Agent Skills

Agent Skills are machine-readable instruction sets that help AI agents, assistants, and RAG pipelines interact safely and effectively with Dzaleka Online Services.

While the Public API provides the data, Agent Skills provide the “rules of engagement”—telling an agent when to use the API, which pages to prioritize in a crisis, and how to avoid common pitfalls like hallucinating contact information.

Discovery

Agents should start by loading the skills index to discover available instructions.

Index Location: https://services.dzaleka.com/.well-known/agent-skills/index.json

The index follows the Agent Skills 0.2.0 schema. Each entry provides a name, a description, and a URL to the specific SKILL.md file.

The Skill Format (SKILL.md)

Each skill is published as a markdown file with YAML frontmatter. This format is designed to be easily parsed by both humans and LLMs.

A typical SKILL.md includes:

• Name & Description: Defining the skill’s purpose.
• Compatibility: The environment needed (e.g., “Standard JSON fetch”).
• When/Do Not Use: Guardrails for the agent’s routing logic.
• Working Sequence: Step-by-step instructions for the task.
• Route/API Map: Key URLs or endpoints relevant to the skill.

Core Skills

Dzaleka Online Services currently publishes three core skills:

1. Help and Safety

Name: dzaleka-help-and-safety
Purpose: Routing urgent protection, rights, and newcomer support requests.

Use this skill when a user is in crisis or needs immediate guidance. It prioritizes the Get Help Now and Rights Navigator routes above all other background information.

2. Public API

Name: dzaleka-public-api
Purpose: Guidance for developers and agents on using the JSON API.

Use this skill to prevent agents from scraping HTML when a structured JSON endpoint exists. It defines the “Preferred Response Style” for data tasks and points to the discovery layer (api-catalog).

3. Site Navigation

Name: dzaleka-site-navigation
Purpose: Helping agents find the right public page for a human user.

Use this skill to map user intent (e.g., “Where are the jobs?”) to the correct page route (/jobs/). It includes “Routing Rules” to ensure agents don’t bury urgent links under generic navigation menus.

Implementation Guide

Loading Skills (JavaScript)

async function loadAgentSkills() {
  const indexUrl = 'https://services.dzaleka.com/.well-known/agent-skills/index.json';
  const response = await fetch(indexUrl);
  const index = await response.json();

  const skills = await Promise.all(index.skills.map(async (skill) => {
    const res = await fetch(`https://services.dzaleka.com${skill.url}`);
    const content = await res.text();
    return { ...skill, content };
  }));

  return skills;
}

Prompt Integration

When integrating these skills into an LLM-based agent, we recommend injecting the SKILL.md content into the system prompt or using it as a few-shot example for the agent’s routing logic.

Example Routing Rule:

“Before answering a question about Dzaleka, check the dzaleka-site-navigation skill. If the user’s intent matches a listed route, prioritize that link in your response.”

Guardrails & Safety

To ensure the safety of the Dzaleka community, all agents using these skills must adhere to these guardrails:

• Trust the Live Page: If a SKILL.md file and a live page disagree, the agent should trust the live response.
• Do Not Invent Contacts: Never hallucinate phone numbers, office hours, or locations. Only use the exact strings provided in the skills or on the matching public pages.
• Prioritize the “Get Help Now” Route: If a user’s request sounds urgent (violence, health, protection), always lead with the Get Help Now link.
• No Scraping when API Exists: Prefer JSON endpoints over HTML scraping whenever the skill indicates an API path is available.

---

Next Steps:

• Read the Agent Access Guide
• Explore the API Reference