From ai-dev-toolkit
Use when setting up a new project's CLAUDE.md, when the user asks to reduce token usage, add token efficiency rules, or optimize Claude Code output verbosity. Also use when creating CLAUDE.md from scratch for any project.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ai-dev-toolkit:token-efficient-setupThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Add token-efficiency rules to a project's CLAUDE.md based on the project type. Rules eliminate sycophantic openers, filler text, restating questions, and over-engineering — reducing output tokens by ~60% with zero signal loss.
Add token-efficiency rules to a project's CLAUDE.md based on the project type. Rules eliminate sycophantic openers, filler text, restating questions, and over-engineering — reducing output tokens by ~60% with zero signal loss.
Based on drona23/claude-token-efficient.
Scan the project root and first two directory levels for signals:
| Signal | Profile |
|---|---|
.ipynb files, pandas/numpy/scipy imports, *.R files, data/ dirs | analysis |
Agent configs, agents/ dir, bot frameworks, heavy CI/CD automation, MCP servers | agents |
Source code (*.ts, *.py, *.swift, *.go, *.rs, etc.), package.json, Cargo.toml, *.xcodeproj | coding |
| User explicitly says "benchmark" | benchmark |
Most projects are coding. When signals overlap (e.g., a coding project with some notebooks), pick the dominant type and mention the overlap to the user.
If CLAUDE.md exists: Append a clearly marked section at the end:
## Token Efficiency
<!-- From https://github.com/drona23/claude-token-efficient -->
<!-- Profile: {profile_name} -->
{profile rules here}
If no CLAUDE.md exists: Create one with the universal rules at the top, then the profile-specific rules below.
Never overwrite existing project-specific instructions. The token efficiency section goes at the bottom because the repo itself says "User instructions always override this file."
Read references/profiles.md for the full text of each profile. The four profiles are:
Tell the user:
npx claudepluginhub elvinouyang/claude-skill-collection --plugin ai-dev-toolkitAudits and restructures CLAUDE.md files to reduce token bloat and improve Claude compliance. Targets 2,500–3,500 tokens with a hard 5,000 limit.
Generates, analyzes, and enhances CLAUDE.md files using best practices, modular architecture, and tech stack customization for new and existing projects.
Generates project-specific CLAUDE.md rules by detecting stack from package.json, Cargo.toml, pyproject.toml, go.mod, git log, and user-selecting categories like response format, library preference, code review stance. Use for new projects, repo onboarding, or establishing conventions.