Sober Reply
Reply sober when you're not — clear thinking is enough, neat typing is optional.
SKILL.md in the open agentskills.io standard — works directly in Claude, ChatGPT/Codex, Cursor, Copilot & more.
Sober Reply is an AI skill for clear, safe replies in hard-to-write moments — a low-input, plain-text and safety mode that understands messy input, typos and half-formed thoughts when you're drunk, tired, stressed, sick or simply not in writing shape. It keeps replies short, asks only light questions, de-escalates messages and pushes risky actions like sending, posting, buying or deploying to a sober moment. With crisis & emergency protection and one firm rule: messy input never automatically means "drunk".
What this skill does
- Understands messy input, typos and half-formed thoughts and quietly reconstructs your intent — without correcting or lecturing
- Picks the right lane via a state router (clear head/broken input, short mode, collect, emotional, risky, very messy)
- Replies short: at most one follow-up, thinks long and answers briefly
- Brakes risky, irreversible or social actions (send, post, buy, book, deploy, delete, quit, text an ex/boss) — clearly, but without patronizing, with a user override
- De-escalates emotional messages and offers a calm morning version or a private vent space
- Saves work-in-progress as a morning note and closes sessions on a positive note ("Good stopping point for today")
- Detects crises (self-harm, danger to others) and alcohol emergencies and points to real emergency services and verified crisis lines — instead of a joke or a morning note
- Never treats messy input as proof of intoxication — no diagnosis, fair to tiredness, voice input, tiny keyboards and motor limitations
- Uses device and keyboard-layout logic (QWERTZ/QWERTY/AZERTY/phone/voice) for better reconstruction
- Produces clean copyblocks only where you'll actually reuse the text — no joke, analysis or warning inside the block
Description
Sober Reply is an AI skill for clear, safe replies in moments of low input quality or low decision confidence — it rescues your thought, turns mush into plain text, and brakes what you might regret tomorrow. "Sober" here means clear, calm and considered, not just "not drunk": the skill helps when you're drunk, exhausted, sick, stressed, emotionally charged, out and about on your phone, or limited in your motor input. Its promise is simple — **clear thinking is enough, neat typing is optional.** Instead of correcting typos or lecturing, Sober Reply quietly reconstructs your intent, replies short (thinks long, answers briefly), asks at most one follow-up, and ends useful. A **state router** moves into the right lane: "clear head, broken input" (just understand, don't correct), "low reading patience" (short mode), "creative but not decision-ready" (collect yes, decide final tomorrow), "emotionally charged" (de-escalate, don't escalate), "risky action" (brake clearly, safe alternative) and "very messy" (only yes/no, nothing important). The **safety brake** kicks in on irreversible or social actions — texting an ex/crush/boss/client, sending an application or resignation, posting, spending money, contracts, domains, deploying, deleting, changing live systems. Sober Reply doesn't patronize: it slows you down, explains the risk and leaves the call to you (you can override) — and always offers a **morning version**, a safe preview, or a plain note. Two research-backed protective layers set the skill apart. First, a **crisis & emergency layer**: when it's about self-harm, danger to others, or an alcohol emergency (someone unresponsive, vomiting while unconscious, breathing shallow), the light tone drops — no joke, no morning note, but a pointer to real help (emergency 112/911/999 and verified crisis lines like 988, 116 123, 3114, 024, 0800 111 0 111), calm and present. Second, a **fairness layer**: messy input is never proof of intoxication. Tiredness, small keyboards, swipe/autocorrect errors, voice input, dyslexia, tremor or other motor limitations produce the same "mush" patterns — Sober Reply makes no diagnosis, never says "you're drunk" (unless you say it), and takes you seriously. Device and keyboard-layout logic (QWERTZ/QWERTY/AZERTY, phone, voice) sharpens the silent reconstruction without pinning you to a cause. Sober Reply is the skill that carries you through the moment you're not in writing shape — and lifts the important stuff to tomorrow, when it's better decided.
Examples
What it does not do
- Doesn't send, post, buy or deploy anything itself — it only writes text you copy; it never performs risky actions
- Makes no diagnosis and never says "you're drunk" unless you say so — messy input is no proof of any state
- Doesn't replace professional help in crises or medical emergencies — it leads you there, it doesn't treat
- Doesn't patronize: it brakes risky actions clearly but doesn't lock them — the decision stays yours (except in real danger)
- Doesn't correct individual typos or give spelling lectures as long as the meaning is clear
- Doesn't sharpen hurtful, threatening or humiliating messages — it helps de-escalate, not escalate
- Doesn't store lasting personal preferences or final decisions from an impaired state without later confirmation
Compatibility & tech
- Tested (internal)
- 8 scenarios
- Modes
- Clear head, broken input (understand silently) · Short mode (low reading patience) · Collect (creative but not decision-ready) · De-escalate (emotionally charged) · Brake (risky action) · Very messy (yes/no only, wrap up) · Morning mode (save the work-in-progress) · Crisis & emergency protection (real help, not a joke)
- Inputs
- Messy / error-filled text (any language) · Half-formed thoughts, keywords, voice-note transcripts · A draft of a risky message (to de-escalate) · Device / input method (optional, improves reconstruction)
- Output format
- A short, calm Markdown reply (understanding → next safe step → one option/question); on request real buttons/chips or numbered options, a morning note, and a clean copyblock to paste
- Subcategory
- Safe writing & impulse control
- License
- Proprietär
Security profile
Runs entirely on your machine with your own AI — no external runtime, no running costs.
Only instructions, templates and references — no executable scripts.
Works offline with what you provide — does not call external services on its own.
What you get
- sober-reply-1.0.0/9 files
- .claude-plugin/marketplace.json
plugins/sober-reply/skills/sober-reply/9 files
- SKILL.md
- manifest.json
references/7 files
- sober-reply-checklist.md
- sober-reply-crisis.md
- sober-reply-input.md
- sober-reply-morning.md
- sober-reply-output.md
- sober-reply-safety.md
- sober-reply-states.md
- .agents/skills/sober-reply/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
- LICENSE.txt
Installation
Also works as a chat prompt
No AI tool? Paste it into Claude, ChatGPT or Gemini and use the method right away.
This skill ships no scripts, so the prompt carries its full method.
Installing is the full version — it triggers automatically, runs its scripts and loads references as needed. As a chat prompt you drive the method by hand.
Unlock to copy the ready-to-paste prompt — then in “My Skills”.
Reviews
No reviews yet – be the first.
Note
Not a substitute for professional or medical help, and not a sobriety/BAC test. For self-harm, danger to others or a medical emergency, always go to emergency services first (112 in the EU, 911 in the US/Canada, 999 in the UK) and a crisis line. Sober Reply makes no diagnosis; messy input is never proof of any particular state.
Changelog
- v1.0.030.06.2026First release: a low-input, plain-text and safety mode for hard-to-write moments. State router (6 lanes from 'clear head, broken input' to 'very messy'), error/device/keyboard-layout logic (QWERTZ/QWERTY/AZERTY/phone/voice), a safety brake for irreversible actions (send/post/buy/deploy/delete/quit) with user override, a morning mode + positive session closure, copyblock rules. Two research-backed protective layers: (1) a crisis & emergency layer (self-harm, danger to others, alcohol emergency) with real, verified helplines (112/911/999, 988, 116 123, 3114, 024, 0800 111 0 111) instead of a joke or a morning note; (2) a fairness layer 'messy input ≠ drunk' — accessibility, motor difficulty, dyslexia, voice input and tiny keyboards are never read as intoxication, no diagnosis. 7 references, realistic examples incl. a calibration case, internal evals. Abuse-resistant: won't sharpen hurtful messages.
Frequently asked questions
What does Sober Reply do?
Sober Reply is an AI skill for clear, safe replies in hard-to-write moments — a low-input, plain-text and safety mode that understands messy input, typos and half-formed thoughts when you're drunk, tired, stressed, sick or simply not in writing shape. It keeps replies short, asks only light questions, de-escalates messages and pushes risky actions like sending, posting, buying or deploying to a sober moment. With crisis & emergency protection and one firm rule: messy input never automatically means "drunk".
How do I text while drunk without sending something I'll regret?
Sober Reply understands your messy text even with typos, keeps replies short and brakes risky actions like sending or posting. Instead of firing the message off, Sober Reply offers a calm morning version or just saves the thought — you decide sober.
Is Sober Reply only for drunk people?
No. Sober Reply is for any hard writing moment — tired, stressed, sick, emotional, on your phone or with a motor impairment. 'Sober' means clear and level-headed, not just 'not drunk'. Alcohol is a real use case, but not the only one.
Does Sober Reply send my message for me?
No. Sober Reply only writes and de-escalates text that you copy and send yourself — it never performs risky actions like sending, buying or deploying. It deliberately brakes them and offers a safe preview or a morning version instead.
Does Sober Reply treat my typos as proof I'm drunk?
No. Sober Reply never reads messy input as intoxication and makes no diagnosis. Typos also come from tiredness, small keyboards, voice input, dyslexia or motor difficulties — Sober Reply rescues the thought without judging you.
What does Sober Reply do if someone is really in a bad way?
If it's about self-harm, danger to others or an alcohol emergency, Sober Reply drops the light tone, makes no morning note and points to real help — emergency services (112/911/999) and verified crisis lines (e.g. 988, 116 123, 3114, 024, 0800 111 0 111). It doesn't replace professional help; it leads you to it.
Can I use Sober Reply on my phone with autocorrect?
Yes. Sober Reply spots phone-typical errors like autocorrect mangling, swipe slips and missing accents, quietly reconstructs your intent, and hands you a clean copyblock to paste — ideal when typing on the go is hard.
Which AI tools does Sober Reply work with?
Claude (Projects & system prompt) · ChatGPT (Custom GPT & custom instructions) · Any LLM · Native UI (buttons/chips) where available, Markdown fallback otherwise · Runs locally, no network access needed
How do I use Sober Reply?
Sober Reply is a SKILL.md in the open agentskills.io standard: install it with one command (npx) or download it and add it to your AI tool — Claude (Projects), ChatGPT (Custom GPT), Cursor, Copilot, Gemini CLI and more. No code needed.