ralph-vault skill
ralph-vault builds a tiered knowledge vault about repositories and a plan for a ralph loop that the agent follows
- 69 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The skill writes vault files and plans into the working folder, and a ralph loop starts an agent for each repository
- The agent reads the code of several repositories and spends tokens on documentation
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/ralph-vaultDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill writes vault files and plans into the working folder, and a ralph loop starts an agent for each repository
- The agent reads the code of several repositories and spends tokens on documentation
This is third-party code. Review the repository files before installing.
What it does
A skill and CLI from Santander AI Lab create and maintain a progressive disclosure knowledge vault, a tiered deepwiki, for one or several repositories. The layout holds an index, repositories, components, infrastructure, technologies and relations. The `scripts/gv.py` script, written on the Python standard library, owns everything deterministic: creating the structure, adding and deleting repositories, validating frontmatter, links and the token budget. The text is written by the agent or a loop following the immutable prompts in `assets/prompts/`. The skill actions are `init`, `add`, `delete`, `update` for new and stale repositories, `relations`, `dependencies` and `components` for the relation graph, `validate`, `check` and `plan`. Validation forbids source code inside the vault and requires a source backlink for every entry. The main result of `plan` is a `plan/plan.md` file with per-repository tasks that a ralph loop executes one repository per iteration, or the agent itself.
Who it is for. Teams with several repositories that build a shared code knowledge base for AI agents and run agents in loops such as ralph.
Good fit when
- Agents need a shared map of a dozen services: who depends on whom and where infrastructure lives
- The vault must be checked for stale and incomplete sections with `check`
- Shared libraries used by two or more repositories move into a separate tier
- A ralph loop needs a ready `plan/plan.md` with tasks per repository
- A vault entry must not contain source code, and `validate` checks that
Not a fit when
- You have one repository with one README: the tiered structure is overkill
- You have no model access for writing text: the vault content is written by an agent
- You need only an architecture diagram as a picture, while the vault gives text and links
Example request
Document these three repositories into the ralph-vault vault and show what is stale or emptyLimitations
The skill only lays out the structure and checks it, while an agent writes the content from fixed prompts and spends tokens. For large vaults the `ralph-loop.sh` loop is recommended, since it keeps each iteration in fresh context, and for one or two repositories the agent alone will do. The `install.sh` installer copies the `ralph-vault/` folder into the skill directories of Codex, Claude Code and Gemini. The last commit dates from October 2026. The README and prompts are in English.
How to disable. Delete the `ralph-vault/` folder from `~/.claude/skills/`, `~/.codex/skills/` or `~/.gemini/antigravity-cli/skills/`. A created vault lives in your project and is deleted separately.
FAQ
How do I install ralph-vault?
Clone the repository and run `./install.sh`: it finds the Codex, Claude Code and Gemini skill directories and copies the `ralph-vault/` folder. `--dry-run` shows the plan without writing.
What is a ralph loop?
A loop where an agent gets a fresh context again and again and performs one task from a plan. ralph-vault prepares `plan/plan.md` and the fixed prompts for such a loop.
Do I need to run the CLI by hand?
Usually not: the agent reads `SKILL.md` and calls `scripts/gv.py` itself. Manual use is described in the advanced section.
What does validate check?
Frontmatter, wikilinks and token budgets, plus the absence of source code in entries and the presence of source backlinks.
What tiers does the vault have?
An index, repositories, components, infrastructure, technologies, relations and meta. Templates and section maps live in `assets/`.
Discussion
No questions or reviews yet.
Related
SkillsLearn from materials
learn-from-materials
A skill that turns books, PDFs, slides and documents into a traceable knowledge base and an interactive learning page with sources, maps and self-check
A marketplace of five Claude Code plugins for researchers: literature search, research design, project memory and manuscript work
The /okf skill builds a Markdown knowledge graph from code and docs in the Open Knowledge Format and keeps it up to date.
A Claude Code skill for scientific literature reviews: an interview, Scopus and OpenAlex search, DOI checks and a final docx file
Research Weave turns a pile of papers into a reading queue, an evidence map and a sharper research question
A library of 280+ research skills: literature search, experiment design, analysis, reference checks and paper preparation
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| ralph-vault (this card) | no | 69 | |
| OpenWiki | no | 17,914 | a CLI for a project wiki, while ralph-vault keeps tiers, a relation graph and a plan for a ralph loop |
| Karpathy LLM Wiki | no | 2,732 | a general LLM wiki approach, while this one brings ready actions, a no-source-code check and a loop plan |
README in short
The README presents the project as a skill and CLI for a knowledge vault that feeds agent loops. It lists the actions, the skill folder layout, ways to invoke it and the split of work between the deterministic script and prompts. Then install through the installer or curl, and finally advanced CLI use.
README badge
Original SKILL.md
--- name: ralph-vault description: >- Create and maintain a project-agnostic deepwiki knowledge vault — a tiered, progressive-disclosure documentation base for one or many code repos/subdirectories, designed as the knowledge source for ralph-loop agents. Use when the user wants to initialize a vault, document or refresh repos, add or remove a repo or subdirectory from the vault, validate the vault, check what is missing or stale, or generate a ralph-loop plan to (re)build it. Triggers include create/init a vault, document this repo, update the vault, add a repo to the vault, validate the vault, vault status, knowledge base for ralph. --- # Ralph Vault Lifecycle tool for a **knowledge vault**: a tiered deepwiki that documents the intent, interfaces and structure of one or more code repositories, kept faithful and current so agents (especially ralph loops) have a reliable knowledge source. Project-agnostic — the documented repos live in a per-vault config registry, nothing is hardcoded. ## How it is split - **Deterministic work → `scripts/gv.py`** (a single stdlib CLI): scaffolding, the repo registry, status, validation, and emitting ralph plan files. Run it directly; it needs no context loaded. - **Content generation → FIXED prompts in `assets/prompts/`** driven by an agent / ralph loop: writing the actual documentation from source. Run `gv.py` from the project that owns the vault. Default vault path is `vault/`; override with `--vault <path>`. ## Action router Load the matching reference file **only** for the action at hand (progressive disclosure). Do not preload all of them. | Intent | Action | Load | |---|---|---| | Create the vault structure + config if missing | **init** | `references/init.md` | | Register a repo or subdirectory to document | **add** | `references/add.md` | | Remove a repo/subdir (optionally purge docs) | **delete** | `references/delete.md` | | Document or refresh repos (incomplete/stale) | **update** | `references/update.md` | | Generate typed edges between repos | **relations** | `references/relations.md` | | Catalogue external infra + providers (reverse index) | **dependencies** | `references/dependencies.md` | | Research public docs for one external provider before filling technology notes | **external-provider-research** | `references/external-provider-research.md` | | Promote components/libraries reused by ≥ 2 repos | **components** | `references/components.md` | | Frontmatter / wikilink / budget gate | **validate** | `references/validate.md` | | Report what is missing / incomplete / stale | **check** | `references/check.md` | | Emit `plan/plan.md` + tasks for a ralph loop | **plan** | `references/plan.md` | | Wire git/CI gates so the vault can't drift | **ci** | `references/ci.md` | Shared contracts (load when authoring or validating content, not for routine CLI calls):