diff --git a/.opencode/README.parallel-agents.md b/.opencode/README.parallel-agents.md
new file mode 100644
index 00000000000..467a6e1306d
--- /dev/null
+++ b/.opencode/README.parallel-agents.md
@@ -0,0 +1,37 @@
+# Parallel Agents Plugin (chat-visible task flow)
+
+## Enable plugin
+
+```json
+{
+  "$schema": "https://opencode.ai/config.json",
+  "plugin": ["file://./.opencode/plugins/parallel-agents.ts"]
+}
+```
+
+## What is fixed
+
+This plugin now queues subagents as native `subtask` parts in the **current chat session**.
+So during execution you see regular `task()` entries with running/completed/error status,
+not only one generic custom-tool spinner.
+
+## Tools
+
+- `agents_start`
+  - input: `tasks[]` (`agent`, `prompt`, optional `description`)
+  - action: pushes subtask parts into current session (`noReply: true`)
+  - output: `run_id`
+- `agents_status`
+  - input: `run_id`
+  - output: per-task status from native `task` tool parts
+- `agents_wait`
+  - input: `run_id`, optional `timeout`
+  - output: waits until all tasks are completed/error
+
+## Typical usage
+
+1. `agents_start`
+2. do other work
+3. `agents_status`
+4. `agents_wait`
+5. summarize results
diff --git a/.opencode/plugins/parallel-agents.ts b/.opencode/plugins/parallel-agents.ts
new file mode 100644
index 00000000000..47fe310fccb
--- /dev/null
+++ b/.opencode/plugins/parallel-agents.ts
@@ -0,0 +1,187 @@
+import { tool, type Plugin } from "@opencode-ai/plugin"
+
+type Item = {
+  agent: string
+  prompt: string
+  description: string
+}
+
+type Run = {
+  id: string
+  session: string
+  items: Item[]
+  started: number
+}
+
+const runs = new Map<string, Run>()
+
+function rid() {
+  return `${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 8)}`
+}
+
+
+export const ParallelAgentsPlugin: Plugin = async ({ client }) => {
+  async function parts(session: string) {
+    const res = await client.session.get({ path: { id: session } })
+    if (res.error || !res.data) return []
+    return res.data.messages.flatMap((msg) => msg.parts)
+  }
+
+  async function state(run: Run) {
+    const all = await parts(run.session)
+    const rows = run.items.map((item) => {
+      const hit = all.findLast(
+        (part) =>
+          part.type === "tool" &&
+          part.tool === "task" &&
+          part.state.input.subagent_type === item.agent &&
+          part.state.input.description === item.description,
+      )
+      if (!hit || hit.type !== "tool") {
+        return {
+          agent: item.agent,
+          description: item.description,
+          status: "pending",
+        }
+      }
+      return {
+        agent: item.agent,
+        description: item.description,
+        status: hit.state.status,
+        output: hit.state.status === "completed" ? hit.state.output : undefined,
+        error: hit.state.status === "error" ? hit.state.error : undefined,
+      }
+    })
+
+    const done = rows.every((item) => item.status === "completed" || item.status === "error")
+    return { rows, done }
+  }
+
+  return {
+    tool: {
+      agents_start: tool({
+        description:
+          "Queue multiple subagents in current chat as task() runs. They will appear in chat with task status updates.",
+        args: {
+          tasks: tool.schema
+            .array(
+              tool.schema.object({
+                agent: tool.schema.string(),
+                prompt: tool.schema.string(),
+                description: tool.schema.string().optional(),
+              }),
+            )
+            .min(1)
+            .max(8),
+        },
+        async execute(args, ctx) {
+          const items = args.tasks.map((item, idx) => ({
+            agent: item.agent,
+            prompt: item.prompt,
+            description: item.description?.trim() || `parallel task ${idx + 1}`,
+          }))
+          const id = rid()
+          runs.set(id, {
+            id,
+            session: ctx.sessionID,
+            items,
+            started: Date.now(),
+          })
+
+          await client.session.prompt({
+            path: { id: ctx.sessionID },
+            body: {
+              noReply: true,
+              parts: items.map((item) => ({
+                type: "subtask" as const,
+                agent: item.agent,
+                prompt: item.prompt,
+                description: item.description,
+              })),
+            },
+          })
+
+          return JSON.stringify(
+            {
+              run_id: id,
+              started: items.length,
+              note: "Subagents are queued as native subtask/task entries in this chat. Use agents_status or agents_wait.",
+              tasks: items.map((item) => ({ agent: item.agent, description: item.description })),
+            },
+            null,
+            2,
+          )
+        },
+      }),
+      agents_status: tool({
+        description: "Show per-agent task() status for a started run.",
+        args: {
+          run_id: tool.schema.string(),
+        },
+        async execute(args) {
+          const run = runs.get(args.run_id)
+          if (!run) return JSON.stringify({ error: "run_id not found" }, null, 2)
+          const data = await state(run)
+          return JSON.stringify(
+            {
+              run_id: run.id,
+              session: run.session,
+              started: run.started,
+              done: data.done,
+              tasks: data.rows,
+            },
+            null,
+            2,
+          )
+        },
+      }),
+      agents_wait: tool({
+        description: "Wait until all task() entries in run are completed or failed.",
+        args: {
+          run_id: tool.schema.string(),
+          timeout: tool.schema.number().int().positive().optional(),
+        },
+        async execute(args) {
+          const run = runs.get(args.run_id)
+          if (!run) return JSON.stringify({ error: "run_id not found" }, null, 2)
+          const end = Date.now() + (args.timeout ?? 600000)
+
+          while (Date.now() < end) {
+            const data = await state(run)
+            if (data.done) {
+              return JSON.stringify(
+                {
+                  run_id: run.id,
+                  done: true,
+                  tasks: data.rows,
+                },
+                null,
+                2,
+              )
+            }
+            await Bun.sleep(400)
+          }
+
+          const data = await state(run)
+          return JSON.stringify(
+            {
+              run_id: run.id,
+              done: data.done,
+              timeout: true,
+              tasks: data.rows,
+            },
+            null,
+            2,
+          )
+        },
+      }),
+    },
+    "experimental.chat.system.transform": async (_input, output) => {
+      output.system.push(`
+When user asks to run multiple subagents and wants visible task() progress in chat,
+use agents_start first, then agents_status / agents_wait.
+This keeps native task status lines visible in this same chat.
+`)
+    },
+  }
+}