console-captured logs readable through `logs.tail`. (#70029) Thanks @Ziy1-Tan.
+- Providers/OpenRouter: treat DeepSeek refs as cache-TTL eligible without injecting Anthropic cache-control markers, aligning context pruning with OpenRouter-managed prompt caching. (#51983) Thanks @QuinnH496.
 - Discord/cron: deliver text-only isolated cron and heartbeat announce output from the canonical final assistant text once, avoiding duplicate Discord posts when streamed block payloads and the final answer contain the same content. Fixes #71406. Thanks @alexgross21.
 - macOS Gateway: wait for launchd to reload the exited Gateway LaunchAgent before bootstrapping repair fallback, preventing config-triggered restarts from leaving the service not loaded. Fixes #45178. Thanks @vincentkoc.
 - TTS/hooks: preserve audio-only TTS transcripts for `message_sending` and `message_sent` hooks without rendering the transcript as a media caption. Thanks @zqchris.

docs/concepts/model-providers.md

@@ -269,7 +269,7 @@ See [/providers/kilocode](/providers/kilocode) for setup details.
 
 Quirks worth knowing:
 
-- **OpenRouter** applies its app-attribution headers and Anthropic `cache_control` markers only on verified `openrouter.ai` routes. As a proxy-style OpenAI-compatible path, it skips native-OpenAI-only shaping (`serviceTier`, Responses `store`, prompt-cache hints, OpenAI reasoning-compat). Gemini-backed refs keep proxy-Gemini thought-signature sanitation only.
+- **OpenRouter** applies its app-attribution headers and Anthropic `cache_control` markers only on verified `openrouter.ai` routes. DeepSeek, Moonshot, and ZAI refs are cache-TTL eligible for OpenRouter-managed prompt caching but do not receive Anthropic cache markers. As a proxy-style OpenAI-compatible path, it skips native-OpenAI-only shaping (`serviceTier`, Responses `store`, prompt-cache hints, OpenAI reasoning-compat). Gemini-backed refs keep proxy-Gemini thought-signature sanitation only.
 - **Kilo Gateway** Gemini-backed refs follow the same proxy-Gemini sanitation path; `kilocode/kilo/auto` and other proxy-reasoning-unsupported refs skip proxy reasoning injection.
 - **MiniMax** API-key onboarding writes explicit text-only M2.7 chat model definitions; image understanding stays on the plugin-owned `MiniMax-VL-01` media provider.
 - **xAI** uses the xAI Responses path. `/fast` or `params.fastMode: true` rewrites `grok-3`, `grok-3-mini`, `grok-4`, and `grok-4-0709` to their `*-fast` variants. `tool_stream` defaults on; disable via `agents.defaults.models["xai/<model>"].params.tool_stream=false`.

docs/reference/prompt-caching.md

@@ -123,7 +123,7 @@ Per-agent heartbeat is supported at `agents.list[].heartbeat`.
 - Anthropic Claude model refs (`amazon-bedrock/*anthropic.claude*`) support explicit `cacheRetention` pass-through.
 - Non-Anthropic Bedrock models are forced to `cacheRetention: "none"` at runtime.
 
-### OpenRouter Anthropic models
+### OpenRouter models
 
 For `openrouter/anthropic/*` model refs, OpenClaw injects Anthropic
 `cache_control` on system/developer prompt blocks to improve prompt-cache
@@ -131,6 +131,11 @@ reuse only when the request is still targeting a verified OpenRouter route
 (`openrouter` on its default endpoint, or any provider/base URL that resolves
 to `openrouter.ai`).
 
+For `openrouter/deepseek/*`, `openrouter/moonshot*/*`, and `openrouter/zai/*`
+model refs, `contextPruning.mode: "cache-ttl"` is allowed because OpenRouter
+handles provider-side prompt caching automatically. OpenClaw does not inject
+Anthropic `cache_control` markers into those requests.
+
 If you repoint the model at an arbitrary OpenAI-compatible proxy URL, OpenClaw
 stops injecting those OpenRouter-specific Anthropic cache markers.
 

extensions/openrouter/index.ts

@@ -27,6 +27,7 @@ const PROVIDER_ID = "openrouter";
 const OPENROUTER_DEFAULT_MAX_TOKENS = 8192;
 const OPENROUTER_CACHE_TTL_MODEL_PREFIXES = [
   "anthropic/",
+  "deepseek/",
   "moonshot/",
   "moonshotai/",
   "zai/",

src/agents/pi-embedded-runner/cache-ttl.test.ts

@@ -16,7 +16,7 @@ vi.mock("../../plugins/provider-runtime.js", async () => {
         return true;
       }
       if (params.context.provider === "openrouter") {
-        return ["anthropic/", "moonshot/", "moonshotai/", "zai/"].some((prefix) =>
+        return ["anthropic/", "deepseek/", "moonshot/", "moonshotai/", "zai/"].some((prefix) =>
           params.context.modelId.startsWith(prefix),
         );
       }
@@ -44,6 +44,7 @@ describe("isCacheTtlEligibleProvider", () => {
 
   it("allows openrouter cache-ttl models", () => {
     expect(isCacheTtlEligibleProvider("openrouter", "anthropic/claude-sonnet-4")).toBe(true);
+    expect(isCacheTtlEligibleProvider("openrouter", "deepseek/deepseek-v3.2")).toBe(true);
     expect(isCacheTtlEligibleProvider("openrouter", "moonshotai/kimi-k2.5")).toBe(true);
     expect(isCacheTtlEligibleProvider("openrouter", "moonshot/kimi-k2.5")).toBe(true);
     expect(isCacheTtlEligibleProvider("openrouter", "zai/glm-5")).toBe(true);

src/agents/pi-embedded-runner/proxy-stream-wrappers.test.ts

@@ -79,6 +79,14 @@ describe("proxy stream wrappers", () => {
     expect(payload.messages[0]?.content).toBe("system prompt");
   });
 
+  it("does not inject Anthropic cache_control markers for automatic OpenRouter DeepSeek cache models", () => {
+    const payload = runSystemCacheWrapper({
+      id: "deepseek/deepseek-v3.2",
+    });
+
+    expect(payload.messages[0]?.content).toBe("system prompt");
+  });
+
   it("injects cache_control markers for native OpenRouter hosts behind custom provider ids", () => {
     const payload = runSystemCacheWrapper({
       provider: "custom-openrouter",