diff --git a/plugins/prompt-task/.claude-plugin b/plugins/prompt-task/.claude-plugin
new file mode 120000
index 00000000..665797f0
--- /dev/null
+++ b/plugins/prompt-task/.claude-plugin
@@ -0,0 +1 @@
+.plugin
\ No newline at end of file
diff --git a/plugins/prompt-task/.codex-plugin b/plugins/prompt-task/.codex-plugin
new file mode 120000
index 00000000..665797f0
--- /dev/null
+++ b/plugins/prompt-task/.codex-plugin
@@ -0,0 +1 @@
+.plugin
\ No newline at end of file
diff --git a/plugins/prompt-task/.plugin/plugin.json b/plugins/prompt-task/.plugin/plugin.json
new file mode 100644
index 00000000..d43391b6
--- /dev/null
+++ b/plugins/prompt-task/.plugin/plugin.json
@@ -0,0 +1,46 @@
+{
+  "name": "prompt-task",
+  "version": "1.0.0",
+  "description": "Run a typed prompt as an OpenHands automation, launched from the agent profile chosen for it",
+  "author": {
+    "name": "OpenHands",
+    "email": "contact@all-hands.dev"
+  },
+  "homepage": "https://github.com/OpenHands/extensions/tree/main/plugins/prompt-task",
+  "repository": "https://github.com/OpenHands/extensions",
+  "license": "MIT",
+  "keywords": [
+    "automation",
+    "prompt",
+    "agent-profile"
+  ],
+  "automation": {
+    "schema_version": "openhands.automation.plugin.v1",
+    "entrypoint": {
+      "runner": "conversation",
+      "command": "run",
+      "arguments": "{{ config.prompt }}"
+    },
+    "runtime": {
+      "agent_profile": {
+        "required": true
+      }
+    },
+    "config_schema": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "prompt"
+      ],
+      "properties": {
+        "prompt": {
+          "type": "string",
+          "title": "Prompt",
+          "description": "What the agent does on every run.",
+          "minLength": 1,
+          "maxLength": 50000
+        }
+      }
+    }
+  }
+}
diff --git a/plugins/prompt-task/README.md b/plugins/prompt-task/README.md
new file mode 100644
index 00000000..8e055991
--- /dev/null
+++ b/plugins/prompt-task/README.md
@@ -0,0 +1,31 @@
+# prompt-task
+
+The stock plugin behind prompt automations. It turns a typed prompt into a
+plugin-spec automation, so the run is launched from the agent profile chosen
+for the automation: its tools (including sub-agents), MCP servers, skills and
+model.
+
+## Usage
+
+Create the automation through the Automation Service's `POST /v1/plugin-spec`:
+
+```json
+{
+  "name": "Weekly issue triage",
+  "plugin": {
+    "source": "github:OpenHands/extensions",
+    "repo_path": "plugins/prompt-task"
+  },
+  "config": {"prompt": "Label every unlabeled issue opened this week"},
+  "agent_profile_id": "<agent profile id>",
+  "trigger": {"type": "cron", "schedule": "0 9 * * 1", "timezone": "UTC"}
+}
+```
+
+Each run sends `/prompt-task:run <prompt>`, where `<prompt>` is the
+`config.prompt` value, to a conversation the service starts from the agent
+profile. The manifest names no default profile, so `agent_profile_id` is
+required; a client normally defaults it to the user's active profile.
+
+`agent_profile_id` requires an Automation Service that advertises
+`pluginSpecAgentProfiles` in `GET /v1/capabilities`.
diff --git a/plugins/prompt-task/commands/run.md b/plugins/prompt-task/commands/run.md
new file mode 100644
index 00000000..355ece32
--- /dev/null
+++ b/plugins/prompt-task/commands/run.md
@@ -0,0 +1,15 @@
+---
+description: Carry out the task an automation was set up with
+argument-hint: <task>
+---
+
+# Automation task
+
+An automation started this conversation. Your task is the text that follows
+`/prompt-task:run` in the message.
+
+- Nobody is watching the conversation live, so complete the task end to end
+  without waiting for a reply.
+- If the message includes an event payload or follow-up messages, treat them
+  as the context the task refers to.
+- End with a short summary of what you did and anything left undone.
diff --git a/tests/test_prompt_task_plugin.py b/tests/test_prompt_task_plugin.py
new file mode 100644
index 00000000..9c106827
--- /dev/null
+++ b/tests/test_prompt_task_plugin.py
@@ -0,0 +1,46 @@
+"""The prompt-task plugin backs prompt automations created as plugin specs."""
+
+import json
+import re
+from pathlib import Path
+
+import yaml
+
+
+REPO_ROOT = Path(__file__).parent.parent
+PLUGIN = REPO_ROOT / "plugins" / "prompt-task"
+
+
+def _automation() -> dict:
+    return json.loads((PLUGIN / ".plugin" / "plugin.json").read_text())["automation"]
+
+
+def test_runs_the_configured_prompt_on_a_chosen_agent_profile():
+    automation = _automation()
+    entrypoint = automation["entrypoint"]
+
+    assert automation["schema_version"] == "openhands.automation.plugin.v1"
+    assert entrypoint["runner"] == "conversation"
+    assert (PLUGIN / "commands" / f"{entrypoint['command']}.md").is_file()
+    assert entrypoint["arguments"] == "{{ config.prompt }}"
+    assert automation["config_schema"]["required"] == ["prompt"]
+    assert automation["runtime"]["agent_profile"] == {"required": True}
+
+
+def test_command_triggers_no_public_skill():
+    message = "/prompt-task:run "
+    matches = []
+    for skill in [
+        *REPO_ROOT.glob("skills/*/SKILL.md"),
+        *REPO_ROOT.glob("plugins/*/SKILL.md"),
+        *REPO_ROOT.glob("plugins/*/skills/*/SKILL.md"),
+    ]:
+        frontmatter = re.match(r"^---\n(.*?)\n---", skill.read_text(), re.S)
+        if not frontmatter:
+            continue
+        for trigger in (yaml.safe_load(frontmatter.group(1)) or {}).get(
+            "triggers"
+        ) or []:
+            if str(trigger).lower() in message:
+                matches.append(f"{skill.parent.name}: {trigger}")
+    assert matches == []