| name | agent-skill-standards |
| description | Guidelines for creating and maintaining granular, high-quality agent skills, shared across the Hora family. |
Agent Skill Standards
To maintain a high-quality codebase and a consistent AI pair-programming experience, skills must follow these standards. Applies to skills in any Hora app repo as well as here in hora-core.
Granularity & Focus
- One topic per skill: Avoid "General UI" or "Common Utils". Use specific titles like
Hero Section Stability or Currency Display Standards.
- Minimal duplication: If a rule applies to multiple components (e.g., currency formatting in charts and lists), create a single central skill and reference it.
- Actionable rules: Skills must provide concrete "MUST" and "MUST NOT" rules, not just general descriptions.
Structure
Every SKILL.md MUST follow this structure:
- YAML frontmatter:
name and description.
- Primary header: title of the skill.
- Rules/principles: the core technical or design requirements.
- Implementation examples: snippets demonstrating the "correct" way.
- Checklist: a short list of verification items to ensure compliance.
Evolution
- Context sync: after completing a major feature or resolving a subtle bug, immediately extract that knowledge into a skill or update an existing one.
- Reversion documentation: if a pattern is tried and reverted, update the relevant skill to document WHY the prior pattern was restored — don't just silently revert.
Metadata guidelines
- Use clear, lowercase-hyphenated directory names matching
name: in the frontmatter.
- Keep descriptions concise for quick indexing by the AI agent.
Hora-family scope note
A skill belongs in an app's own repo if it documents that app's specific implementation (class names, package paths, one-off bugs). It only belongs in hora-core once it's been generalized to a convention 2+ apps actually share — see docs/conventions.md and the hora-repo-map skill for that boundary.