Skillboxing
Persistent project memory for Claude. A CoWork plugin that maintains a brain/ folder in your project, capturing what matters across sessions.
The problem
Every time you start a new Claude session on a project, you lose context. You re-explain the background, re-state your preferences, re-describe decisions you already made. If someone else picks up the project, they start from zero.
What skillboxing does
Skillboxing gives your project a memory. It creates and maintains a collection of short markdown files that capture decisions, key facts, open questions, preferences, and other useful project information. At the start of each session, Claude reads the brain to ensure continuity across sessions.
Anyone who opens the project can say "Review the brain and summarize where things stand". Both human teammates and AI agents can be brought up to speed in seconds. It's also great if you do a lot of context switching and need a reminder about where you left off on a project.
Install
Tell Claude:
Install the skillboxing plugin from https://github.com/zenzen-sol/skillboxing
Or install manually from the CLI:
claude plugin marketplace add zenzen-sol/skillboxing
claude plugin install skillboxing
Usage
Starting a session
Just start working. Skillboxing automatically loads brain/index.md at the beginning of every session. If no brain exists yet, it tells you how to create one.
Capturing knowledge
At the end of a session (or any time something important happens):
/skillboxing:reflect
Claude scans the conversation for things worth remembering — decisions, corrections, key facts, open questions — and writes them to brain/. It filters aggressively: only durable, non-obvious, useful information makes it in.
Cleaning up
Periodically, or after a major milestone:
/skillboxing:meditate
Claude reads the entire brain and audits it. It proposes changes — deleting stale notes, merging duplicates, tightening verbose files, filling gaps — and waits for your approval before applying them.
Getting someone up to speed
When you (or someone else) opens the project in a new session:
Review the brain and summarize where things stand.
Claude reads brain/index.md, follows the links to relevant files, and gives a summary. No re-explaining needed.
What ends up in the brain
Whatever matters for your project. Skillboxing has no opinions about your domain. Examples:
- A contracts project: key terms, party positions, open issues, concession history
- A startup project: formation status, cap table decisions, hiring pipeline
- A research project: sources, key findings, open questions
- A product launch: timeline, stakeholders, blockers, decisions made
The brain organizes itself around your project, not a predetermined schema.
How it works
Skillboxing has four components:
Two hooks (automatic)
SessionStart — When a session begins, inject-brain.sh loads brain/index.md into Claude's context. Claude sees what knowledge exists and reads relevant files before acting.
PostToolUse — When any tool modifies a file in brain/, auto-index-brain.sh checks whether brain/index.md is out of sync with the actual files on disk. If it is, it rebuilds the index automatically.
Two skills (user-invoked)
/skillboxing:reflect — Scans the current conversation and writes what's worth remembering to brain/. Creates the brain folder on first use. Each item must be durable (matters in future sessions), non-obvious (Claude wouldn't know without being told), and useful (changes behavior or provides essential context).
/skillboxing:meditate — Reads the entire brain and proposes cleanup: delete stale notes, merge duplicates, tighten verbose files, fill coverage gaps, link orphans. Presents all proposed changes for approval before applying.
Brain structure
The brain is just markdown files. There's no required structure — it adapts to your project. A typical brain might look like:
brain/
├── index.md # Auto-maintained table of contents
├── project-context.md # Background and objectives
├── key-decisions.md # What was decided and why
├── open-questions.md # Unresolved items
├── stakeholders.md # Who's involved and their roles
└── templates/
└── weekly-update.md # Reusable formats
Files are short (under 50 lines), scannable (bullets over prose), and connected (wikilinks like [[open-questions]] link related files). The index rebuilds itself automatically.
The brain folder is also browseable as an Obsidian vault — just open the brain/ directory in Obsidian to get a linked graph of your project knowledge.
Plugin structure