Kms — AI skill for Claude Code

Vivantel KMS

*Give your AI coding agent durable, checkable knowledge — decisions, facts, and guardrails it won't forget, contradict, or re-litigate.*

[](LICENSE) [](https://code.claude.com/docs/en/plugins.md)

**The problem isn't that your agent forgets — it's that it has no language for knowledge.** A vector database can retrieve a similar-looking snippet, but it can't tell a settled decision from a stale guess, or a hard constraint from a passing note, so it keeps re-suggesting what you already rejected. `kms` gives agents that vocabulary: four typed artifacts — facts, decisions, guardrails, procedures — each with its own fields, lifecycle, and role, so knowledge stays durable, checkable, and shareable instead of a half-remembered guess. [Read the full case for it →](https://medium.com/@strebulaev/taxonomy-is-a-language-why-your-ai-agent-needs-a-vocabulary-not-more-memory-000b8023a40a)

Vivantel KMS (Knowledge Management System) — for any git-based project, capture decisions, facts, guardrails, and procedures as durable, traceable artifacts, maintained automatically as the project evolves. Works for software repos, technical documentation, even long-form writing like research papers, not just code.

The skill content itself is agent-neutral — no product-specific language or tooling assumptions — so any AI coding agent that can read and follow instructions from a file can use it. `kms` additionally ships one-command install packaging for three: a Claude Code plugin marketplace, a native Codex plugin, and a self-hosted remote-skills manifest for Kilo Code CLI.

Skill	What it does	Examples
quickstart	Sets up the knowledge system and captures one real decision live, in the same sitting — for a first-time user.	examples
brainstorm	Genera