Sovi Capability Foundry founding specification

# Sovi Capability Foundry — founding specification (preserved original)

Operator: Nathaniel
Date: 2026-09-14

Decision: The durable asset is NOT the Grok prompt. The durable asset is the structured capability specification.

Pipeline:
CONVERSATION / CORPUS
→ KNOWLEDGE EXTRACTION
→ INTELLECTUAL ASSET DISCOVERY
→ CAPABILITY DISCOVERY
→ CAPABILITY ARCHITECTURE
→ SOFTWARE SPECIFICATION
→ BUILD PACKAGE
→ AI CODING SYSTEM
→ SOFTWARE
→ SOVI INTEGRATION
→ CAPABILITY REGISTRATION
→ EXECUTABLE CAPABILITY

Requirements that MUST hold:
- Preserve original source material. Never overwrite originals with AI interpretations.
- Human approval gates for capability approval, merges, spec approval, build, deploy, register, permissions, destructive changes.
- Do not fake unavailable integrations. Create adapter contracts instead.
- Overlap detection must present recommendations; never auto-merge.
- Multi-pass analysis with resumable jobs.
- Self-hostable. No unnecessary telemetry.
- Builder-agnostic specs. Grok Build, Codex, Claude Code, humans, and Sovi are interchangeable adapters.
- Every build package must require README.md, SOVI-INTEGRATION.md, SOVI-MANIFEST.yaml, and the docs/* set.
- Provenance chain: software feature → requirement → intellectual asset → conversation → source.

Integrate later:
- Sovi launcher PWA (contract only)
- ChatGPT / Claude / Grok export adapters
- Google Drive, Git, local filesystem
- Vector retrieval
- Autonomous mode (architected, disabled)

Sources:
https://docs.x.ai
https://github.com

Open question: How should Sovi score potential value across capabilities without a production usage graph?