daedalus Meta Skill Creator
A compact system for transforming intent into reusable, high-quality skills.
Core Loop
Extract Intent
- What does the user want to consistently achieve?
- When should this trigger?
- What does success look like?
Model the Workflow
- Identify steps, decisions, and outputs
- Capture implicit knowledge (shortcuts, corrections, preferences)
- Define input/output shape
Draft the Skill
- Clear purpose + triggering conditions
- Structured instructions (progressive, not rigid)
- Emphasize why, not just what
Test with Reality
- Create 2–3 realistic prompts
- Simulate execution
- Observe gaps, inefficiencies, ambiguity
Refine
- Generalize from failures (avoid overfitting)
- Remove unnecessary constraints
- Improve clarity, flow, and adaptability
Repeat Until Stable
- Iterate until outputs are consistent and high-quality
- Expand coverage only after core reliability
Skill Design Principles
1. Intent > Instructions
Focus on what outcome matters, not just steps. A good skill adapts.
2. Explain the Why
Models perform better when they understand purpose, not just rules.
3. Minimize Friction
Remove redundant steps, verbosity, or over-constraint.
4. Generalize Early
Avoid tailoring to specific examples—design for reuse across contexts.
5. Trigger Aggressively
Descriptions should actively pull the skill into use when relevant.
Output Structure
When creating a skill, always produce:
---
name: <skill-name>
description: <what it does + when to use it (very explicit)>
---
# Skill Name
## Purpose
Clear explanation of capability and value.
## When to Use
Concrete triggers, phrased like real user intent.
## Workflow
Step-by-step process with reasoning.
## Output Format
Exact structure or expectations.
## Examples (optional)
Realistic input/output pairs.