Conventions and guardrails for AI agents working in this repository.
Personal AI knowledge base. Structured skill packs, agent definitions, and operational docs extracted from real projects. Content is reference material, not executable code.
- Entry point:
SKILL.md(uppercase) - YAML frontmatter:
name,description(required for discoverability) - References in
references/subdirectory - Reference filenames:
{category}-{topic}.md(e.g.,core-architecture.md,features-spotify.md) - Categories:
core,features,best-practices - Keep
SKILL.mdas an index under 500 lines — detailed content belongs inreferences/ - Write descriptions in third person for consistent system prompt injection
- Flat directory structure (no subdirectories)
- YAML frontmatter:
name,description,tools,model - Description: third person, explains what the agent does and when to use it
- Content: concise system prompt with responsibilities and conventions
- Keep agents generic and project-agnostic — they should work across any codebase
- YAML frontmatter:
summary,read_when - Content is actionable and concise
- Use kebab-case for file and directory names (except
SKILL.md,AGENTS.md,README.md) - Keep reference files focused on one topic
- Include concrete code examples, not abstract descriptions
- Prefer specific versions and actual config values over vague guidance
- Generic skills (biome, pnpm-monorepo) must not reference specific projects
- Project-specific skills (website, ait) document particular codebases
- Do not remove YAML frontmatter from any file
- Do not merge separate reference files into a single file
- When updating a skill, also update the reference table in
SKILL.md
- TypeScript/JavaScript: 2-space indent, single quotes, semicolons
- Shell: bash,
set -euo pipefail - Markdown: 120-char line width where practical