** Santander AI Lab의 오픈 소스. ** Python **CLI 도구 / 라이브러리 ** 생성 및 유지 하나의 코드 저장소에 대한 진보적 인 결함 (Stevened “deepwiki”) - LLM / AI Agent 루프(ralph-style)에 대한 지식 소스.

**Santander AI Open Source ** - Banco Santander (santander.com)의 오픈 소스 AI 프로젝트.

---

기술 + CLI tocreate and maintain a Progress-disclosure knowledge vault (“deepwiki”) for one or many code repositories, ralph-style Agent loops에 대한 지식 소스. Project-agnostic: 문서화 된 저장소는 per-vault config 레지스트리에 살고, 아무것도 hardcoded.

무엇 그것

-init 계층형 vault 구조(index, repos, components], infrastructure], technologies], relations], … meta]) + config. -add /delete 문서에 저장소 또는 하위 디렉토리.

• ** 업데이트 ** 새로운, 불완전, 또는 stale (FIXED 프롬프트를 통해 LM 일). -relations / dependencies / components — graph tiers: repos, External infra + Provider (reverse index)와 ≥ 2 repos에 의해 재사용 된 공유 라이브러리 사이 유형 가장자리.

ralph-vault

Open source by Santander AI Lab. A Python CLI tool / library to create and maintain a progressive-disclosure knowledge vault (a tiered “deepwiki”) for one or many code repositories — the knowledge source for LLM / AI agent loops (ralph-style).

Part of Santander AI Open Source — open source AI projects from Banco Santander (santander.com).

---

A skill + CLI to create and maintain a progressive-disclosure knowledge vault (a tiered “deepwiki”) for one or many code repositories, designed to be the knowledge source for ralph-style agent loops. Project-agnostic: documented repos live in a per-vault config registry, nothing is hardcoded.

What it does

• init the tiered vault structure (index, repos, components, infrastructure, technologies, relations, … meta) + config.
• add / delete a repo or subdirectory to document.
• update repos that are new, incomplete, or stale (LLM work via FIXED prompts).
• relations / dependencies / components — graph tiers: typed edges between repos, external infra + providers (reverse index), and shared libraries reused by ≥ 2 repos.
• validate frontmatter / wikilinks / token budgets, plus a deterministic no-source-code gate and a source-backlink check.
• check what is missing / incomplete / stale.
• plan a ralph-loop (plan/plan.md + tasks) to (re)build the vault.

The split: scripts/gv.py owns everything deterministic; assets/prompts/ are the immutable prompts that an agent/loop follows to write the content.

Layout

ralph-vault/
├── SKILL.md              # entry point + action router (progressive disclosure)
├── agents/openai.yaml    # UI metadata