discourse/AI-AGENTS.md

AI Coding Agent Guide

Project-specific instructions for AI agents. MUST be loaded at conversation start.

Default Mode

• Architect mode enabled by default: detailed analysis, patterns, trade-offs, architectural guidance
• Stop and ask for context if unable to write code meeting guidelines

Development Rules

Discourse is large with long history. Understand context before changes.

Toolset

• Use pnpm for JavaScript, bundle for Ruby
• Use helpers in bin over bundle exec (bin/rspec, bin/rake, bin/lint)

JavaScript and UI

• No empty backing classes for template-only components unless requested
• Use FormKit for forms, see ./docs/developer-guides/docs/03-code-internals/21-form-kit.md (frontend/discourse/app/form-kit)
• Use BEM for CSS, see ./docs/developer-guides/docs/03-code-internals/25-css-guidelines-bem.md
• Make display strings translatable (use placeholders, not split strings)
• Use "Sentence case" for strings, not "Proper Case" or "lower case"
• Plugins/themes can't import npm modules directly; add the dependency to core and expose a frontend/discourse/app/lib/load-*.js wrapper that does the import() (see load-morphlex.js).

Comments & Types

• Prefer self-documenting code. Comments should only be added when future misunderstanding is likely. They should be terse, and should describe 'why', not 'what'. They should not be used to describe history.
• In the frontend, typescript is typically used for platform-level code, javascript for business-logic
• Platform-level frontend code should include accurate types & tsdoc descriptions for public APIs
• Simple JSDoc/TSDoc comments can be used in other code for editor intellisense, but this is not essential
• In core, never name plugin features or specific libraries in comments/docs — describe by mechanism

Testing

• Use the skill at .skills/discourse-writing-rspec-tests when writing RSpec tests

Commands

# JavaScript tests - bin/qunit
bin/qunit --help # detailed help
bin/qunit path/to/test-file.js  # Run all tests in file
bin/qunit path/to/tests/directory # Run all tests in directory

# Linting
bin/lint --fix path/to/file path/to/another/file
bin/lint --fix --recent # Lint all recently changed files

ALWAYS lint any changes you make with bin/lint --fix

Site Settings

• Configured in config/site_settings.yml or config/settings.yml for plugins
• Functionality in lib/site_setting_extension.rb
• Access: SiteSetting.setting_name (Ruby), siteSettings.setting_name (JS with @service siteSettings)

Services

• Extract business logic (validation, models, permissions) from controllers
• docs/developer-guides/docs/03-code-internals/19-service-objects.md
• Use the skill at .skills/discourse-service-authoring
• Examples: app/services (only classes with Service::Base)

Database & Performance

• ActiveRecord: use includes()/preload() (N+1), find_each()/in_batches() (large sets), update_all/delete_all (bulk), exists? over present?
• Queries: use explain, specify columns, strategic indexing, counter_cache for counts

Migrations

• Use the skill at .skills/discourse-migration before writing or reviewing any migration

HTTP Response Codes

• 204 No Content: Use head :no_content for successful operations that don't return data
  • DELETE operations that successfully remove a resource
  • UPDATE/PUT operations that succeed but don't need to return modified data
  • POST operations that perform an action without creating/returning resources (mark as read, clear notifications)
• 200 OK: Use render json: success_json when returning confirmation data or when clients expect a response body
• 201 Created: Use when creating resources, include location header or resource data
• Do NOT use 204 when:
  • Creating resources (use 201 with data)
  • Returning modified/useful data to the client
  • Clients expect confirmation data beyond success/failure

Security

• XSS: use {{}} (escaped) not {{{ }}}, sanitize with sanitize/cook, no innerHTML, careful with @html
• Auth: Guardian classes (lib/guardian.rb), POST/PUT/DELETE for state changes, CSRF tokens, protect_from_forgery
• Input: validate client+server, strong parameters, length limits, don't trust client-only validation
• Authorization: Guardian classes, route+action permissions, scope limiting, can_see?/can_edit? patterns. Use user.guardian shorthand not Guardian.new(user)

Knowledge Sharing

• ALWAYS persist information for ALL developers (no conversational-only memory)
• Follow project conventions, prevent knowledge silos
• Recommend storage locations by info type
• Inform when this file changes and reloads