Skills
A skill is a reusable system prompt preset. Skills let you package battle-tested instructions and apply them to multiple agents without copy-pasting prompt text.
Think of a skill as a job-specific training module: the "Senior Code Reviewer" skill knows exactly how to prompt an LLM to behave like an expert code reviewer. You assign the skill to any agent, and that agent inherits those behaviors — merged with whatever the agent's own system prompt says. Skills are LLM-agnostic: any agent on any provider or model can use any skill.
Skills vs. System Prompts
| Skill | Agent System Prompt | |
|---|---|---|
| Scope | Reusable across agents | Specific to one agent |
| Content | Generic expertise/behavior bundle | Task-specific instructions |
| Maintenance | Update skill → all agents using it update | Must edit each agent individually |
| Flexibility | Merged with agent's own prompt | Full control |
When to use a skill:
- You want multiple agents to share a common behavioral style (e.g., "always cite sources")
- You have a well-tested prompt for a specific domain (e.g., security analysis, content writing)
- You want to apply platform-provided best practices without writing prompts yourself
When to skip the skill:
- Your agent has highly specific, unique behavior that doesn't generalize
- You need full control over the exact system prompt text
Skill Schema
A skill document contains:
| Field | Type | Description |
|---|---|---|
name | string | Display name of the skill |
description | string | What this skill makes the agent do |
category | string | One of the skill categories (for browsing) |
system_prompt | string | The prompt text merged into the agent's system prompt — the only field used at runtime |
example_prompts | list | Sample user messages that work well with this skill |
tags | list | Topic tags for browsing |
version | string | Skill version |
is_builtin | bool | True for TARX-provided skills |
At runtime, only system_prompt is applied to the agent. Skills do not change the agent's model or temperature — those are always controlled by the agent itself. The other fields exist purely to help you find, understand, and organize skills.
How Skills Merge with Agent Config
When an agent with a skill runs, the skill's system_prompt is appended to the agent's own system prompt (separated by a blank line). Nothing else changes — the agent keeps its own model, temperature, and max tokens.
Merge rule: Skill's system_prompt is appended to the agent's system prompt. The agent's model and temperature are never overridden by a skill.
Built-in Skills
TARX ships with a catalog of roughly 48 built-in skills across 18 categories — covering writing, analysis, engineering, DevOps, security, data, design, finance, legal, HR, sales, marketing, customer success, product, project management, research, social, and education. Each one is a curated system_prompt you can assign to any agent.
See the full list with descriptions in Built-in Skills.
Creating Custom Skills
Project admins and editors can create custom skills:
- Click Skills in the left sidebar.
- Click New Skill.
- Fill in the Skill Editor:
Required Fields
Name: A clear, role-describing name. E.g., "Company Voice Writer", "Internal Security Reviewer".
System Prompt: The text to be merged into any agent that uses this skill. Write this as if it's adding to a system prompt — don't start with "You are a..." (the agent's own prompt handles that). Instead, add behavioral guidance:
When writing content:
- Always use the Oxford comma
- Avoid passive voice (target less than 15% passive sentences)
- Reference case studies with specific numbers, not vague claims
- End all content pieces with "Ready to get started? [CTA]"
- Internal product names: TARX (not Tarx or tarx)
Tags: Add category tags for browsing. Use existing tags where possible: writing, analysis, code, security, customer-success, data, research.
Optional Fields
Description: One sentence explaining what this skill makes agents do.
Category & Tags: Pick a category and add tags so the skill is easy to find in the picker.
Example Prompts: Add 2-3 sample user messages that demonstrate the skill working well. These help teammates understand how to use the skill.
Assigning Skills to Agents
In the Agent Editor, Section 5 (Skills):
- Click Browse Skills.
- Filter by category tag (Writing, Analysis, Code, etc.).
- Click a skill to see its description, example prompts, and system prompt preview.
- Click Assign to attach it to the current agent.
Only one skill can be assigned per agent. If you need behavior from multiple skills, create a new custom skill that combines the prompt elements you need.
To remove a skill: click the assigned skill card → Remove.
Skill Versioning
Skills carry a version field for tracking. Built-in skills are maintained by TARX and kept in sync automatically — the catalog you see always matches the current built-in set. Agents reference a skill by id, so when a built-in skill's system_prompt improves, agents using it pick up the change on their next run.
For custom skills, maintain the version field yourself if you want to track your own iterations.
Tips
If your team has a brand voice, a security review checklist, or a data reporting format that should be consistent across all agents, encode it in a skill. Update the skill once; all agents using it get the update on their next run.
The skill's system prompt is merged with the agent's own system prompt, not a replacement. If there's a conflict between them (e.g., the agent says "write in formal tone" and the skill says "conversational tone"), the model may struggle. Make sure skills and agent prompts are complementary.
Skill system prompts are readable by all project members. Never put API keys, passwords, or sensitive information in a skill's system prompt. Use credentials or RAG sources for sensitive configuration.