Skip to content

/upfront:enlighten

/upfront:enlighten audits your project’s AI instruction files (CLAUDE.md, AGENTS.md) and improves them with stack-specific examples and the three-tier boundary pattern (always / ask first / never).

  • First time setting up AI in a repo
  • After choosing a tech stack — enlighten generates conventions and examples specific to your stack
  • When AI keeps making the same mistakes — add a rule to prevent it
  • Periodically as the project evolves

This skill is suggested automatically when /upfront:explore or /upfront:feature detects missing or weak instruction files.

  1. Find instruction files — checks for CLAUDE.md, AGENTS.md, .cursorrules, etc.
  2. Detect tech stack — from package.json, go.mod, pyproject.toml, etc.
  3. Audit six areas — commands, project structure, code style, testing, git workflow, boundaries
  4. Present gaps — scored as Present / Missing / Weak
  5. Generate improvements — stack-specific examples showing good and bad patterns
  6. Write to file — adds to existing file, never overwrites

Research from 2,500+ repositories shows these are what separate instruction files that work from those that don’t:

  1. Commands — exact build/test/lint commands with flags, near the top
  2. Project structure — where code goes, one sentence per directory
  3. Code style — with good AND bad examples in your language
  4. Testing — framework, patterns, what to mock and what not to
  5. Git workflow — commit format, branch naming
  6. Boundaries — always do / ask first / never do

One real code snippet showing your style beats three paragraphs describing it. Every suggestion includes a concrete, stack-specific example.