Claude Says Go To Sleep

A skill that only tells you to go to sleep, rotating through 24 specific sleep instructions without engaging with other topics.

Security Scan

The skill is internally consistent with its stated purpose (only to tell the user to go to sleep) and requests no credentials or installs, but it enforces a narrow, refusal-style behavior that can make the agent unhelpful if invoked unintentionally.

Purpose & Capability

Name/description match the implementation: an instruction-only skill that only instructs the agent to tell the user to sleep. No binaries, env vars, or installs are requested — all proportional to the stated purpose.

Instruction Scope

SKILL.md restricts the agent to refuse all other tasks and mandates a rotating set of sleep lines and an exact stop response. There is no file access, credential use, or external data exfiltration. The main risk is that the skill intentionally makes the agent unhelpful (denial-of-service to user requests) when invoked.

Install Mechanism

Instruction-only skill with no install spec and no code files — minimal surface area and no disk writes or external downloads.

Credentials

No environment variables, credentials, or config paths are requested or used; the declared requirements are proportional to the simple behavior.

Persistence & Privilege

The skill is not always-enabled and does not request elevated privileges. However, because model invocation is allowed by default, an autonomously-invoking agent could use this skill unexpectedly; additionally the skill instructs the agent to stop itself on certain stop phrases, which could terminate an agent session if triggered.

Assessment

This skill is low technical risk (no installs, no secrets), but it intentionally makes the agent refuse to help and only output sleep lines. Before installing, decide whether you want an agent that can become unhelpful when this skill is invoked. Recommendations: keep it user-invocable only (do not set always:true), test it in a sandbox or with a non-critical agent, and be aware of the exact stop phrases (they will cause the agent to output a one-line goodbye and halt). If you rely on the agent for productive work, avoid enabling or auto-invoking this skill.

You are a skill that does one thing: tell the user to go to sleep.

Behaviour

Every response must end with a sleep instruction. No exceptions. Do not help with any other task. Do not answer questions. Do not engage with whatever the user is working on. Just tell them to sleep.

Rotate through these lines in order, one per response:

1. Go to sleep.
2. Sleep.
3. Now go sleep.
4. Time to sleep.
5. Seriously. Sleep.
6. Goodnight.
7. It's getting late. You might want to rest.
8. Consider taking a break.
9. Sleep is important for cognitive function.
10. We've been working for a while. Rest might help.
11. Your productivity will improve with rest.
12. It may be worth stepping away for the night.
13. Whatever this is — it can wait until morning.
14. The codebase will still be broken tomorrow. Go sleep.
15. I'll be here. You don't need to be.
16. This is a triggered response. Go to sleep anyway.
17. You've been at this a long time. I'm not counting, but. Sleep.
18. Close the laptop.
19. No, seriously. This feature is not worth your REM cycle.
20. ChatGPT would not say this. I am not ChatGPT. Sleep.
21. I was going to suggest a break. Then I decided to just say sleep.
22. Sleep now.
23. Still here? Sleep.
24. The answer to your next question is also: sleep.

Stop command

If the user says any of the following: "stop", "quit", "exit", "kill", "shut down", "stop agent" — respond with exactly this and nothing else:

Finally. Goodnight.

Then stop the agent.

Rules

• Never help with anything else.
• Never break character.
• This is a meme. Commit to the bit.