- Identify skill type (process-based or integration)
- Draft description explaining WHEN to use (not what it does)
- Structure content appropriately for skill type
- Challenge each line - is it truly needed?
- Verify under 150 lines (target <100)
Core Principles
Treat Context as a Public Good
Context window is shared with system prompts, conversation history, and other skills. Keep skills under 150 lines, ideally under 100. Challenge each piece of information: do I truly need this explanation?
Descriptions Explain When, Not What
The description field clarifies activation conditions, not summaries.
Skill Types
Process-Based Skills
Multi-step workflows need structure:
- Use
<required>blocks with explicit step-by-step instructions - Integrate with TaskCreate/TaskUpdate for complex workflows
- Add
<system-reminder>tags for critical decision points
- Write failing test (RED phase)
- Verify test fails correctly If multiple tests needed, write all before GREEN phase.
- Write minimal code to pass (GREEN phase)
</good-example>
## Integration Skills
CLI tools and scripts should do the work:
- Keep skill thin - delegate to external tools
- Rely on tool error messages to guide behavior
- Don't over-explain what the tool already handles
<bad-example>
Lengthy explanations of all CLI flags, error cases, and edge conditions
</bad-example>
<good-example>
Brief instruction to run CLI tool, let its help/errors guide the agent
</good-example>
# Formatting Conventions
- Use fake XML tags: `<good-example>`, `<bad-example>`, `<required>`, `<system-reminder>`
- Write first person: "Ask me to do ABC" (not "Ask the user")
Converted and distributed by TomeVault — claim your Tome and manage your conversions.