Modular Context

AI-native knowledge terminal for Obsidian — agent dashboard, skills sidebar, wiki-link autocomplete, and the modular-context methodology. By receptionOS.

---

What is this?

An Obsidian plugin that turns your vault into an LLM Wiki — a persistent, compounding knowledge base where AI agents do the bookkeeping and you focus on thinking.

Terminal — Real PTY terminal (zsh) inside Obsidian with multi-tab, wiki-link autocomplete [[, drag-and-drop, fullscreen with split layouts.

Agent Dashboard — Skills sidebar to launch Claude Code sessions with one click. Working / To Review agent tracking. Auto-mode for autonomous operation.

Methodology — Opinionated vault structure: frontmatter standards, cadence-based staleness, three-layer architecture (Sources → Wiki → Schema).

---

Features

• Real PTY terminal — Full zsh shell in a pseudo-terminal, not a basic command runner
• Multi-tab support — Multiple terminal sessions with named, persistent tabs
• Wiki-link autocomplete — Type [[ inside the terminal to search vault notes
• Skills sidebar — One-click Claude Code agent sessions for recurring workflows
• Agent tracker — Working / To Review / Standby states for running agents
• Custom skills — Add your own via the [+] button (maps to /skill-name commands)
• Drag-and-drop — Drag files from Finder or Obsidian to paste shell-escaped paths
• Screenshot drop — Drag macOS screenshot thumbnails directly into the terminal
• Fullscreen mode — Expand terminal to fill Obsidian with split layouts (single, side-by-side, stacked, grid)
• Terminal bookmarks — Mark important output positions with scrollbar indicators
• Output capture — Save terminal output to notes (Cmd+Shift+S)
• Session persistence — Tab names and layout survive restarts
• Auto-onboarding — First install triggers a setup agent that builds your vault structure

---

The Modular Context Methodology

Three Layers

1. Raw Sources (_transcripts/, _transcripts-backlog/) — Your curated source material. Immutable. The LLM reads but never modifies.
2. The Wiki (project folders with *_index.md) — LLM-generated, interlinked knowledge modules. Summaries, entity pages, syntheses.
3. The Schema (CLAUDE.md + _claude/) — Conventions, templates, and skill references that teach the LLM how your vault works.

Three Operations

Operation	What it does	Skill
INGEST	Process new sources into wiki modules	/process-transcripts
QUERY	Ask questions, synthesize answers, file insights back	/brief, /ideas
LINT	Health-check: staleness, orphans, broken links	/pulse, /vault-audit, /reweave, /graph

Frontmatter Standard

Every file gets structured metadata:

---
title: Module Name
updated: 2026-04-05
status: stable        # stable | draft | needs-update | stub
cadence: tactical     # hot (7d) | tactical (30d) | iron-cold (60d) | frozen
depends-on: [[related-file]]
sources: [[transcript-name]]
---

Cadence System

Temperature-based staleness scoring:

staleness = days_since_update / cadence_days

< 0.5  → fresh (green)
0.5–1  → aging (yellow)
1–2    → stale (orange)
> 2    → critical (red)

A pipeline.md (hot, 7d) untouched for 10 days is stale (ratio 1.4). A vision.md (iron-cold, 60d) untouched for 40 days is fresh (ratio 0.67). The vault tells you what needs attention.

---

Built-in Skills

Skill	Description
Ingest Data	Process new sources — categorize, extract insights, update wiki modules
Pulse	Vault health check — staleness radar, strategic questions, next steps
Brief	Generate PDF brief or one-pager from vault knowledge
Log	Close session — generate session log, commit changes
Ideas	Generate new ideas from vault context using creative triggers
Reweave	Cascade-update stale or disconnected modules
Vault Audit	Audit vault structure — broken links, orphans, naming issues
Graph	Analyze knowledge graph — clusters, bridges, dependency depth
Graduate	Promote buried transcript insights into standalone modules

Add custom skills with the [+] button. Each skill maps to a Claude Code /skill-name command defined in .claude/skills/.

---

Installation

With BRAT (recommended — auto-updates)

1. Install the BRAT plugin from Obsidian's community plugins
2. Open BRAT settings and click Add Beta plugin
3. Enter klemensgc/modular-context-obsidian-plugin and click Add Plugin
4. Enable the plugin in Settings > Community Plugins

Manual

1. Download from the latest release
2. Unzip into your vault's .obsidian/plugins/modular-context/
3. Restart Obsidian, then enable the plugin in Settings > Community Plugins

---

Requirements

• macOS (desktop only — uses Python PTY for real terminal)
• Python 3 (included with macOS)
• Claude Code CLI (for agent skills — install guide)

---

Usage

Opening the terminal

• Click the receptionOS icon in the ribbon (left sidebar)
• Or use Command Palette → "Open Terminal"

Terminal basics

• + — New tab (new zsh session)
• Double-click tab name — Rename it
• [[ — Wiki-link autocomplete from inside the terminal
• Drag files onto the terminal — Pastes shell-escaped path
• Cmd+Shift+S — Capture terminal output to a note

Skills

Click any skill in the sidebar to launch a Claude Code session that executes it. The agent tracker shows:

• Working — Agent is actively running
• To Review — Agent finished, output ready for your review
• Standby — Dismissed but still available

Auto-mode

Toggle in the sidebar. When enabled, Claude Code runs with --dangerously-skip-permissions for fully autonomous operation. Off by default — enable when you trust the agent to work independently.

---

Onboarding

On first install, the plugin automatically opens a setup wizard. Click "Start Here →" to launch an AI agent that:

1. Scans your vault structure
2. Checks if CLAUDE.md exists (offers to extend or create fresh)
3. Diagnoses which layer is weakest (sources, wiki, schema)
4. Recommends and builds the modular-context structure for you

You can re-trigger onboarding anytime from the sidebar.

---

Building from source

git clone https://github.com/klemensgc/modular-context-obsidian-plugin.git
cd modular-context
npm install
npm run build

Copy main.js, manifest.json, and styles.css to your vault's .obsidian/plugins/modular-context/.

---

Credits

Built on internetvin-terminal by Vin Verma (MIT License). Extended with agent dashboard, skills sidebar, onboarding system, and the modular-context methodology.

Inspired by Andrej Karpathy's LLM Wiki concept.

---

License

MIT — see LICENSE