louisbrillet/toolbox
Standalone engineer helpers independent of the COR workflow: show-me visual explanations, token-caveman terse mode, authoring-skills
Authors, audits, and refactors Agent Skills against Anthropic's skill authoring best practices — frontmatter validity, description discoverability, token budgets, progressive disclosure, workflows, and evaluations. Use when creating a new SKILL.md, reviewing or improving an existing skill, auditing a skill collection for compliance, or when the user mentions skill authoring, SKILL.md, skill descriptions, or skill token budgets.
Explains the current topic visually instead of in prose: picks the smallest fitting view (pseudocode, call tree, component tree, file tree, Mermaid diagram, diff sketch, or a single self-contained HTML artifact) grounded in the real code. Use when the user says show me, draw, diagram, sketch, visualize, map this out, what does the flow look like, or asks how pieces connect, what changes, or where something lives.
Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy. Active by default. Invoke /toolbox:token-caveman to re-assert the mode or remind Claude to stay terse.