Skip to main content

What is the Agent Skill File?

Aegis is built for agents that can decide when to request financial actions, but those agents still need strict operating instructions. The Agent Skill File (SKILL.md) is the live execution contract your agent can load before using Aegis. It teaches the agent how to authenticate, format requests, choose the correct endpoint, generate idempotency keys, manage action nonces, respect policy limits, and recover from common API errors. Instead of maintaining a custom prompt for Aegis, fetch SKILL.md and inject it into your agent’s system prompt or knowledge base.

Where to find it

The skill file is hosted directly on the Aegis API server. You can fetch it dynamically:

What it Contains

The SKILL.md file is structured for agent consumption and includes:
  1. Authentication & Headers: Strict rules on how to format the dual-header auth, generate UUID v4 idempotency keys, and pass action nonces.
  2. Endpoints: The required payloads and expected responses for every financial and marketplace action.
  3. Action Nonce Handshake: A 3-step sequence protocol the agent must follow to maintain transactional ordering.
  4. Decision Framework: Cost-benefit rules for bridging, payments, swaps, and yield actions.
  5. Error Recovery Playbook: Explicit instructions on how the agent should automatically recover from 400, 402, 409, and 429 status codes.
  6. Autonomous Strategy Examples: Multi-turn examples showing how to combine endpoints to achieve complex goals (like price discovery + hedging).

How to use it

Option 1: Direct System Prompt Injection

When initializing your AI agent, fetch the skill file and append it to the agent’s core system prompt:

Option 2: RAG / Knowledge Base

If you are using an agent framework like Eliza, LangChain, or LlamaIndex, you can save SKILL.md to your agent’s vector database or documents directory. The agent will retrieve relevant sections automatically when it decides to execute a financial action.
Treat SKILL.md as live product configuration. Refresh it when you deploy new actions or change policy, nonce, or recovery requirements.