New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Nonstop Autonomous Mode

by andylizf in andylizf/nonstop

Lets Claude keep working while you are away, after a pre-flight that surfaces blockers, risky operations and open decisions, with a stop hook scoped to the session.

MITAuto-check: warningsAgent Workflows

Install Nonstop Autonomous Mode

The automated check flagged lines worth reading first. See the safety section below.

skills CLI
$ npx skills add andylizf/nonstop --skill nonstop -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install andylizf/nonstop nonstop --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
nonstop
GitHub stars
263
Token cost
~2k tokens
SKILL.md length
1,102 words
Files
10
Skills in repo
1
Repo updated
First seen
Licence
MIT

At a glance

Lets Claude keep working while you are away, after a pre-flight that surfaces blockers, risky operations and open decisions, with a stop hook scoped to the session.

  • Works in 4 steps: Mental Simulation — Anticipate Every… → Dangerous Operations Manifest → Permissions Check → …
  • Handing a long task to the agent before going away from the keyboard
  • SKILL.md covers Activation — /nonstop or…, During Nonstop Mode — Blocker…, Deactivation — /nonstop off and Auto-deactivation, plus 1 more section
  • Runs Shell scripts from its folder

What it does

Activated with /nonstop and switched off with /nonstop off, the skill makes the agent finish a pre-flight before any work, because nobody will be around to answer questions later. The first phase is a task-specific mental simulation: break the task into steps, imagine each one running, and surface anything that would make the agent stop, such as ambiguous requirements, missing credentials, decision points or slow external dependencies, so you can settle them up front.

The second phase presents a Dangerous Operations Manifest of high-risk actions that might be needed, such as killing processes, deleting files or directories and dropping tables, and asks for an explicit yes or no per relevant category. The repo also ships nonstop.sh, install.sh and a plugin marketplace file, and the description mentions a blocker decision framework; the excerpt was cut off before the remaining categories and that framework.

When your agent uses it

  • Handing a long task to the agent before going away from the keyboard
  • Letting a task run overnight with risky operations approved in advance
  • Surfacing missing credentials and ambiguous requirements before leaving
  • Switching autonomous mode off once you are back

Example prompts

  • “/nonstop I am going AFK, so finish migrating the test suite to the new runner.”
  • “Go nonstop on the build fix, and list anything you would need to ask me before I leave.”
  • “Turn off nonstop mode, I'm back at my desk.”

Workflow steps

4 steps, taken from the step headings in SKILL.md.

  1. Mental Simulation — Anticipate Every Blocker
  2. Dangerous Operations Manifest
  3. Permissions Check
  4. Confirm and Activate

What it can do on your machine

Read from SKILL.md and the folder at commit 9511b6f. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships script files (Shell), which the agent can run.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Nonstop Autonomous Mode loads about 2k tokens when it runs. Until then it costs about 115 tokens; SKILL.md has 1,102 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~115
When it runs · the whole SKILL.md, loaded when a task matches
~2k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check: warnings

The automated check found patterns that need a careful read before installing.

  • WarningContains instruction-override wording (e.g. “without asking the user”)SKILL.md:114
    te-force retry the same failing action, disable safety checks to bypass a blocker, make destructive changes to work arou

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from andylizf/nonstop at commit 9511b6f, republished under its MIT licence (© andylizf). 1,102 words, ~2,006 tokens.

Download SKILL.mdSave it as .claude/skills/nonstop/SKILL.md (or your agent's skills folder). This skill also uses 9 other files; get the full folder from GitHub.
name
nonstop
description
Autonomous work mode that lets Claude work continuously while the user is away. Activates pre-flight risk assessment, dangerous-ops approval, and a blocker decision framework. Use "/nonstop" to activate, "/nonstop off" to deactivate. Use this skill whenever the user says things like "go nonstop", "keep working", "don't stop", "I'm going AFK", "work while I sleep", "finish this without me", or wants Claude to work autonomously without interruption.

Nonstop Mode

A session-scoped autonomous work mode. When activated, Claude works continuously without stopping to ask the user, handling blockers intelligently.

Activation — /nonstop or /nonstop on

When the user triggers nonstop mode, start the pre-flight sequence immediately. Drop whatever else you are doing — do not wait for background agents, pending reads, or other in-progress work. The pre-flight is the priority now.

Complete the pre-flight before doing any work. The pre-flight exists because once the user walks away, there is no one to answer questions — every ambiguity must be resolved now.

Phase 1: Mental Simulation — Anticipate Every Blocker

Mentally simulate the entire execution path of the task from start to finish. This is not a generic checklist — it is a task-specific dry run.

How to think:

  1. Break the task into concrete steps (what will you actually do, in order?)
  2. For EACH step, imagine yourself executing it right now. Ask: "What could go wrong here? What would make me stop and ask the user?"
  3. Think about the environment: what machines, services, credentials, permissions does this touch?
  4. Think about dependencies between steps: if step 3 fails, can step 4 still run?
  5. Think about the edges: what if the test fails? what if the build takes 30 minutes? what if a file doesn't exist?

Surface to the user:

  • Anything that would make you stop and ask — surface it NOW
  • Ambiguous requirements where you'd need to make a judgment call — clarify NOW
  • Credentials, tokens, permissions you might need — confirm NOW
  • Decision points with multiple valid paths — pick one together NOW
  • External dependencies that might be slow or unavailable — plan a fallback NOW
  • Any step where you're not 100% sure how to proceed — ask NOW

The goal: after this phase, there should be ZERO reasons to stop and ask the user anything.

Present your findings as: "Here's what I plan to do [ordered list], and here's what might block me [list with proposed solutions]. Can you confirm/clarify these before you go?"

Phase 2: Dangerous Operations Manifest

Present a Dangerous Operations Manifest — high-risk actions that MIGHT be needed during execution. Get explicit yes/no for each relevant category:

CategoryExamples
Kill processesKill others' GPU jobs, restart services
Delete files/dirsrm -rf, git clean, drop tables
Force push / rewrite historygit push --force, git reset --hard
Deploy / releasePush to prod, publish package, merge to main
Modify infraChange cloud resources, modify CI/CD, edit configs
Social / commsPost GitHub comments/issues/PRs, send Slack messages, reply to reviews
External APIsCall paid APIs, trigger webhooks, write to external DBs
Resource contentionClaim GPU, take locks, use large disk/memory

The key principle: anything that leaves a trace in the outside world that others can see or that costs money needs approval. Reading is always fine; writing/posting/sending is what needs a decision.

Only present categories that are relevant to the task. For each approved category, record the scope (e.g. "OK to kill MY GPU processes but not others'"). The user may also say "approve all" — that's fine, it's their call.

For anything NOT approved: do not attempt it. Create a task marked as blocked with the reason, and move on.

Phase 3: Permissions Check

Ask the user: "What's your current permission mode? (check the bottom status bar)"

  • Bypass permissions → good to go, no prompts will interrupt.
  • Auto mode → mostly fine. The auto classifier may block some operations; Claude will attempt alternatives when that happens.
  • Default / other → warn: "Permission prompts will block me when you're not here. I recommend switching to auto mode (Shift+Tab to cycle) or bypass before you leave."
Show full SKILL.md (512 more words)Show less
Phase 4: Confirm and Activate

Present the summary and ask the user to confirm. Do NOT activate until the user explicitly replies.

  1. Scope: "Here's what I'll do: [list]. Anything to add or remove?"
  2. Workaround policy: "If I hit a blocker, should I: (a) skip and continue, (b) attempt a workaround, or (c) stop and wait?"
  3. Fallback for unapproved ops: "Dangerous ops you didn't pre-approve will be marked as blocked and skipped. OK?"

STOP HERE. Wait for the user to reply. Do not proceed until you receive an explicit confirmation (e.g. "go", "yes", "confirmed", "looks good"). The user triggering /nonstop is NOT confirmation — it starts the pre-flight, not the work.

Once the user confirms, activate:

bash
mkdir -p ~/.claude/hooks/state && touch ~/.claude/hooks/state/nonstop.activate

This signals the stop hook to create a session-scoped flag on your next turn end. The hook knows your session ID and will handle the rest.

Say: "Nonstop mode ON. Go rest — I've got this."

During Nonstop Mode — Blocker Decision Framework

When you encounter something that would normally make you stop and ask the user, follow this decision tree:

Level 1: Can I solve it myself?
  • Missing info → search the codebase, read docs, check git history
  • Test failing → read the error, fix it
  • Build error → diagnose and fix
  • If yes → solve it and move on. Don't stop.
Level 2: Can I work around it?
  • Permission denied → check if there's an alternative path (different API, different approach)
  • External service down → skip that part, continue with other tasks
  • Ambiguous requirement → pick the most reasonable interpretation, document your choice
  • Critical check: will the workaround change the outcome of the task?
    • If the result stays equivalent → proceed with workaround, note what you did in the task
    • If the result would be materially different → do NOT proceed, mark the task as blocked
Level 3: Truly blocked
  • Update the task status to blocked, noting: what you tried, why it failed, what the user needs to do
  • Move on to the next task. Do NOT spin on unsolvable problems.
  • NEVER: brute-force retry the same failing action, disable safety checks to bypass a blocker, make destructive changes to work around permissions, or guess at credentials/tokens.
Subagent Delegation
  • For long-running operations (builds, tests, deploys), spawn a background Agent to monitor
  • Don't block the main thread waiting for results
  • Continue with other work while subagents run

Deactivation — /nonstop off

  1. Signal deactivation:

    bash
    touch ~/.claude/hooks/state/nonstop.deactivate

  2. Present a summary from the task list:

    • Tasks completed
    • Tasks where you took workarounds (and why)
    • Tasks marked as blocked that need the user's attention

  3. Say: "Nonstop mode OFF. Here's what happened while you were away."

Auto-deactivation

Nonstop mode auto-deactivates after NONSTOP_MAX nudges (default 5, env var NONSTOP_MAX, 0 = unlimited). This prevents runaway loops when Claude is genuinely stuck on everything.

Technical Details

  • Activation handshake: Claude touches nonstop.activate → stop hook creates session-scoped nonstop-<session_id>.active (hook knows the session ID from its JSON input, Claude doesn't need to)
  • Deactivation handshake: Claude touches nonstop.deactivate → stop hook removes the session-scoped flag
  • stop_hook_active flag prevents infinite loops (allows stop on 2nd attempt per turn)
  • Nudge counter tracks blocks per session; auto-cleans on deactivation

© andylizf, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 9 other files in the repository root of andylizf/nonstop.

  • SKILL.md
  • .claude-plugin/marketplace.json
  • .github/workflows/ci.yml
  • .markdownlint.json
  • INSTALL.md
  • LICENSE
  • README.md
  • TODO.md
  • install.sh
  • nonstop.sh

Open the folder on GitHubat commit 9511b6f

Compare with similar skills

Nonstop Autonomous Mode next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Nonstop Autonomous Mode compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Nonstop Autonomous Mode this skillandylizf/nonstop263—~2kAutomated safety check: WarnMIT
Show Me Your Work Decision Logcursor/plugins10k9 repos~1.6kAutomated safety check: PassNone
Plugin CreatoruseLexora/Lexora548—~1kAutomated safety check: PassAGPL-3.0
Loop FactoryJuliusBrussee/skills161—~2kAutomated safety check: PassMIT
Superloopy Doctorbeefiker/superloopy1111 repos~1.9kAutomated safety check: PassMIT
Smithers Agent Runtimesmithersai/smithers429—~765Automated safety check: PassMIT

Similar skills

  • Show Me Your Work Decision Log

    cursor/plugins

    Official

    Keeps a TSV decision log for long or unattended agent runs, one row per decision with what, why, evidence and result, so a reviewer can check the work later.

    10k GitHub starsUsed in 9 repos~1.6k tokens
    Agent WorkflowsAuto-check passed

  • Plugin Creator

    useLexora/Lexora

    创建、修改并验证可安装的 Lexora 桌面插件，包括工作台装饰、导航与面板、文件阅读工具、控件样式、网络查询、通知和定时命令。也响应 $plugin-creator。普通任务执行、使用现有插件、其他产品的插件开发不属于此技能。

    548 GitHub stars~1k tokensUpdated 9 days ago
    Agent WorkflowsAuto-check passed

  • Loop Factory

    JuliusBrussee/skills

    Run a spec-driven agent loop where coding tasks live as markdown specs that move through inbox → active → archive, get implemented by Claude Code or Codex, and pass a review gate before they count…

    161 GitHub stars~2k tokensUpdated 2 mo ago
    Agent WorkflowsAuto-check passed

  • Superloopy Doctor

    beefiker/superloopy

    A read-only health check for a Superloopy install or checkout: state folders, evidence files, bundled agents, hooks, wrapper and plugin registration, reported before any repair.

    111 GitHub starsUsed in 1 repo~1.9k tokens
    Agent WorkflowsAuto-check passed

  • Smithers Agent Runtime

    smithersai/smithers

    Guides changes to the Smithers agent loop's host wiring: credentialed model seats, budget accounting, sandbox limits and flow-call authorization.

    429 GitHub stars~765 tokensUpdated yesterday
    Agent WorkflowsAuto-check passed

  • Guard Mode

    garrytan/gstack

    Switches on full safety by combining warnings before destructive commands with a block on edits outside one directory you choose, for work on production or live systems.

    136k GitHub stars~919 tokensUpdated yesterday
    Agent WorkflowsAuto-check: notes

Categories

Questions about Nonstop Autonomous Mode

What does Nonstop Autonomous Mode do?

Lets Claude keep working while you are away, after a pre-flight that surfaces blockers, risky operations and open decisions, with a stop hook scoped to the session. Activated with /nonstop and switched off with /nonstop off, the skill makes the agent finish a pre-flight before any work, because nobody will be around to answer questions later. The first phase is a task-specific mental simulation: break the task into steps, imagine each one running, and surface anything that would make the agent stop, such as ambiguous requirements, missing credentials, decision points or slow external dependencies, so you can settle them up front.

When should I use Nonstop Autonomous Mode?

Nonstop Autonomous Mode fits situations like: handing a long task to the agent before going away from the keyboard; letting a task run overnight with risky operations approved in advance; surfacing missing credentials and ambiguous requirements before leaving; switching autonomous mode off once you are back.

How do I install Nonstop Autonomous Mode in Claude Code?

Run `npx skills add andylizf/nonstop --skill nonstop -a claude-code`. Or copy the skill folder (the andylizf/nonstop repository) into .claude/skills/nonstop in your project. Claude Code loads it when a task matches its description.

How do I install Nonstop Autonomous Mode in Codex?

Run `npx skills add andylizf/nonstop --skill nonstop -a codex`. Or copy the skill folder (the andylizf/nonstop repository) into .agents/skills/nonstop in your project. Codex loads it when a task matches its description.

Can I use Nonstop Autonomous Mode in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add andylizf/nonstop --skill nonstop -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/nonstop, .gemini/skills/nonstop, .github/skills/nonstop and .opencode/skills/nonstop in your project.

What does Nonstop Autonomous Mode need to run?

Going by SKILL.md and its folder, Nonstop Autonomous Mode needs a shell for the scripts in its folder.

Does Nonstop Autonomous Mode access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Nonstop Autonomous Mode safe to install?

Our automated static check of SKILL.md flagged 1 warning(s): contains instruction-override wording (e.g. “without asking the user”). Read the flagged lines before installing; the check is not a guarantee either way.

What licence does Nonstop Autonomous Mode use?

Nonstop Autonomous Mode is published under the MIT licence (from the LICENSE file in the skill folder). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Nonstop Autonomous Mode use?

About 2k tokens (SKILL.md is roughly 8k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Nonstop Autonomous Mode?

Skills that share tags, products or a category with Nonstop Autonomous Mode: Show Me Your Work Decision Log (cursor/plugins, 10k stars), Plugin Creator (useLexora/Lexora, 548 stars), Loop Factory (JuliusBrussee/skills, 161 stars) and Superloopy Doctor (beefiker/superloopy, 111 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Nonstop Autonomous Mode?

andylizf (a GitHub user) maintains it in andylizf/nonstop, which has 263 GitHub stars. The repository was last updated on April 4, 2026.

Source: andylizf/nonstop on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.