claude-plugins

A personal Claude Code plugin marketplace by @nicholas1513.

Plugins

Plugin	Version	Description
rag	0.1.2	Two-layer RAG memory system — System Knowledge (durable, versioned) and Issue Memory (active cards). Skills: /rag:init, /rag:card, /rag:trace, /rag:promote, /rag:context, /rag:memory.
decision-engine	0.1.3	Personal goal-indexed decision engine. Treats goals as first-class indexed objects, evaluates new inputs against them via reusable workflows (e.g. house_analyzer), and persists runs for longitudinal analysis. Ships a session-start agent (@decision-engineer), a reactive skill, and an MCP server for state management.

Install

Add this marketplace to Claude Code:

claude plugin marketplace add nicholas1513/claude-plugins

Then install a plugin from it:

claude plugin install rag@nicholas1513-claude-plugins
claude plugin install decision-engine@nicholas1513-claude-plugins

Or, in a Claude Code session:

/plugin marketplace add nicholas1513/claude-plugins
/plugin install rag@nicholas1513-claude-plugins
/plugin install decision-engine@nicholas1513-claude-plugins

If the marketplace was added before a plugin was published, refresh the local cache first:

claude plugin marketplace update nicholas1513-claude-plugins

Usage — rag

After install, the /rag:* skills are available in any Claude Code session. Typical flow:

/rag:init        → stand up the rag-memory/ directory in the current project
/rag:card        → open a new investigation card
/rag:trace       → append a finding to the current card
/rag:promote     → move a confirmed, durable finding to system/ knowledge
/rag:context     → resume a card in a new session
/rag:memory      → orchestrator; routes to the right skill given current state

See plugins/rag/ for skill source and plugins/rag/features/plugin-support.md for the full reference on Claude Code's plugin host model that this plugin was built against.

Usage — decision-engine

After install, the @decision-engineer agent is registered and the decision-engine MCP server exposes goal/workflow/run tools.

First-session note: the MCP server is built from TypeScript on first launch via a SessionStart hook (hooks/ensure-build.sh). The build is idempotent — it only runs when mcp-server/dist/server.js is missing, so subsequent sessions start instantly. Requires Node.js ≥ 20 on PATH.

Typical flow:

@decision-engineer        → session-start agent: surfaces active goals, recent runs
                            and pending decisions

Goals live in plugins/decision-engine/goals/ as markdown files indexed by the MCP server. Workflows (e.g. house_analyzer) define how new inputs are evaluated against a goal. Runs are persisted under runs/ for longitudinal analysis.

See plugins/decision-engine/ARCHITECTURE.md for the full design.

Versioning

Plugins in this marketplace follow semver. Releases are git-tagged using Claude Code's required format:

{plugin-name}--v{version}

For example: rag--v0.1.2, decision-engine--v0.1.1. The double-dash separator is part of the spec — it lets each plugin in this monorepo carry an independent version line.

The marketplace pins each plugin to its tag using a git-subdir source, which sparse-clones only that plugin's subdirectory over HTTPS:

{
  "name": "rag",
  "source": {
    "source": "git-subdir",
    "url": "https://github.com/nicholas1513/claude-plugins.git",
    "path": "plugins/rag",
    "ref": "rag--v0.1.2"
  }
}

Development

Local install during development (no marketplace needed) — point Claude Code at any plugin's directory:

git clone https://github.com/nicholas1513/claude-plugins.git
claude --plugin-dir claude-plugins/plugins/rag
claude --plugin-dir claude-plugins/plugins/decision-engine

Or add the local checkout as a marketplace so installs resolve through the same flow as the published version:

claude plugin marketplace add ./claude-plugins

License

MIT © 2026 Nicholas DeTore