Your First Skill
Skills are prompt-based agents that shape how PIE responds. They come in two modes:
- Always-on skills are injected into every conversation (personality, formatting, tone)
- On-demand skills (prompt tools) are invoked by the AI only when relevant (domain expertise, analysis)
What Skills Do
- Add personality traits (always-on)
- Set response formatting rules (always-on)
- Provide domain expertise on demand (prompt tool)
- Give specialized analysis capabilities (prompt tool)
Always-On Skills
Always-on skills modify how PIE responds. They're lightweight prompts included in every conversation.
Create an Always-On Skill
- Go to Agents in PIE
- Click Create Agent
- Select Skill tier
- Fill in the details:
Name: friendly-assistant
Display Name: Friendly Assistant
Description: Makes PIE respond in a warm, friendly manner
Prompt (the body of your skill):
You are a friendly and warm assistant. Always:
- Start responses with a friendly greeting when appropriate
- Use conversational language
- Show enthusiasm when helping
- Offer encouragement and positive feedback
- End with a helpful follow-up question when relevant
Keep your personality consistent but natural - don't be over-the-top.Manifest:
{
"trigger": "always"
}How Always-On Skills Work
- The skill's prompt is added to PIE's system prompt
- It's included in every conversation
- Multiple skills combine together
On-Demand Skills (Prompt Tools)
On-demand skills provide domain expertise that the AI invokes only when relevant. Instead of burning tokens on every message, the knowledge is delivered as a tool result when the AI decides it's needed.
This is ideal for:
- Legal document analysis (NDAs, contracts)
- Code review checklists
- Compliance frameworks
- Industry-specific expertise
- Any knowledge that's deep but not always relevant
How Prompt Tools Work
- User asks: "Can you review this NDA for me?"
- AI sees the
analyze_ndatool in its available tools - AI calls the tool with the document text
- PIE returns the skill's prompt (expert knowledge) as the tool result
- AI uses that expertise to formulate its analysis
No code runs. No sandbox. The prompt template IS the tool result.
Create an On-Demand Skill
The .pie file format is the same as always-on skills, but the manifest includes a tool: definition:
---
name: nda-analyzer
displayName: NDA Analyzer
description: Expert NDA analysis with risk assessment and recommendations
tier: skill
version: 1.0.0
manifest:
trigger: auto
tool:
name: analyze_nda
description: >-
Analyze a Non-Disclosure Agreement for risks, unusual clauses, red flags,
and missing protections. Call when a user asks to review an NDA.
parameters:
type: object
properties:
document_text:
type: string
description: The full text of the NDA to analyze
focus_areas:
type: string
description: "Optional: specific concerns like 'non-compete scope'"
required:
- document_text
---
You are an expert contract attorney specializing in NDAs. When analyzing:
## Structure your analysis as:
1. **Agreement Type**: Mutual vs one-way, identify the parties
2. **Key Terms**: Duration, scope, exclusions
3. **Risk Assessment**: Rate overall risk as Low / Medium / High
4. **Red Flags**: Non-standard or unusually broad clauses
5. **Missing Protections**: Standard clauses that are absent
6. **Recommendations**: Specific suggested modifications
## Pay special attention to:
- Overly broad definitions of "confidential information"
- Non-compete or non-solicitation clauses hidden in NDAs
- Asymmetric obligations
- Jurisdiction and governing law implicationsThe key differences from an always-on skill:
| Always-On Skill | On-Demand Skill (Prompt Tool) | |
|---|---|---|
| Manifest | { "trigger": "always" } | Has a tool: definition with name, description, parameters |
| When active | Every conversation | Only when the AI decides to call it |
| Token cost | Paid on every message | Only when invoked |
| Best for | Personality, tone, formatting | Domain expertise, analysis, knowledge bases |
| Body content | Short behavioral prompt | Can be very large (knowledge bases, frameworks) |
When to Use Which
Use always-on when the skill is:
- A personality or tone modifier
- A formatting preference
- Short (under ~500 tokens)
- Relevant to virtually every conversation
Use on-demand when the skill is:
- Domain-specific expertise
- A large knowledge base
- Only relevant to certain topics
- Something with clear input parameters (a document to analyze, a topic to research)
Tips for Good Skills
- Be specific - Clear instructions work better than vague ones
- Don't conflict - Avoid always-on skills that contradict each other
- Keep always-on skills lightweight - Save large knowledge bases for on-demand
- Write good tool descriptions - For on-demand skills, the description tells the AI when to invoke it
- Define clear parameters - Help the AI understand what inputs your skill needs
Next Steps
Now that you've created a skill, let's build something more powerful: Your First Tool.