Skip to main content

Skills Overview

A skill is a reusable system prompt preset that you can apply to any agent. Skills encode domain expertise, writing styles, analysis frameworks, or behavioral patterns that you want multiple agents to share. They are LLM-agnostic — any agent on any provider or model can use any skill.

Think of skills as plug-in modules for agents: the "Security Analyst" skill makes any agent think and respond like a security analyst. The "SEO Writer" skill makes any agent produce SEO-optimized content.


What a Skill Contains

FieldDescription
nameDisplay name
descriptionWhat this skill makes the agent do
categoryCategory for browsing
system_promptText merged into the agent's system prompt — the only field used at runtime
example_promptsSample user messages that work well with this skill
tagsTopic tags for browsing
versionSkill version
is_builtinTrue for platform-provided skills
Prompt-only, at runtime

Only system_prompt is applied when an agent runs. A skill never changes the agent's model or temperature.


Skills vs. Agent System Prompts

SkillAgent System Prompt
ScopeReusable across many agentsSpecific to one agent
OwnershipShared with the projectBelongs to the agent
ContentBehavioral framework, style, methodologyTask-specific instructions
MergingAppended to agent's own promptIs the base prompt
Update effectUpdate skill → all agents using it benefitUpdate agent → only that agent changes

Use skills for: Cross-cutting concerns that apply to many agents (writing style, analysis framework, output format standards)

Use agent prompts for: The specific task this particular agent does (write SEO blog posts about SaaS, analyze security logs from Splunk)


The Three-Layer Prompt Stack

When an agent with a skill runs, the LLM receives a combined system prompt:

Layer 1: Agent system prompt
"You are a technical content writer specializing in developer tools..."

Layer 2: Skill system prompt addition
[SEO Content Writer skill]
"Optimize content for search: include the primary keyword in the title
and first paragraph. Use H2 headings containing keyword variants..."

Layer 3: RAG retrieved context (if configured)
"--- Retrieved Context ---
[Relevant documentation chunks]
--- End Retrieved Context ---"

The model sees all three layers as one combined system message.


Skill Categories

TARX ships ~48 built-in skills across 18 categories. A representative sample:

CategoryExample skills
Content / MarketingSEO Writer, Technical Docs Writer, Email Copywriter, Social Media Writer
Research / DataResearch Analyst, Data Analyst, Market Research
Developer / DevOpsCode Reviewer, Python Developer, DevOps Advisor
CybersecuritySecurity Analyst, Threat Triage
Customer Success / SalesCustomer Support Agent, Sales Outreach
Finance / Legal / HRFinancial Analyst, Contract Reviewer, Recruiting Assistant

Other categories include design, operations, project management, product, education, and social. See Built-in Skills for the full catalog.


Next Steps