vector-md-context-injector
Vector-md-plugin provides hooks to automatically inject sub-tree scoped context files upon reading a file from said sub-tree. The injections
Open source Open in the app JSON README (API)
About
Vector-md-plugin provides hooks to automatically inject sub-tree scoped context files upon reading a file from said sub-tree. The injections are deduplicated per agent context. The skill file and CLAUDE.md loading hook provide guidance towards maintaining these files as a natural part of the work Claude Code does. The idea is that these vector.md files form a knowledge graph that's automatically available to claude as needed.
Details
- Kind
- Plugins
- Topic
- AI, RAG & memory
- Publisher
- lleukkun
- Origin
- marketplace
- Category
- ferramentas
- Last push
- 2026-03-24T12:26:03Z
- Repository state
- ativo
- Language
- Shell
- License
- MIT
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
lleukkun/vector-md-plugin/vector-md-context-injector
README
# vector-md — Claude Code Plugin Auto-inject the nearest `vector.md` file into Claude's context whenever it reads a file, so design constraints and conventions are always visible. ## What it does Many projects use `vector.md` files throughout the directory tree to document design intent, architectural constraints, and review rules for each subsystem. This plugin ensures Claude always sees the relevant `vector.md` when working in that subtree — without duplicate injections. - **PostToolUse (Read)** — after every file read, walks up the directory tree to find the nearest `vector.md` and injects it into context - **PostCompact** — clears dedup state after context compaction so constraints are re-injected - **Agent-aware dedup** — parent and subagents maintain separate seen-sets, since their contexts are isolated - **Workspace-bounded** — the search stops at the workspace root and never traverses above it ## Install ``` /plugin install lleukkun/vector-md-plugin ``` ## What goes in a vector.md Each `vector.md` serves as the index of its directory — capturing *why* the code is the way it is, not implementation details. ```markdown # path/to/directory Brief description of this directory's purpose. ## Files - `foo.ts` — what this file is responsible for ## Design - Constraints and patterns that must hold - Key architectural decisions and rationale ## Review - What a reviewer should verify when changes touch this directory ## See Also - [path/to/related/vector.md] — relationship ``` Sections are optional — omit any that would be empty. The plugin includes a `/vector-md:vector-author` skill that guides Claude through creating and updating these files. Invoke it with: ``` /vector-md:vector-author ``` ## Requirements - [Claude Code](https://claude.com/claude-code) with plugin support - `jq` (pre-installed on most dev machines)