Apply when the user is working on CLAUDE.md, AGENTS.md, context files, agent memory, or AI coding assistant instructions. Teaches the minimal-context approach backed by empirical research.
Creates minimal context files for AI agents by applying research-backed principles to eliminate unnecessary instructions.
npx claudepluginhub maxnoller/claude-code-pluginsThis skill inherits all available tools. When active, it can use any tool Claude has access to.
Applies to: CLAUDE.md (Claude Code), AGENTS.md (Jules / other agents), and any equivalent agent instruction file.
Developer-written context files improve task success by only ~4% while increasing compute cost and time by 20%+. LLMs are proficient at exploring codebases independently by reading package.json, source files, and schemas. Bloated context files cause important rules to get buried and ignored.
For every line, ask: "Would removing this cause the agent to make mistakes?"
package.json / Cargo.toml — especially verification steps to run after changes| Content | Why to cut |
|---|---|
| Project description / tech stack | Agent reads package.json |
| Architecture explanations | Belongs in ADRs / docs/ |
| Directory structure | Agent explores the filesystem |
| Key file lists | Agent searches the codebase |
| Environment variable templates | Agent reads .env.example |
| Database schema docs | Agent reads schema files |
| Deployment / CI pipeline info | Irrelevant during coding |
| Anything discoverable by reading code | Agent will find it |
When an agent makes a mistake, prefer fixing the codebase over adding a context file rule:
Context files are a temporary patch. A well-structured codebase with good types and tests needs almost none of it.
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.