WorkspaceBrowse, save, and manage patterns without leaving the shell

Context-Minimalist Workflow

A deliberately tiny CLAUDE.md (under 50 lines) paired with aggressive /clear usage, plan mode, and letting the codebase speak for itself.

minimalWeb App
Q
Quzr27@quzr27

software engineer

1 followerquzr27
AI Tools
claude-code
File structure2 files
13
Pros
Maximizes instruction-following for the few rules that matter
Lets the codebase teach the AI through in-context learning
Zero maintenance burden — nothing to keep in sync
Cons
Requires a well-structured, consistent codebase to work — messy codebases will confuse Claude
New team members get no onboarding context from the AI setup
Hard to debug when Claude makes mistakes — no documented expectations to point to
Avoid if
Your codebase has inconsistent patterns that would confuse the AI
New team members frequently onboard and need AI context as a learning tool
You have critical rules that must always be followed — put those in writing

Description

Research shows that LLMs are in-context learners — if your codebase follows consistent patterns, Claude will pick them up from a few file reads without being told. This pattern takes that insight to the extreme: CLAUDE.md contains only build/test commands, critical gotchas, and things Claude consistently gets wrong. Everything else is discovered at runtime. Use /init to bootstrap, then delete everything Claude could infer. Run /clear between tasks to reset context. Use plan mode for complex changes. The HumanLayer team runs 60 lines and still sees inconsistent adherence. The lesson: fewer, higher-signal instructions beat comprehensive documentation. Never send an LLM to do a linter's job.

Who is this for

Experienced developers who prefer Claude to learn from their code rather than from instruction files. Best for well-structured codebases with consistent patterns.

Why files are split this way

Context window tokens are a finite resource. Claude Code's system prompt uses ~50 instructions. Each instruction you add dilutes all others uniformly — the model does not just ignore new ones, it begins to follow everything worse. A minimalist CLAUDE.md preserves instruction-following quality for the rules that actually matter.

What changed from the previous setup

A long, comprehensive CLAUDE.md that Claude increasingly ignored as instruction count grew.