Core Concepts

A mental model of how the toolkit works, before you start using it.

Sessions

Every Claude Code session starts cold — no memory of past work. The toolkit fixes this by reading your project's context files at session start (via a shell hook) and printing a one-line summary. From that point, Claude knows the project, stack, and what was being worked on last time.

This happens automatically. You don't trigger it.

Memory

The toolkit stores project context in Markdown files in your vault, not inside the project repo. There are two layers:

• Per-project (.claude/memory/ inside the project repo) — what the project is, what stack it uses, what's been built, what's next, and what was learned
• Global (vault/05-areas/claude-memory/) — cross-project learnings, model preferences, plans, specs, and the active project pointer

Memory is read once at session start, in parallel. Within a session, Claude works from one-line summaries. Files are never re-read unless a specific section is explicitly needed.

See Memory for the full protocol.

Routing

You don't need to memorize slash commands. The toolkit detects intent from plain English and routes to the right workflow automatically:

You say	What runs
"add [feature]"	/implement — full 5-phase workflow
"fix [bug]"	superpowers:systematic-debugging
"push this"	/git-push
"new project"	/new-project
"review this"	/standards-check

See Routing for the full table.

Standards Gate

Before Claude can write any code in a session, the standards for your stack must be loaded. A shell hook (pre-tool-standards-gate.sh) enforces this — it blocks file edits until a session flag is set by /load-standards.

The gate is active in any project that has a .claude/memory/MEMORY.md file (created by /new-project). Projects without that file bypass the gate silently.

Sub-agent Architecture

Long workflows (like /implement) run as orchestrated sub-agents. The main session acts as a coordinator — it holds file paths and one-line summaries, not file content. Each phase runs as a fresh sub-agent that reads only the files it needs, does its work, and writes a bounded output file. The main session reads only the status line.

This keeps the main context window clean across long sessions and makes workflows fully resumable — if Claude is interrupted after Phase 2, the next session reads the existing plan.md and picks up at Phase 3.