This document explains the thinking behind this workflow system.
Claude Code sessions are stateless. When you start a new session, Claude doesn't know:
- What project you were working on
- Where you left off
- What decisions were made
- What's next on your list
You end up re-explaining context every time.
This workflow creates persistent context through:
- CLAUDE-INSTRUCTIONS.md - Per-project context that Claude reads at session start
- Session notes - Documentation of what happened
- /hello and /bye commands - Standardized start/end rituals
Code Projects
- Live in your projects folder (e.g.,
~/projects/) - Have git repositories
/hellopulls latest code/byehandles commits and pushes- Example: Mobile apps, web apps, APIs, libraries
Documentation Projects
- Live in your cloud sync folder
- No git required
- Great for non-code work
- Example: Research, planning, writing, brainstorming
MVP/Startup Projects
- Live in your projects folder (like code projects)
- Enable "AI Co-Founder" mode with autonomous execution
- Include decision frameworks, metrics tracking, and structured reviews
- Claude operates more proactively: proposing solutions, executing independently within bounds
- Example: Startups, MVPs, side projects where you want maximum AI leverage
Documentation exists at two levels:
Layer 1: Brief Summaries (Root)
- Location:
[docs_path]/SessionNotes/ - Purpose: Quick overview of all work
- Content: What projects, brief accomplishments, links to details
- Format: One file per session date
Layer 2: Detailed Notes (Per Project)
- Location:
[project]/SessionNotes/ - Purpose: Technical deep-dive
- Content: Code changes, commits, decisions, blockers
- Format: Multiple files per project, by topic
Why two layers?
- Quick context: Scan root SessionNotes to see what happened recently
- Deep dive: Read project SessionNotes only when needed
- Efficient: Don't load unnecessary detail
Every project has a CLAUDE-INSTRUCTIONS.md file. This is the single source of truth for project context.
What it contains:
- Setup instructions (how to run, dependencies)
- Current status (what's working, what's not)
- Next steps (priorities for upcoming work)
- Notes & context (decisions, constraints, gotchas)
- Last session info (date, summary)
Why it works:
- Claude reads it at session start via
/hello - Claude updates it at session end via
/bye - Context persists without you re-explaining
START SESSION
│
├─→ Run /hello
│ ├─ List available projects
│ ├─ User selects project
│ ├─ Read CLAUDE-INSTRUCTIONS.md
│ ├─ Pull latest code (if git)
│ └─ Read NEXT-SESSION-START-HERE.md
│
├─→ Work on project
│ ├─ Make changes
│ ├─ Create commits
│ └─ Document decisions
│
└─→ Run /bye
├─ Create brief summary (root)
├─ Create detailed notes (project)
├─ Update CLAUDE-INSTRUCTIONS.md
├─ Handle git push (ask user)
└─ Update NEXT-SESSION-START-HERE.md
This file answers: "Where do I pick up?"
Contains:
- Current state (branch, commit, build status)
- Last session summary with link to detailed notes
- Next priorities in order
- Blockers or waiting items
Updated by: /bye at the end of each session
The workflow includes a DEVELOPMENT-STANDARDS.md file that gets installed to ~/.claude/. This provides consistent guidelines for:
- Code principles: DRY, YAGNI, SOLID, Clean Code
- Git conventions: Commit message format, branching strategy
- Code style: Formatting, linting, testing
- Security: Secrets management, input validation
- AI guidelines: How Claude should work with your code
Projects can reference these standards, and Claude will follow them when making changes.
For code projects, the workflow integrates with git:
On /hello:
- Check if behind remote
- Offer to pull latest
On /bye:
- Show uncommitted changes
- Offer to commit
- Show unpushed commits
- Offer to push
- Document what was/wasn't pushed
Never automatic: The workflow always asks before git operations.
- Session notes capture everything
- Git status is always checked
- Uncommitted work is documented
- NEXT-SESSION-START-HERE.md provides quick context
- CLAUDE-INSTRUCTIONS.md has full project state
- Session notes have detailed history
- Documentation syncs via cloud
- Code syncs via git
- Config is per-machine (paths can differ)
/helloand/byeare simple to type- Prompts have sensible defaults
- Workflow adapts to your choices
- All files are markdown
- Can browse/edit without Claude
- History is preserved
Good fit:
- Multiple projects you switch between
- Long-running work over many sessions
- Need to track decisions and progress
- Work on multiple machines
Maybe overkill:
- One-off quick tasks
- Single project you never leave
- Don't care about history