Guide · Agent skill

How to use the Run Without You agent skill

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

Run Without You is an agent skill for Claude that takes an agent you have already built and graded and makes it run on its own. It can set up a recurring cron schedule, an event-driven web trigger, or on-demand use, and it writes a roadmap file for later upgrades. This guide explains what the skill does, how to install it, how to ask for it in plain language, and what to check before you leave an agent running unattended.

Run Without Youalirezarezvani/claude-skills · MIT Download the skill

Quick start

1Download and unzip the skill. Download the ZIP from the directory page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps, upload it under Settings → Capabilities → Skills.
2Prepare your agent and build sheet. Confirm you have a graded Claude Managed Agent with its agent ID and environment ID, plus the build-sheet.json from the earlier phases.
3Validate the cron and timezone. Run cron_validator.py with your 5-field cron expression and IANA timezone. Read the DST note and fix any exit-code error before continuing.
4Build the deployment payload. Run deployment_builder.py with --nest-outcome to write the POST /v1/deployments payload and get the BYOK curl commands.
5Test with one manual run. Use the curl command for the manual run endpoint, read the verdict, and only then keep the cron schedule in place.
6Write the roadmap and hand off. Run next_directions_writer.py to create NEXT-DIRECTIONS.md, then move the goal state to wrap-up and invoke the wrap-up skill.

What is an agent skill, and what does Run Without You do?

An agent skill is a folder of instructions and scripts that an AI agent loads when a task matches it. The skill's own description says it is Phase 4 of building a Claude Managed Agent, the step where the agent stops needing you to start it.

The skill takes a graded agent and turns it into one of three shapes: a recurring scheduled deployment that uses POSIX cron, an event-driven curl trigger, or on-demand use where you re-send a message yourself. For the recurring case, each scheduled run can include an outcome so that every firing grades itself.

Recurring cron deployment: for requests like 'every morning', 'weekly' or 'nightly'.
Event-driven curl trigger: for 'when X happens'. The source describes this as documented, not scheduled.
On-demand: for 'only when I ask'. No deployment is created.

Install the Run Without You skill

Download the ZIP for the skill from the directory page. Then choose one of two locations. For a personal install available to all your projects, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps instead of Claude Code, you can upload the skill under Settings → Capabilities → Skills.

Personal: ~/.claude/skills/
Project only: .claude/skills/ inside the project folder
Claude apps: Settings → Capabilities → Skills → upload

Check what you need before you start

The skill's scripts use only Python 3 standard library, so there are no packages to install for them. You also need an existing Claude Managed Agent with its agent ID and environment ID, an Anthropic API key that you supply yourself for the curl commands, and a build-sheet.json file produced by the earlier phases of the agent build.

The scripts do not handle your API key. The deployment is created by you, using curl with your own key (BYOK).

Python 3
An existing Claude Managed Agent, with agent ID and environment ID
Your own Anthropic API key
A build-sheet.json from the earlier build phases

Set up a recurring schedule step by step

Start by validating the cron expression and timezone. The validator checks a 5-field cron expression and an IANA timezone, and prints a note about daylight saving time. An invalid cron or timezone makes the script exit with code 1.

Next, build the deployment payload. The builder writes a POST /v1/deployments payload to a file and prints the BYOK curl commands to create the deployment and to test it once with the manual run endpoint. Use --nest-outcome so that each firing includes the rubric and self-grades.

Before you trust the schedule, fire one manual run and read the verdict. Only then leave the cron in place. Once it passes, the skill advises pinning the agent version in the deployment.

Validate the schedule and read the DST note.
Build the payload with --nest-outcome.
Run the test once manually and read the verdict.
Pin the agent version after the test passes.

python3 scripts/cron_validator.py --cron "0 9 * * *" --timezone Europe/Berlin
python3 scripts/deployment_builder.py \
  --sheet ./my-agent/build-sheet.json --agent-id agent_123 --env-id env_456 \
  --nest-outcome --out ./my-agent/payloads/deployment.json

Finalize the roadmap file

The last build step writes NEXT-DIRECTIONS.md, a versioned roadmap for future upgrades. It is generated from your build sheet and the loop shape you choose. After it is written, the skill tells you to advance the goal state to wrap-up and invoke the wrap-up skill.

python3 scripts/next_directions_writer.py \
  --sheet ./my-agent/build-sheet.json --loop-shape cron-loop --last-verdict satisfied --out-dir ./my-agent

Safety notes: scripts, network and unattended runs

The skill runs Python scripts, so read them before you run them. The skill's own metadata says it runs scripts. No tool in the skill makes API calls itself, so the deployment only exists after you run the curl command with your key.

Unattended agents can spend money and take actions without you. The skill says Claude Managed Agents has no spend cap of its own, so you need to set safety rails. The skill recommends an always_ask setting for MCP tools, limited networking where possible, read_only untrusted memory, a max_iterations limit per firing, and a workspace spend limit.

No tool makes API calls; you create the deployment with curl and your own key.
There is no spend cap inside Claude Managed Agents, so set a workspace spend limit.
Set max_iterations per firing and use always_ask for MCP tools.
Limit networking where you can and keep untrusted memory read_only.

Tips for schedules that behave

Daylight saving time changes the wall clock. Spring-forward skips some local times and fall-back repeats others. The skill advises avoiding 02:00 to 03:00 in DST zones if you need each run to happen exactly once.

Deployments are capped at 1,000 per organization, according to the skill notes. The skill also recommends starting on demand and moving to a cadence only after the agent has been graded.

Avoid 02:00 to 03:00 local time in DST zones.
Start on-demand, then schedule once the agent is graded.
Keep one deployment per purpose and check the count against the 1,000 limit.

What people use it for

Run a research or monitoring agent every morning in your own timezone and have each run grade its own output.
Set up a weekly agent that produces a report on a fixed schedule.
Create an event-driven trigger that starts the agent from an outside web request, using the documented curl pattern.
Keep an agent on on-demand use while it is still being tuned, then move it to a schedule once it passes grading.

Frequently asked questions

How do I install the run-without-you skill?

Download the ZIP from the directory page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project. In the Claude apps, upload it under Settings → Capabilities → Skills.

What do I need to use run-without-you?

You need Python 3, an existing Claude Managed Agent with its agent ID and environment ID, your own Anthropic API key for the curl commands, and a build-sheet.json from the earlier build phases. The scripts use only the Python standard library.

Is run-without-you free, and what is its licence?

The skill is published under the MIT licence. The source does not describe any pricing, and Claude Managed Agents usage is billed through your own Anthropic account, which the source does not detail.

Why does my cron schedule fail or fire twice?

Daylight saving time can skip or repeat local times. The skill's validator prints a DST note, and it advises avoiding 02:00 to 03:00 in DST zones if you need exactly-once runs. An invalid cron expression or timezone makes cron_validator.py exit with code 1.

Does the skill create the deployment for me?

No. No tool in the skill makes API calls. The builder writes the payload and prints the curl commands, and you run them with your own API key.

How does run-without-you compare with other ways to schedule an agent?

The source does not compare this skill with other scheduling tools. Its stated role is the final phase of a Claude Managed Agent build, and it is distinct from the grade-iterate loop and the wrap-up closeout.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Run Without You suits developers who already have a working Claude Managed Agent and want it to run on a schedule, with a validated timezone and a roadmap file at the end of the build. It is a poor fit for anyone without Managed Agents access or who is uncomfortable running curl with an API key, and for unattended agents that have no spend or permission limits.

Download the skill