mancitrus/system-prompts-and-models-of-ai-tools

mirror of https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git synced 2026-06-18 15:29:36 +00:00

Sami Assiri c6efca32b5 docs(governance): expand operating fabric index and governance library

Made-with: Cursor

2026-04-16 07:58:36 +03:00

2.3 KiB

Raw Blame History

Data plane and connector facades

Canonical: MASTER_OPERATING_PROMPT.md.

Data and knowledge layer

Operational store: relational DB as system of record for business entities; strict tenant boundaries (tenant_id and access checks).
Embeddings: colocate or pipeline near operational data when semantic search or routing benefits; do not let embeddings become an ungoverned shadow DB of PII.
Graph stores: add only when relationship-heavy reasoning is a proven requirement (cost and ops complexity are real).
Document ingestion: contracts, DD materials, board packs, PDFs — through controlled pipelines with retention and sensitivity classes.
Lineage / metadata catalog: choose one primary cataloging path; a second overlapping system needs explicit justification.

Semantic metrics

Do not let multiple systems independently define “revenue”, “pipeline velocity”, “contribution margin”, “synergy realization”, or other executive metrics.

Prefer:

Central semantic definitions (single module or analytics layer).
Data quality checks and documented downstream impact.
Visible lineage from metric definition to dashboards and agent outputs.

Connector facade rule

Agents and ad-hoc scripts must not sprawl direct calls to many vendor APIs.

Each connector facade should specify:

Contract (inputs/outputs, error model)
Version
Retry policy
Timeout policy
Idempotency key handling
Approval policy linkage (Class B where needed)
Audit field mapping
Observability hooks (logs, metrics, traces)
Rollback or compensation notes when side effects are possible

Treat every external API as a changing contract, not a fixed truth.

Dealix pointers

Model / provider routing: salesflow-saas/backend/app/services/model_router.py.
LLM provider abstraction: salesflow-saas/backend/app/ai/llm_provider.py, salesflow-saas/backend/app/services/local_inference.py.
Integrations and CRM direction: search integrations and services for connector-style modules; extend via facades rather than new raw SDK calls from agent nodes.

2.3 KiB Raw Blame History