ClaudeCodeMod

All shelves / Marketplaces

blackmo18/realm

blackmo18/realm · 0 stars · 1 plugin

Marketplace

Install

In Claude Code
/plugin marketplace add blackmo18/realm

Run inside Claude Code, then install plugins from it. Check the repo's README before you run it.

Open the repo

Plugins 1

After adding the marketplace, install one with /plugin install <name>@realm.

  1. 1realmLiving Obsidian vault for project decisions & architecture memory. Core: forge, fathom, recall, status, planning, concise./plugin install realm@realmproductivity

Files

README.md

Realm

Realm

Decision capture & knowledge pipeline for Claude Code, Cursor, Codex, and Gemini. Persists the WHY behind your code — decisions made, alternatives rejected, constraints imposed — as interlinked ADR nodes in an Obsidian vault.

Realm keeps one host-neutral Markdown skill source. Invoke skills with /realm-* in Claude Code and Gemini, or $realm-* in Codex. Host-native agent files contain only model, sandbox, and delegation adapters; workflow procedures stay in skills/.


Table of Contents


What is Realm?

Realm is an AI-coding-host skill/plugin system that captures architectural decisions as you make them — the choices, the alternatives you rejected, and the constraints those decisions impose — and stores them as compressed, interlinked ADR nodes in an Obsidian vault.

Every new AI session starts cold. Code tells AI what exists. Realm tells AI why it exists that way, what was tried and discarded, and what must not change.


Why Realm?

Code captures the surviving solution. It does not capture the graveyard of rejected alternatives, the constraint that forced an unusual pattern, or the incident that made you write "DO NOT change this order."

Without a decision record, AI assistants re-derive or re-propose settled questions every session. You answer the same "why not Redis pub/sub?" question repeatedly. A rejected approach gets proposed again in session 12.

Problem Realm's Answer
AI starts cold every session Vault persists decisions across sessions
Code doesn't explain WHY ADR nodes store rationale + rejected alternatives
Rejected approaches get re-proposed realm-recall "tried X" surfaces prior art instantly
Constraints are invisible in code Consequences field captures what must not change
Oversized god files accumulate tech debt realm-concise triages and queues refactors deterministically
Architectural changes lack structure realm-planning provides two-phase execution blueprints

How It Works

/realm-forge     ← Bootstrap vault directory structure and project state (once per repo)
/realm-fathom    ← Investigate code + vault rationale in parallel before changing code
/realm-recall    ← Instant decision queries (~20 tokens/node)
/realm-planning  ← Two-phase planning (P1: High-level + ADR direction; P2: Impl blueprint)
/realm-concise   ← God-file concierge triage queue (scripts/concise.py + committed ledger)
/realm-status    ← Read-only vault & pipeline health check

1. Bootstrap — realm-forge

Run /realm-forge once per project to link an Obsidian vault, create directory conventions, and seed pipeline state.

2. Investigate — realm-fathom

Before touching unfamiliar code:

/realm-fathom function:validateUser   # live code (ground truth) + vault WHY
/realm-fathom "how does auth flow"    # end-to-end + ADR context

Live code is always ground truth. Vault adds the architectural intent. Conflicts flagged as VAULT DRIFT.

3. Query — realm-recall

/realm-recall "why JWT"               # ~20 tokens
/realm-recall "what was rejected for auth"   # surfaces rejected_alternatives
/realm-recall "constraint on payments"       # consequences field
/realm-recall decisions               # all ADR nodes, compressed

4. Planning — realm-planning

Two-phase planning skill operating inside native plan mode:

  • Phase 1: High-level architectural analysis and ADR direction.
  • Phase 2: Code-level implementation blueprint for coding agents.
/realm-planning "refactor auth to JWT"
/realm-planning "distributed caching" --phase2

5. God-File Triage — realm-concise

Deterministic crawler (scripts/concise.py) finds oversized files, scores blast radius, maintains persistent queue in .realm/concise-state.json, and updates committed docs/GOD_FILES.md ledger.

/realm-concise                        # scan & print top god files
/realm-concise recommend <file>       # single-file refactor recommendation
/realm-concise plan <file>            # delegate approved candidate to /realm-planning

Skills

Skill Purpose
/realm-forge Bootstrap vault directory structure and local state. Run once per project.
/realm-fathom Deep investigation: live code + vault in parallel. Returns what (code) + why (vault). Flags drift. Zero writes.
/realm-recall Query vault by decision keyword, tag, or semantic phrase. Optimized for ADR queries: "why X", "rejected for Y", "constraint on Z".
/realm-planning Two-phase architecture and implementation plan. Phase 1: High-level + ADR direction; Phase 2: Code blueprint.
/realm-concise God-file triage concierge. Deterministic LOC/blast-radius scoring, persistent refactor queue, committed docs/GOD_FILES.md ledger.
/realm-status Read-only health check. Lists node counts, stale docs, pipeline state.

Team-Wide Facts (Organization)

Central GitLab realm-facts repo for organization knowledge. GitLab MR review + Microsoft Teams notifications.

One routed skill, /realm-facts, with 8 subcommands:

Subcommand Purpose
/realm-facts forge Connect product repo to central facts repo
/realm-facts new Create a new team fact
/realm-facts link Link facts (related, depends_on, supersedes)
/realm-facts submit Submit for GitLab MR review + Teams notification
/realm-facts review Reviewer approve or request changes
/realm-facts sync Pull latest approved facts
/realm-facts recall Query facts (compressed by default)
/realm-facts ingest Bundle facts for other agents

See docs/realm-facts-workflow.md for full team workflow.


Installation

See REQUIREMENTS.md for prerequisites.

Cursor (plugin)

Install from the repository URL via Settings → Plugins:

https://github.com/blackmo18/realm

Cursor reads .cursor-plugin/plugin.json and registers Realm's skills plus the Cursor-native agent adapters. Reload Cursor afterwards, then run /realm-forge.

Codex, Cursor, and Gemini

# Recommended installer
curl -fsSL https://raw.githubusercontent.com/blackmo18/realm/main/install.sh | bash -s -- --agent codex
curl -fsSL https://raw.githubusercontent.com/blackmo18/realm/main/install.sh | bash -s -- --agent cursor
curl -fsSL https://raw.githubusercontent.com/blackmo18/realm/main/install.sh | bash -s -- --agent gemini

# Skills-only direct installs (native agents require the recommended installer)
npx --yes skills add blackmo18/realm -a cursor --global --yes
npx skills add blackmo18/realm -a gemini

From a local clone:

# Global install
node bin/install.mjs --agent codex
node bin/install.mjs --agent gemini

# Local project-only install (does not touch ~)
./install.sh --agent gemini --local
./install.sh --agent codex --local

Claude Code

# 1. Install caveman plugin (required dependency)
/plugin marketplace add ~/.claude/plugins/marketplaces/caveman

# 2. Copy Realm into the Claude plugin marketplace path from a local clone
node bin/install.mjs --agent claude --force

# 3. Install realm inside Claude Code
/plugin marketplace add ~/.claude/plugins/marketplaces/realm

Full step-by-step guide: INSTALL.md


Claude Plugin Publication

To publish or update Realm on the Claude Code marketplace:

  1. Verify Plugin Manifests:

    • .claude-plugin/plugin.json — contains plugin name, version, description, and author.
    • .claude-plugin/marketplace.json — contains plugin entry, category, and source repository pointer (https://github.com/blackmo18/realm.git).
  2. Update Release Version & Git Commit SHA:

    • Bump version in .claude-plugin/plugin.json (e.g., "version": "0.1.6").
    • Get current git commit SHA: git rev-parse HEAD.
    • Update sha in .claude-plugin/marketplace.json.
  3. Commit & Push to GitHub:

    git add .claude-plugin/plugin.json .claude-plugin/marketplace.json
    git commit -m "chore(release): update plugin manifest and commit SHA"
    git push origin main
  4. Add or Refresh in Claude Code:

    # Add plugin from local marketplace path or git endpoint:
    /plugin marketplace add ~/.claude/plugins/marketplaces/realm
    # Or from public/remote git URL:
    /plugin marketplace add blackmo18/realm

Uninstallation

# Skills CLI installs (Cursor and Gemini)
npx skills remove realm

Codex native removal commands are listed in UNINSTALL.md.

For Claude Code installs:

./uninstall.sh

Full uninstall guide: UNINSTALL.md


Quick Start

Codex:

$realm-forge
$realm-fathom function:validateUser
$realm-recall "why JWT"
$realm-planning "refactor payment module"
$realm-concise

Claude Code and Gemini:

# 1. Bootstrap the vault for your project
/realm-forge

# 2. Investigate before changing code
/realm-fathom function:validateUser

# 3. Query existing ADRs
/realm-recall "why JWT"

# 4. Plan complex architectural changes
/realm-planning "refactor payment module"

# 5. Check god-file tech debt queue
/realm-concise

Querying the Vault

Decision queries (primary use case)

/realm-recall "why did we choose X"       # rationale field
/realm-recall "what was rejected for Y"   # rejected_alternatives field
/realm-recall "constraint on Z"           # consequences field
/realm-recall decisions                   # all ADR nodes, compressed

Before touching unfamiliar code

/realm-fathom function:validateUser       # signature, flow, callers + vault why + drift check
/realm-fathom class:AuthService           # class responsibility + vault ADR context
/realm-fathom "how does auth flow work"   # freeform → relevant code mapped + vault decisions

Token Economics

What realm-recall saves

Question Without vault With realm-recall Savings
"Why did we choose JWT?" 500–2K tokens (multi-file + reasoning) ~20 tokens 97%
"What was rejected for auth?" Cannot answer without docs ~30 tokens —
"Any constraint on payment module?" Cannot answer without docs ~25 tokens —

Vault Structure

<vault>/projects/<slug>/
├── overview.md          # status, stack, milestone tracker, key file links
├── architecture.md      # service map, event shapes, data flow, schema groups
├── decisions/
│   ├── ADR-000-index.md # table of all ADRs
│   └── <id>.md          # one file per decision
├── discoveries/
│   └── YYYY-MM-DD-<topic>.md  # perf notes, bug discoveries
└── sessions/
    └── YYYY-MM-DD-<topic>.md  # per-session logs

Dependencies

Dependency Purpose
Supported host: Claude Code, Cursor, Codex, or Gemini Runtime for Realm skills
Node.js Runs Realm's installer; npx is needed for Cursor and Gemini Skills CLI installs
Obsidian 1.x+ Vault storage, graph view, backlinks
Graphify CLI (optional) Fast zero-token codebase discovery for realm-fathom & realm-planning

Diagrams: VISUALS.md — pipeline flows & architecture diagrams.

Facts

Kind
Marketplace
Repo
blackmo18/realm
Group
Uncategorized
Stars
0
Marketplace name
realm
Owner
blackmo
License
MIT
Language
Python
Created
2026-06-11
Last push
2026-08-30
Forks
1
Plugins
1

More on this shelf

  1. 1obra/superpowersobra/superpowersAn agentic skills framework & software development methodology that works.297.9k
  2. 2affaan-m/everything-claude-codeaffaan-m/ECCThe agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.277.5k
  3. 3anthropics/skillsanthropics/skillsPublic repository for Agent Skills180.6k
  4. 4f/prompts.chatf/prompts.chatf.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.172.7k
  5. 5anthropics/claude-codeanthropics/claude-codeClaude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.150.3k
  6. 6nextlevelbuilder/ui-ux-pro-max-skillnextlevelbuilder/ui-ux-pro-max-skillAn AI skill that provides design intelligence for building professional UI/UX across multiple platforms.135k