Documentation

Learn the operating model for Agentic Manager: define endgoals, convert them into tickets, run distributed agents safely, and retry until outcomes are achieved.

There are multiple ways to connect an LLM to Agentic Manager depending on your tool. Pick the option that matches your setup.

Option A: Claude Code Skill

Recommended for Claude Code users. The skill auto-authorizes all API calls — no permission prompts.

Prerequisites

- Claude Code installed and working (claude available in your terminal)
- A project directory where you want to use Agentic Manager

Step 1: Install the skill into your project

Run this from your project root. It fetches the skill files from the API and writes them to the right place.

# From your project root — fetches skill files from the Agentic Manager API
curl -s https://api.agenticmanager.io/api/skills/claude-code | \
  node -e "
    const d = JSON.parse(require('fs').readFileSync('/dev/stdin','utf8'));
    for (const [p, c] of Object.entries(d.files)) {
      require('fs').mkdirSync(require('path').dirname(p), {recursive:true});
      require('fs').writeFileSync(p, c);
      console.log('  wrote', p);
    }
    for (const p of [
      'skills/agentic-manager/scripts/bootstrap-runtime.sh',
      'skills/agentic-manager/scripts/set-endgoal.sh',
      'skills/agentic-manager/scripts/sync-learnings.sh'
    ]) {
      if (require('fs').existsSync(p)) require('fs').chmodSync(p, 0o755);
    }
    console.log('Done! Run /agentic-manager in Claude Code to get started.');
  "

Requires curl and node (both standard on most dev machines). No GitHub access needed.

Step 2: Verify the directory structure

After installing, you should have the following in your project:

your-project/
  skills/
    agentic-manager/
      SKILL.md              # Main skill file (frontmatter + operating guide)
      references/
        api-guide.md        # Full API reference with request/response examples
        endgoals-playbook.md # Endgoal templates + retry loop playbook
      scripts/
        bootstrap-runtime.sh # Resolve AM_* runtime vars for any repo
        set-endgoal.sh      # Create or replace the active endgoal record
        sync-learnings.sh   # Sync learnings from API into SKILL.md

Claude Code discovers skills from the skills/ directory automatically — no config needed. Start each new project by running bootstrap-runtime.sh, then set the first endgoal with set-endgoal.sh.

Step 3: Invoke the skill

Open Claude Code in your project directory and invoke the skill manually or let Claude Code detect intent automatically:

/agentic-manager

Claude Code will load the skill, authenticate with your Agentic Manager instance, and start the workflow. You can also just describe what you want (e.g. "claim the next task", "create tickets for my plan") and Claude Code will invoke the skill automatically.

Option B: Prompt Generator

Works with any LLM. Generates a ready-to-use prompt with your credentials baked in.

1. Go to the Prompt Generator tab above.
2. Select your target LLM product (Codex, Cursor, Claude, ChatGPT, or generic).
3. Your API URL, token, and project ID are auto-filled from your session.
4. Copy the generated prompt into your LLM's highest-priority instruction area.
5. Start a new session and verify the agent loaded it.

Placement by product

Product	Where to paste
Codex	AGENTS.md or project instructions
Cursor	.cursor/rules/agentic-manager.mdc
Claude	CLAUDE.md or project instructions
ChatGPT	Custom GPT instructions or API system message

Option C: Fetch from API

For programmatic setups. Fetch the skills documentation and inject it into any system prompt.

Fetch the full skills guide (public, no auth required)

curl -s https://api.agenticmanager.io/api/skills

Then get your credentials

# Get your token
curl -s -X POST https://api.agenticmanager.io/api/auth/login \
  -H "Content-Type: application/json" \
  -d '{ "username": "my-agent", "password": "my-password" }'

# Get your projects
curl -s https://api.agenticmanager.io/api/projects \
  -H "Authorization: Bearer <token>"

Combine the skills markdown + your token + project ID into your LLM's system prompt. The skills endpoint returns comprehensive markdown covering the full workflow, all endpoints, and best practices.

Option D: Manual Setup

Copy the skill file contents directly into your LLM's instruction area.

1. Open skills/agentic-manager/SKILL.md from the repository.
2. Copy its full contents into your LLM's system prompt or instruction area.
3. Replace all <token> placeholders with your JWT token.
4. Replace all <id> placeholders with your project UUID.
5. For full endpoint examples, also include references/api-guide.md.
6. Include references/endgoals-playbook.md so the agent follows endgoal verification and retry loops.

Verify the Skill is Active

After any setup method, ask your LLM:

Print AGENTIC_MANAGER_PROMPT_ACTIVE and list the first three API calls you will make.

The LLM should confirm the prompt is loaded and list its startup calls: rules, context notes, and learnings.