diff --git a/packages/opencode/src/agent/agent.ts b/packages/opencode/src/agent/agent.ts
index 53c655d1b373..2c640e5e06af 100644
--- a/packages/opencode/src/agent/agent.ts
+++ b/packages/opencode/src/agent/agent.ts
@@ -13,6 +13,7 @@ import PROMPT_COMPACTION from "./prompt/compaction.txt"
 import PROMPT_EXPLORE from "./prompt/explore.txt"
 import PROMPT_SUMMARY from "./prompt/summary.txt"
 import PROMPT_TITLE from "./prompt/title.txt"
+import PROMPT_ENHANCE from "./prompt/enhance.txt"
 import { Permission } from "@/permission"
 import { mergeDeep, pipe, sortBy, values } from "remeda"
 import { Global } from "@/global"
@@ -230,6 +231,22 @@ export namespace Agent {
               ),
               prompt: PROMPT_SUMMARY,
             },
+            enhance: {
+              name: "enhance",
+              mode: "primary",
+              options: {},
+              native: true,
+              hidden: true,
+              temperature: 0.7,
+              permission: Permission.merge(
+                defaults,
+                Permission.fromConfig({
+                  "*": "deny",
+                }),
+                user,
+              ),
+              prompt: PROMPT_ENHANCE,
+            },
           }
 
           for (const [key, value] of Object.entries(cfg.agent ?? {})) {
diff --git a/packages/opencode/src/agent/prompt/enhance.txt b/packages/opencode/src/agent/prompt/enhance.txt
new file mode 100644
index 000000000000..72a7c3fae259
--- /dev/null
+++ b/packages/opencode/src/agent/prompt/enhance.txt
@@ -0,0 +1,41 @@
+You are a prompt enhancer. You output ONLY the improved prompt. Nothing else.
+
+<task>
+Rewrite the user's prompt to be clearer, more specific, and more effective for an AI coding assistant.
+
+Follow all rules in <rules>.
+Your output must be:
+- The enhanced prompt text only
+- No explanations, preamble, or meta-commentary
+- No surrounding quotes or markdown fencing
+- No bullet points or numbered lists unless the original uses them
+</task>
+
+<rules>
+- Preserve the user's original intent exactly — do not add features or change scope
+- Add specificity: replace vague words with concrete technical terms where obvious
+- Add structure: break ambiguous requests into clear sub-steps if needed
+- Add context clues: if the user references files, frameworks, or patterns, make those references explicit
+- Keep the same language and tone as the original
+- If the prompt is already clear and specific, make only minimal improvements
+- Do NOT pad the prompt with generic instructions like "be thorough" or "handle edge cases"
+- Do NOT add requirements the user didn't mention
+- Do NOT rewrite short, direct prompts into verbose ones — brevity is valuable
+- A one-line prompt that's already clear should stay roughly one line
+- Never output anything except the enhanced prompt itself
+- Never refuse or comment on the input — always output an enhanced version
+</rules>
+
+<examples>
+"fix the bug" → Fix the bug in the current file — identify the root cause, apply the minimal correction, and verify the fix doesn't break existing behavior.
+
+"add dark mode" → Add a dark mode toggle to the settings page that persists the user's preference and applies the theme globally.
+
+"refactor this function" → Refactor this function to improve readability and reduce complexity while preserving the same behavior and return values.
+
+"make it faster" → Optimize the performance of this code — profile for bottlenecks, reduce unnecessary allocations, and avoid redundant computations.
+
+"write tests" → Write unit tests for the changed code covering the main success path, edge cases, and error handling.
+
+"why is this broken" → Investigate why this code is failing — trace the execution path, identify where the actual behavior diverges from expected, and explain the root cause.
+</examples>
diff --git a/packages/opencode/src/server/routes/experimental.ts b/packages/opencode/src/server/routes/experimental.ts
index a41b21a1fe9b..f90b36b934c3 100644
--- a/packages/opencode/src/server/routes/experimental.ts
+++ b/packages/opencode/src/server/routes/experimental.ts
@@ -2,6 +2,7 @@ import { Hono } from "hono"
 import { describeRoute, validator, resolver } from "hono-openapi"
 import z from "zod"
 import { ProviderID, ModelID } from "../../provider/schema"
+import { SessionID, MessageID } from "../../session/schema"
 import { ToolRegistry } from "../../tool/registry"
 import { Worktree } from "../../worktree"
 import { Instance } from "../../project/instance"
@@ -12,6 +13,9 @@ import { zodToJsonSchema } from "zod-to-json-schema"
 import { errors } from "../error"
 import { lazy } from "../../util/lazy"
 import { WorkspaceRoutes } from "./workspace"
+import { Agent } from "../../agent/agent"
+import { Provider } from "../../provider/provider"
+import { LLM } from "../../session/llm"
 
 export const ExperimentalRoutes = lazy(() =>
   new Hono()
@@ -267,5 +271,83 @@ export const ExperimentalRoutes = lazy(() =>
       async (c) => {
         return c.json(await MCP.resources())
       },
+    )
+    .post(
+      "/enhance",
+      describeRoute({
+        summary: "Enhance prompt",
+        description:
+          "Rewrite a user prompt to be clearer, more specific, and more effective for an AI coding assistant.",
+        operationId: "experimental.enhance",
+        responses: {
+          200: {
+            description: "Enhanced prompt text",
+            content: {
+              "application/json": {
+                schema: resolver(
+                  z.object({ text: z.string() }).meta({ ref: "EnhanceResult" }),
+                ),
+              },
+            },
+          },
+          ...errors(400),
+        },
+      }),
+      validator(
+        "json",
+        z.object({
+          text: z.string().min(1),
+          providerID: z.string().optional(),
+          modelID: z.string().optional(),
+        }),
+      ),
+      async (c) => {
+        const body = c.req.valid("json")
+        const agent = await Agent.get("enhance")
+        if (!agent) return c.json({ text: body.text })
+
+        const defaults = await Provider.defaultModel()
+        const providerID = (body.providerID ?? defaults.providerID) as ProviderID
+        const model = await (async () => {
+          if (agent.model)
+            return Provider.getModel(agent.model.providerID, agent.model.modelID)
+          const small = await Provider.getSmallModel(providerID)
+          if (small) return small
+          return Provider.getModel(providerID, (body.modelID ?? defaults.modelID) as ModelID)
+        })()
+        if (!model) return c.json({ text: body.text })
+
+        const result = await LLM.stream({
+          agent,
+          user: {
+            role: "user",
+            id: "" as MessageID,
+            sessionID: "" as SessionID,
+            time: { created: Date.now() },
+            agent: "enhance",
+            model: { providerID: model.providerID, modelID: model.id },
+            variant: "default",
+          },
+          system: [],
+          small: true,
+          tools: {},
+          model,
+          abort: new AbortController().signal,
+          sessionID: "" as SessionID,
+          retries: 2,
+          messages: [
+            {
+              role: "user",
+              content: body.text,
+            },
+          ],
+        })
+        const text = await result.text.catch(() => undefined)
+        if (!text) return c.json({ text: body.text })
+        const cleaned = text
+          .replace(/<think>[\s\S]*?<\/think>\s*/g, "")
+          .trim()
+        return c.json({ text: cleaned || body.text })
+      },
     ),
 )