Skills

Skills package domain knowledge and multi-step workflows into units that Amara loads only when a task calls for them. Rather than stuffing every procedure your agent might need into the system prompt — bloating context and diluting attention — you author each capability as a self-contained skill and let the agent pull it in on demand. This keeps the base conversation lean while making deep expertise available the moment it’s relevant.

What a skill contains

A skill has a name, a description that tells Amara when the skill applies, and a body of instructions, references, or procedures the agent reads once it decides the skill is needed. The description is the most important part: it’s what the agent matches against the current task to decide whether to load the skill, so write it to clearly signal the trigger conditions.

import { query } from "@syntic/agent-sdk";
 
const response = query({
  prompt: "Cut a new release.",
  options: {
    skills: [
      {
        name: "release-process",
        description: "Use when cutting a release: version bumps, changelog, tagging, and publish steps.",
        instructions: "1. Bump the version in package.json. 2. Update CHANGELOG.md from merged PRs. 3. Tag and push. 4. Publish to the registry.",
      },
    ],
  },
});

from syntic_agent_sdk import query
 
response = query(
    prompt="Cut a new release.",
    options={
        "skills": [
            {
                "name": "release-process",
                "description": "Use when cutting a release: version bumps, changelog, tagging, and publish steps.",
                "instructions": "1. Bump the version. 2. Update CHANGELOG.md. 3. Tag and push. 4. Publish to the registry.",
            }
        ]
    },
)

On-demand loading

The value of skills is progressive disclosure. Amara sees only the lightweight list of skill names and descriptions until a task matches one; then it loads that skill’s full body into context. This means you can register dozens of specialized procedures without paying the token cost for all of them on every turn, and without confusing the agent with instructions that don’t apply to the task at hand.

Authoring effective skills

Keep each skill focused on one coherent capability, and make the trigger description precise so the agent loads it at the right time and not otherwise. Skills can reference files, link to other skills, and encode branching procedures. When a set of skills belongs together — say, everything for a particular framework or internal system — bundle them into a plugin so other projects can install the whole collection at once.