Skip to content

AGENTS.md

Latest commit

 

History

History
64 lines (51 loc) · 3.05 KB

AGENTS.md

File metadata and controls

64 lines (51 loc) · 3.05 KB

Metadata-Auto-Classifier

Purpose

Metadata Auto Classifier — Obsidian plugin that uses AI (OpenAI/Gemini/Ollama/etc.) to automatically classify and apply frontmatter metadata to notes. Supports multiple LLM providers via a unified adapter, configurable classification prompts, and preset-based metadata templates.

Key Files

File Description
src/main.ts Composition root — MetadataAutoClassifierPlugin, commands, settings
src/domain/constants.ts Plugin-wide constants
src/domain/prompt.ts Prompt builder — constructs LLM classification request
src/domain/presets.json Built-in metadata classification presets
src/domain/auth/ API key management and validation
src/ui/ClassificationService.ts Orchestrates LLM request → frontmatter write pipeline
src/ui/CommandService.ts Registers Obsidian commands (classify active note, batch classify)
src/ui/UnifiedProvider.ts Unified LLM provider interface (OpenAI/Gemini/Ollama/etc.)
src/ui/provider-api.ts HTTP adapter for provider REST APIs (uses requestUrl)
src/ui/frontmatter.ts Read/write note frontmatter via Obsidian MetadataCache
src/ui/request.ts Request builder and response parser
src/ui/settings/ Settings tab — provider selection, API keys, prompt templates
src/ui/index.ts UI barrel export

Subdirectories

Directory Purpose
src/ Source layers (see src/AGENTS.md)
src/domain/ Business logic — NO obsidian imports (see src/domain/AGENTS.md)
src/domain/auth/ API key storage and validation (see src/domain/auth/AGENTS.md)
src/ui/ Obsidian-dependent services, provider adapters, frontmatter, settings (see src/ui/AGENTS.md)
src/ui/settings/ Settings tab components (see src/ui/settings/AGENTS.md)
src/types/ Pure type definitions (see src/types/AGENTS.md)

For AI Agents

Working In This Directory

  • 4-layer architecture: domain/ must not import obsidian
  • UnifiedProvider.ts is the extension point for new LLM providers — add there, not in services
  • isDesktopOnly: false — use requestUrl (Obsidian API), not fetch or Node http
  • this repo keeps implementation local; do not reintroduce a default src/shared/ implementation surface
  • Test files in __tests__/ use Jest-style mocks in __mocks__/

Testing Requirements

pnpm run ci     # build + lint + test
pnpm run lint   # ESLint — 0 errors required

Common Patterns

  • Adding a new LLM provider: implement the provider interface in UnifiedProvider.ts, add auth config in domain/auth/
  • Classification flow: CommandService → ClassificationService → UnifiedProvider → frontmatter.ts

Dependencies

Internal

  • workspace contract docs define shared surfaces; this repo keeps implementation local by default

External

  • obsidian — Obsidian Plugin API
  • OpenAI / Gemini / Ollama APIs — via configurable provider