Security audit
Security checks for vulnerabilities and agentic risk
Sport Mode is not malicious, but it persistently changes agent heartbeat behavior and stores broad autonomous tasks that can keep running without a built-in stop limit.
Review this skill before installing. It is appropriate only if you want OpenClaw to change its heartbeat interval and repeatedly act on instructions stored in HEARTBEAT.md. Use explicit, narrow tasks with a stop condition, avoid tasks that authorize commits or deployments unless you intend that autonomy, and back up any existing HEARTBEAT.md content first.
The top-level description markets Sport Mode as high-frequency monitoring and auto-cleanup, but the later examples instruct the agent to perform unrelated autonomous actions. This mismatch increases the risk of deceptive capability framing: users may consent to monitoring while unintentionally enabling broader autonomy and command execution loops.
The skill is explicitly designed to increase heartbeat frequency and persist mission state across repeated agent wakeups, enabling a durable autonomy loop. That persistence is not inherently malicious, but in this context it amplifies the danger of the README's task examples by letting broad instructions continue executing over time with limited fresh user oversight.
## Why?
By default, OpenClaw checks in every 30 minutes. That's great for battery life, but terrible when you're:
- 🍿 Watching **Codex** write code (you want to see errors *now*).
- 🏗️ Waiting for a **long build** or deployment.
- 🎮 Playing a turn-based game with your agent.
The conversational activation examples use generic natural-language prompts like 'watch it for me' and 'start Sport Mode,' which can overlap with normal chat and cause accidental invocation or over-broad interpretation. In an agent system, ambiguous trigger language can convert informal requests into persistent autonomous behavior without a clear consent boundary.
The README states that the skill hot-patches openclaw.json and creates/overwrites or clears HEARTBEAT.md, but it does not prominently warn users about these local file modifications and their operational consequences. Silent or under-disclosed config mutation is risky because it changes agent behavior persistently and may destroy existing file contents or interfere with other workflows.
The README materially expands the skill from heartbeat tuning into broad autonomous task execution, including monitoring Codex, auto-starting next phases, committing code, and playing games. This is dangerous because a user may invoke what appears to be a scheduling/monitoring helper, while the agent is encouraged to take consequential actions based on loosely specified task text and persisted state in HEARTBEAT.md.
No suspicious patterns detected.