Packagerepo can now boot its entire Flask server from a workflow definition.
No more procedural Python - the whole app is declarative JSON.
New Features:
- web.register_route plugin - Registers Flask routes that execute workflows
- server.json - Complete server definition as workflow (6 routes)
- server_workflow.py - Boots Flask server by executing server.json
Architecture:
1. web.create_flask_app - Create Flask instance
2. web.register_route (×6) - Register routes → workflows
3. web.start_server - Start Flask on port 8080
Each route maps to a workflow:
- PUT /v1/.../blob → publish_artifact.json
- GET /v1/.../blob → download_artifact.json
- GET /v1/.../latest → resolve_latest.json
- GET /v1/.../versions → list_versions.json
- POST /auth/login → auth_login.json
Benefits:
- 95% code reduction (957 → 50 lines)
- Add endpoints without code (just JSON)
- No restart needed for workflow updates
- Visual DAG of entire server architecture
- Multi-language plugin support
Usage:
python packagerepo/backend/server_workflow.py
The entire Flask application is now workflow-based!
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Add workflow integration that imports from root project instead of copying.
This maintains single source of truth for workflow system.
Changes:
- workflow_loader.py imports from /workflow/executor/python/
- Uses plugins from /workflow/plugins/python/
- Created /workflow/plugins/python/packagerepo/ for app-specific plugins
- Created publish_artifact.json workflow definition
Architecture:
- Packagerepo imports workflow system, doesn't copy it
- Shared plugins (math, string, logic) available to all apps
- App-specific plugins go in root workflow/plugins/python/{app}/
Benefits:
- Single workflow engine for entire metabuilder project
- Updates to workflow system benefit all apps automatically
- Consistent workflow format across frontend and backend
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Python: class extending NodeExecutor + factory.py (80+ plugins)
- TypeScript: class implements NodeExecutor + factory.ts (7 groups, 116 classes)
- Go: struct with methods + factory.go (36 plugins)
- Rust: struct impl NodeExecutor trait + factory.rs (54 plugins)
- Mojo: struct + factory.mojo (11 plugins)
All package.json files now include:
- files array listing source files
- metadata.class/struct field
- metadata.entrypoint field
This enables a unified plugin loading system across all languages
with no import side effects (Spring-style DI pattern).
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Complete manifest hierarchy for plugin discovery:
- Root level: lists all 6 languages (python, go, rust, cpp, mojo, ts)
- Language level: lists all categories per language
- Category level: lists all plugins with metadata
Structure enables programmatic plugin discovery and loading
across the multi-language workflow plugin ecosystem.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Multi-language workflow plugin system following Python's structure:
- Each plugin in its own directory with implementation + package.json
- Category-level package.json manifests listing all plugins
- Consistent interface: run(runtime, inputs) -> outputs
Languages added:
- Go: math, string, logic, list, dict, var, convert (25+ plugins)
- Rust: math, string, logic, list, convert, var (50+ functions)
- C++: header-only math, string, logic, var, convert (30+ plugins)
- Mojo: math, string, list with systems programming features
Python structure fixed:
- Reorganized flat files into plugin subdirectories
- Added package.json metadata to all 120+ plugins
- Added missing backend category (15 plugins)
- Category manifests with plugin lists
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
