Contributing to Hermes SuperCode Skills

Adding a New Skill

1. Create a directory under skills/ with your skill name (lowercase-with-hyphens)
2. Write a SKILL.md following the structure below
3. Add trigger eval queries to tests/trigger-eval.json
4. Open a PR with a description of what the skill does and example usage

SKILL.md Structure

---
name: your-skill-name
description: |
  What the skill does and when it should trigger. Be specific about
  trigger patterns — list exact phrases users might say. Make the
  description slightly "pushy" to avoid under-triggering.
---

The body should include:

• Identity — one paragraph explaining the role
• Workflow — numbered phases the skill follows
• Output format — JSON schema for structured output
• Safety rails — red/yellow/green tier classification
• Examples — at least 2 realistic usage scenarios

Quality Bar

• Keep SKILL.md under 500 lines
• Every workflow phase must be actionable (not just descriptive)
• Safety rails must cover destructive operations
• Output format must be valid JSON

Modifying an Existing Skill

• Open an issue first to discuss the change
• Preserve the existing safety rails — relaxing them requires explicit justification
• Add test cases that cover your change
• Run the trigger eval to make sure your changes don't break triggering

Trigger Evaluation

The tests/trigger-eval.json file contains queries that should and shouldn't trigger each skill. Before submitting, verify your skill triggers correctly on the should-trigger queries and doesn't trigger on the should-not queries.

Code of Conduct

• No skills that facilitate unauthorized access to systems
• No skills that help bypass security measures
• No skills that collect or process PII without explicit user consent
• No skills that provide financial, medical, or legal advice as recommendations