Skip to content

classification-system.md

Latest commit

 

History

History
98 lines (74 loc) · 2.7 KB

classification-system.md

File metadata and controls

98 lines (74 loc) · 2.7 KB

Classification System

This document defines the canonical taxonomy for repository entries.

Design goals

  • keep categories stable over time
  • separate what a project is from how it is deployed
  • support both quick browsing and structured querying
  • keep contributor decisions consistent

Canonical categories

Every entry must use exactly one category from this list:

  1. Agent Frameworks
  2. Agent SDKs / Libraries
  3. Skills Systems
  4. Memory / Retrieval / RAG
  5. Tool Use / Integrations
  6. Multi-Agent Systems
  7. Coding Agents
  8. Research Agents
  9. Browser / Computer Use Agents
  10. Voice / Multimodal Agents
  11. Evaluation / Benchmarking
  12. Workflow / Automation
  13. Observability / Monitoring
  14. Sandboxing / Execution
  15. Knowledge Bases

Subcategory guidance

Use subcategory for narrower placement. Examples:

  • Agent Frameworks -> graph orchestrators, role-based orchestrators
  • Agent SDKs / Libraries -> Python SDK, TypeScript SDK
  • Tool Use / Integrations -> MCP, plugin runtime, tool routers
  • Evaluation / Benchmarking -> online evals, offline eval harnesses
  • Knowledge Bases -> enterprise knowledge hub, semantic indexers

Keep subcategories short (2-5 words) and reusable.

Supporting dimensions

These fields refine comparison without creating category sprawl:

  • maturity: experimental, beta, production
  • activity_status: active, maintenance, inactive
  • docs_quality: low, medium, high
  • beginner_friendly: true or false
  • deployment_type: one or more of local, cloud, hybrid, api, self-hosted, managed

Tagging rules

  • use lowercase kebab-case tags
  • prefer specific tags over broad tags (browser-automation over automation)
  • use 3-10 tags per entry
  • avoid duplicating category names as tags

Language rules

  • choose the dominant implementation language
  • use multi when no single language dominates
  • use lowercase names (python, typescript, go, rust, multi)

Beginner-friendly criteria

Mark beginner_friendly: true only when all are true:

  • setup instructions are clear
  • docs include runnable example
  • core concepts are explained
  • common errors are documented

Review cadence

  • every entry should be reviewed periodically
  • update last_reviewed in YYYY-MM-DD
  • archive or downgrade stale entries instead of deleting immediately

Folder mapping

Data is stored in practical browsing folders under data/repositories/. Those folders help navigation and do not replace canonical category values.

Current allowed data folders:

  • agent-frameworks
  • agent-sdks-libraries
  • skills-systems
  • coding-agents
  • browser-agents
  • research-agents
  • multimodal
  • multi-agent
  • evaluation
  • memory
  • tooling
  • automation