Let Me Know

Notify the user before starting any long-running task and keep them updated. Use when a task will take noticeable time (>2-3 minutes). Send a start message, schedule a 5‑minute heartbeat update, and send a completion message immediately when done.

Security Scan

The skill's instructions broadly match a 'notify on long tasks' purpose, but they ask the agent to read/write state files, manage cron jobs, and even perform a gateway restart without declaring or restricting those privileges — the declared metadata doesn't justify or enumerate those capabilities.

Purpose & Capability

The skill name/description (notify for long tasks) is coherent with the instructions (send start/heartbeat/completion). However, the runtime instructions require capabilities beyond a simple notifier: reading progress/state files or logs, creating and removing cron jobs, and performing gateway restarts. The registry metadata declares no required binaries, env vars, config paths, or permissions, so there is a mismatch between what the skill says it needs and what the instructions actually rely on.

Instruction Scope

SKILL.md explicitly instructs the agent to: read latest progress from a state file/logs, write heartbeatJobId to a state file, run sleep loops inside an agent turn, call 'cron add/list/remove', send messages with 'message send', and (as a last resort) use gateway restart with commands.restart: true. These are concrete file, scheduling, and privileged-control actions that reach beyond merely composing notifications. The skill also leaves unspecified where logs/state live and which cron/gateway API is used, which could lead to the agent accessing or modifying unexpected files or system services.

Install Mechanism

No install spec and no code files — the skill is instruction-only. That minimizes the risk of external code being fetched or installed.

Credentials

The skill declares no environment variables or credentials, yet the instructions require access to system-like capabilities (cron, state file paths, gateway restart). The absence of declared required permissions or config paths is disproportionate to the operational needs described and obscures which credentials or platform privileges are actually necessary.

Persistence & Privilege

The skill doesn't set always:true nor disableModelInvocation, but the instructions encourage creating persistent cron jobs and using gateway restart (a privileged action). Persistent scheduling combined with the ability to restart gateway components is a high-privilege design; the skill metadata does not document or limit these privileges, which is a notable mismatch.

What to consider before installing

This skill's behavior (start/heartbeat/completion notifications) is reasonable, but SKILL.md expects the agent to read/write state files, manage cron jobs, and perform gateway restarts — none of which are declared in the metadata. Before installing, ask or verify: (1) which platform APIs/permissions will the agent actually use for cron, message send, and gateway restart; (2) what filesystem paths will it read/write (to ensure it won't access unrelated files); and (3) who can invoke cron/gateway restart and whether those actions are allowed. If those answers are unclear or the platform can't safely grant tightly scoped permissions for cron and restart, treat this skill as risky and avoid installing it until the capabilities and required permissions are explicitly documented.

Purpose

Ensure the user is informed before long-running tasks start, gets periodic updates on a configurable interval (default 5 minutes), and receives an immediate completion/failure notice. Heartbeat messages must reflect real-time progress, not a repeated template.

Trigger

Use this skill whenever a task will take noticeable time (>2–3 minutes) or involves long-running installs/builds/tests.

Workflow (required)

1. Pre-flight message (before starting):

• Send a short message: what will run, estimated time, and explicitly state:
  • “完成或失败都会立刻通知你；期间我每 X 分钟 发一次进度心跳，您也可以修改心跳时间间隔。”

2. Start a heartbeat (configurable interval, with pre-check)

• Default interval = 5 minutes (everyMs=300000). If the user specifies a different interval, use it.
• Schedule repeating updates while the task runs.
• Before each heartbeat message, read the latest progress (state file/logs) and send current progress (no repeated template):
  • Running → include latest step, progress metrics, and next step.
  • Failed → send failure notice and stop the heartbeat.
• 优先推荐：同一条 agentTurn 内“原地心跳”（不创建额外 cron）：
  • 在长任务执行期间，用循环 sleep <interval> → 读取进度 → message send 发一次动态进度。
  • 任务结束自然停止，不会遗留心跳任务。
• 只有在必须脱离当前执行流时才用 cron 心跳，并且必须满足：
  • 通过 cron add 创建心跳 job 时，payload.deliver=false（避免“收到/启动”之类消息被转发给用户）。
  • 心跳 job 内部用 message send 主动推送进度。
  • 创建后把返回的 heartbeatJobId 写入状态文件（例如 <task>-state.json），供清理使用。
  • 创建前先 cron list，若已存在同名心跳 job，先 remove（去重）。
• Content template (dynamic):
  • Running: 进度：<最新步骤/阶段>（<关键指标>）。下一步：<next>。完成/失败会立刻通知你。
  • Failed: 失败：<task> 发生错误（简述原因）。已停止心跳提醒。

3. Run the task

• Execute the long-running command(s).

4. Completion message (immediately after finish)

• Send result summary (success/failure + key output).

5. Stop heartbeat（必须做到）

• 如果你使用了“原地心跳”（推荐）：任务结束即可，不会遗留任何 cron。
• 如果你使用了 cron 心跳：
  • 在任务成功/失败的 finally 里调用 cron remove <heartbeatJobId>。
  • 若 remove 失败（gateway timeout）：至少重试 2 次（指数退避 2s/8s）。
  • 仍失败：创建一个 2 分钟后的一次性 cleanup cron 再次 remove（避免永远刷屏）。

Heartbeat interval (user-configurable)

• Default: 5 minutes.
• If the user specifies an interval (e.g., “每 2 分钟/10 分钟”), use that value.
• If the user changes the interval mid-task, update the cron schedule and acknowledge in the next heartbeat.

Message Delivery

Prefer outbound normal chat messages:

• Use message send with the correct target format.
• Example for Discord DM: user:<id>.

Safety

• Do not start long tasks without the pre-flight message.
• If blocked/failed, notify immediately, set state=failed, and stop the heartbeat.
• If cron removal fails due to gateway timeout, retry removal; if still stuck, use gateway restart (requires commands.restart: true) and retry.

Example (Discord DM)

Start message:

• 即将开始：安装依赖并运行测试（预计 5–10 分钟）。完成或失败都会立刻通知你；期间我每 5 分钟发一次进度心跳，你也可以修改心跳时间间隔。

Heartbeat (every 5 min, example):

• 进度：已完成安装依赖（1/2），测试运行中（已用时 4 分钟）。下一步：汇总测试结果。完成/失败会立刻通知你。

Completion:

• 完成：安装成功，测试通过。