Skip to content
Get packs

Guide · Agent skill

How to use the Agentic OS Builder agent skill (agentic-os)

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

The agentic-os skill teaches Claude Code to work as a long-running assistant system instead of a single chat session. It describes a project layout with a central instruction file that routes work to specialist agent prompts, slash commands for repeatable routines, and a memory store made of plain JSON and Markdown files. After this guide you will know what the skill sets up, how to install it, how to trigger it, and what to check before you schedule anything to run unattended.

Agentic OS Builderaffaan-m/ECC · MIT

Quick start

  1. 1Get the skill. Download the ZIP from this page and confirm it contains SKILL.md.
  2. 2Install it. Unzip into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder. In the Claude apps, upload it under Settings → Capabilities → Skills.
  3. 3Create the project folders. In your project root, create CLAUDE.md, agents/, .claude/commands/, scripts/ and data/ as shown in the skill's layout.
  4. 4Write the kernel and first agent. Add an agent registry table to CLAUDE.md and create one specialist file such as agents/dev.md with its identity, memory scope and constraints.
  5. 5Add a slash command. Save a workflow like daily-sync as .claude/commands/daily-sync.md, then run /daily-sync in Claude Code to test it.
  6. 6Test before scheduling. Run the command by hand, review the output in data/, and only then set up a LaunchAgent, systemd timer or pm2 job.

What does the agentic-os skill make Claude Code do?

Normally each Claude Code session starts from scratch. This skill asks Claude to treat the project as a persistent runtime: it reads stored context at the start of a session, routes requests to the right specialist, and writes results back to files so the next session can continue.

The skill defines four layers, each a directory in your project root. The kernel is a CLAUDE.md file that acts as a coordinator. Specialist agents are Markdown prompt files in agents/. Slash commands live in .claude/commands/. Scripts and state sit in scripts/ and data/.

  • Kernel (CLAUDE.md): identity, routing rules, model policy and an agent registry
  • Agents (agents/): specialist roles such as @dev, @writer, @researcher and @ops, each with a memory scope, tool access and constraints
  • Commands (.claude/commands/): repeatable workflows such as /daily-sync
  • Scripts (scripts/): daemon scripts triggered by cron or webhooks
  • State (data/): append-only logs, project files and decision records, with no external database
project-root/ ├── CLAUDE.md # kernel: identity, routing, agent registry ├── agents/ # specialist agent prompts (Markdown) ├── .claude/commands/ # slash commands ├── scripts/ # scheduled or event-driven scripts └── data/ # state: JSON/Markdown files, no external DB

How do you install the agentic-os skill?

Download the ZIP file from this page. The package contains a single file, SKILL.md. You can install it in one of two ways.

For Claude Code, unzip the download into ~/.claude/skills/ to make it available everywhere, or into .claude/skills/ inside a specific project to limit it to that project. In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

  • Personal install: unzip into ~/.claude/skills/
  • Project install: unzip into the project's .claude/skills/ folder
  • Claude apps: Settings → Capabilities → Skills, then upload the skill

How do you trigger the skill with natural requests?

You do not need a special command to use the skill. Describe what you want in plain language. The skill's description matches requests about building multi-agent workflows, persistent automation, a 'personal OS' or 'agentic OS', agent coordinators, and long-term projects whose context must survive between sessions.

Good trigger phrases include asking Claude to set up a multi-agent workflow, to keep project context across restarts, or to structure a long-running project. Be specific about the parts you want, such as the agent roles or the routines.

  • Set up a multi-agent workflow for my project with a developer and a writer agent.
  • Create a persistent personal OS for my weekly routines.
  • Structure this long-term project so context carries over between sessions.
  • Build an agent coordinator that routes my requests to specialists.

What does a typical setup look like?

The kernel file CLAUDE.md holds an agent registry table. Each row names an agent, its role, and the words in a request that trigger it. Claude parses the request, matches the trigger column, loads the matching file from agents/, and then returns an integrated answer.

A daily briefing is the example the skill uses most. The /daily-sync command reads the last sync log, checks git status, pending pull requests and CI health, reviews the inbox, and appends a summary to a dated log file. Save the command as .claude/commands/daily-sync.md and run it with /daily-sync.

# .claude/commands/daily-sync.md # /daily-sync Run the morning briefing: 1. Read data/logs/last-sync.md for context 2. Check project status: git status, pending PRs, CI health 3. Check data/inbox/ for new tasks or decisions needed 4. Produce a summary of blockers, priorities and next actions 5. Append the briefing to data/logs/daily/<date>.md

How does the file-based memory work?

Memory is plain files. The skill uses JSON for structured state and Markdown for notes meant to be read by people, and it creates no vector database, Redis or PostgreSQL. Subfolders under data/ hold daily logs, project context, decision records in ADR format, an inbox, contacts and reusable templates.

The skill also suggests recording a short reflection at the end of each session: what worked, what did not, and what to change. Over time this is meant to improve the system without code changes. When you evolve a JSON schema, add new fields and mark old ones as deprecated rather than renaming them, so older files still read correctly.

  • data/daily-logs/: append-only daily activity logs
  • data/projects/: one context file per project
  • data/decisions/: architecture and business decisions
  • data/inbox/: new tasks and ideas waiting for triage
  • data/contacts/ and data/templates/

How do you schedule routines to run automatically?

The skill schedules tasks with an external scheduler, not Claude Code's built-in cron, because a built-in schedule stops when the session ends. The source gives three options: a macOS LaunchAgent plist in ~/Library/LaunchAgents/, a systemd user timer on Linux, or pm2 on any platform, which needs Node.js. Each one runs the claude command with --cwd pointing at your project and --command set to a slash command such as /daily-sync.

Write these files yourself based on the skill's examples and adjust the paths and times for your machine. Test the command manually before you schedule it.

  • macOS: LaunchAgent in ~/Library/LaunchAgents/ with StartCalendarInterval
  • Linux: systemd user service and timer in ~/.config/systemd/user/
  • Cross-platform: pm2 with cron_restart in an ecosystem.config.js file
# ~/.config/systemd/user/agentic-daily-sync.timer [Unit] Description=Run the daily sync every morning [Timer] OnCalendar=*-*-* 8:00:00 Persistent=true [Install] WantedBy=timers.target

What safety points should you check first?

Scheduled runs happen without anyone watching them. A routine that is allowed broad file or tool access can change files or take actions while you are away. Read every agent prompt and slash command before you schedule it, and limit what each agent is allowed to touch.

The skill also warns against hardcoding credentials in agent files or the kernel. The skill itself does not run scripts and does not use the network. However, the commands it describes can call external services if you configure them, such as analytics from Stripe or GitHub or MCP servers listed in .claude/mcp.json, so review those connections as well. Git tracking is recommended so you can see and revert changes made by agents.

  • Review all agent files and commands before scheduling them
  • Give agents only the tool access they need
  • Never store API keys or passwords in agent or kernel files
  • Keep the data folder under Git or a backup, since there are no database safeguards

What people use it for

  • A developer keeps project context and decision records across many Claude Code sessions.
  • A solo founder runs a daily briefing that summarises git status, pull requests and blockers.
  • A freelancer splits work between a writer agent for proposals and a researcher agent for client background.
  • Anyone wanting a worked example of organising several Claude Code agent roles with routing rules.

Frequently asked questions

Is the agentic-os skill free to use?

The skill is published under the MIT licence, which permits use and modification. You still need your own Claude Code setup to run it, and the optional scheduler and Node.js for pm2 are separate tools.

What do I need to run the agentic-os skill?

You need Claude Code. Git is recommended for tracking the kernel, agents and commands. A scheduler is optional: macOS LaunchAgent, systemd or pm2, where pm2 requires Node.js.

Where do I put the agentic-os skill files?

Unzip the downloaded ZIP into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings → Capabilities → Skills.

Does agentic-os need a database?

No. Its memory uses JSON and Markdown files in the data/ folder and explicitly avoids vector databases, Redis and PostgreSQL. That keeps setup simple but offers no database safeguards, so it is not suited to concurrent or shared access.

Why is my scheduled Claude task not running?

Check the scheduler first: confirm the LaunchAgent, systemd timer or pm2 job is loaded and that the --cwd path and slash command are correct. Test the same command manually in Claude Code and check the log file you configured, such as /tmp/agentic-daily-sync.log in the LaunchAgent example.

How does agentic-os compare with a normal Claude Code chat?

A normal chat usually starts fresh each time. Agentic-os adds stored context, routing to specialist agent prompts, repeatable slash commands and scheduled runs, which means more files to maintain and more to review for safety.

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

Use agentic-os if you want Claude Code to keep project context, routines and a file-based decision log across sessions, and you are comfortable writing and reviewing Markdown prompts by hand. Pick something else if you need a ready-made app, shared or access-controlled data, or unattended automation you cannot closely supervise.

Download the skill