majiayu000 enterprise_bot Skill
Purpose
Lazy-load access to the enterprise_bot codebase. Load this skill instead of reading 60 files.
Quick Start
To use enterprise_bot in new code:
from protocols import cfg, get_auth_service, CogTwin, MemoryNode
# That's it. 12 exports. Everything else is internal.
Nuclear Elements (protocols.py)
These 12 exports are the ONLY stable API. Everything else is implementation detail.
| Export | Module | Purpose |
|---|---|---|
cfg(key, default) |
config_loader | Get any config value |
load_config(path) |
config_loader | Load YAML config |
get_auth_service() |
auth_service | Singleton for auth |
authenticate_user(email) |
auth_service | SSO → database user |
User |
auth_service | Auth user dataclass |
get_tenant_service() |
tenant_service | Singleton for tenant/dept |
TenantContext |
enterprise_tenant | Request context carrier |
CogTwin |
cog_twin | The brain (query pipeline) |
DualRetriever |
retrieval | Memory retrieval system |
create_adapter(provider) |
model_adapter | LLM factory |
MemoryNode |
schemas | Memory chunk dataclass |
EpisodicMemory |
schemas | Episode dataclass |
Rules
- New code imports from
protocolsonly - never import internal modules directly - Modifying internals? - Extract from src.zip, edit, test, re-zip
- Adding new nuclear elements? - Update protocols.py + this table
- Implementation changes - Fine, as long as protocol signatures stay stable
File Tree
enterprise_bot/
├── protocols.py # ← START HERE (12 exports)
│
├── ══ CONFIGURATION ══
├── config.yaml # App config (tier, features, model)
├── config_loader.py # cfg() helper
├── config.py # Settings class
│
├── ══ ENTRY POINTS ══
├── main.py # FastAPI app + WebSocket
├── claude_chat.py # SDK agent REPL
├── claude_run.py # One-shot executor
│
├── ══ AUTH ══
├── auth_service.py # User CRUD, permissions
├── azure_auth.py # Azure AD SSO
├── sso_routes.py # OAuth callbacks
│
├── ══ TENANT ══
├── tenant_service.py # Department content
├── enterprise_tenant.py # TenantContext dataclass
│
├── ══ COGNITIVE ══
├── cog_twin.py # Main brain
├── retrieval.py # DualRetriever
├── model_adapter.py # LLM factory
├── venom_voice.py # Personality injection
│
├── ══ MEMORY ══
├── memory_backend.py # Abstract + FileBackend
├── postgres_backend.py # PostgreSQL backend
├── schemas.py # MemoryNode, EpisodicMemory
│
├── ══ SEARCH (internal) ══
├── hybrid_search.py # Vector + keyword
├── scoring.py # Relevance scoring
├── embedder.py # Embedding generation
│
├── ══ DATA ══
├── data/memories/ # File-based memory storage
├── Manuals/Driscoll/ # Department manuals + chunks
│
├── ══ DB ══
├── db/migrations/ # PostgreSQL migrations
│
└── frontend/ # SvelteKit app
Token Budget
| Load | Tokens |
|---|---|
| This SKILL.md | ~600 |
| protocols.py | ~150 |
| Extract 1 file | 200-800 |
| Full cold start | ~800 |
vs old way: 70,000 tokens to load everything
Common Tasks
Add a new API endpoint:
unzip -p src.zip main.py > main.py
# Edit main.py
# Test
zip -u src.zip main.py
Modify retrieval logic:
unzip -p src.zip retrieval.py > retrieval.py
# Edit - but DON'T change DualRetriever's public interface
# Test
zip -u src.zip retrieval.py
Check what's in a module:
unzip -p src.zip cog_twin.py | grep "def \|class "
Version
- Skill: 1.0.0
- Codebase: enterprise_bot @ commit
b601f3b - Last updated: 2024-12-19