fix: disable tools for deepseek-reasoner and strip <think> tags from text parts

DeepSeek R1 (deepseek-reasoner) does not honour the tools parameter on the
standard api.deepseek.com endpoint despite models.dev reporting tool_call: true.
When tools are sent, R1 ignores them and writes the invocation as markdown text,
causing tool calls to appear as plain text instead of being executed.

- llm.ts: disable canTool for deepseek-reasoner so tools are never sent to it,
  avoiding the markdown-text fallback entirely
- processor.ts: strip <think>...</think> blocks from text parts for reasoning
  models that emit chain-of-thought inline rather than as dedicated reasoning
  events, preventing raw thinking tokens from appearing in the UI

Author: lgarceau768

packages/opencode/src/session/llm.ts

@@ -193,7 +193,15 @@ const live: Layer.Layer<
         },
       )
 
-      const canTool = input.model.capabilities.toolcall
+      // DeepSeek R1 (deepseek-reasoner) does not honour the `tools` parameter on the
+      // standard api.deepseek.com endpoint despite models.dev reporting tool_call: true.
+      // When tools are sent, R1 ignores the definitions and writes the invocation as
+      // markdown text inside its response — exactly the wrong behaviour. Disable tools
+      // for it so the model falls back to conversational output instead.
+      const isDeepSeekR1 =
+        input.model.providerID === "deepseek" &&
+        input.model.api.id.toLowerCase().includes("reasoner")
+      const canTool = input.model.capabilities.toolcall && !isDeepSeekR1
       const tools = canTool ? resolveTools(input) : {}
 
       // LiteLLM and some Anthropic proxies require the tools parameter to be present

packages/opencode/src/session/processor.ts

@@ -442,6 +442,13 @@ export const layer: Layer.Layer<
               },
               { text: ctx.currentText.text },
             )).text
+            // Some reasoning models (e.g. DeepSeek R1) emit chain-of-thought inside
+            // <think>…</think> tags in the text stream rather than as dedicated
+            // reasoning events. Strip those blocks so they don't appear as visible
+            // text in the UI — they are stored separately via reasoning-start/delta.
+            if (ctx.model.capabilities.reasoning) {
+              ctx.currentText.text = ctx.currentText.text.replace(/<think>[\s\S]*?<\/think>\s*/g, "").trimStart()
+            }
             {
               const end = Date.now()
               ctx.currentText.time = { start: ctx.currentText.time?.start ?? end, end }