Rulebook — Personalized CLAUDE.md & AGENTS.md Compiler

Turn your team's actual recurring agent mistakes — not generic advice — into a short, enforced CLAUDE.md/AGENTS.md rulebook, then check every diff against it before merge.

What this Claude Skill does

• Runs a structured interview that pulls out this repo's real failure patterns — where the agent overreaches, what it keeps guessing wrong, which mistake has recurred more than once — instead of starting from a list of generic best practices
• Classifies every answer into scope, assumption, regression, or verification, and drops anything that would apply equally to any codebase, so nothing in the output is filler
• Compiles a rulebook under 60 lines using the included templates, with every rule ending in a concrete way to check it — a command, a pattern to grep, a file to inspect
• Ships a working diff-auditor script that reads the compiled rulebook, scans a diff, flags pattern-checkable violations automatically, and lists everything else as a manual check with the exact rule attached — so nothing gets silently skipped

Built for

• Teams who installed a generic agent-instructions file and found it still doesn't catch their repo's specific recurring mistakes
• Engineering leads who want agent output checked against agreed conventions before it merges, not just read once and forgotten
• Anyone maintaining both CLAUDE.md and AGENTS.md across different tools who wants them compiled from the same source instead of drifting apart

What's inside

• SKILL.md — the full skill file with frontmatter and the two-phase methodology: compile (interview → rulebook) and audit (diff → violation report)
• scripts/diff_auditor.py — a standard-library-only Python script that parses a compiled rulebook's verification lines and checks a diff against them
• templates/claude-md-template.md and templates/agents-md-template.md — output templates so both files compile from one interview
• reference/interview-questions.md — the structured interview the skill runs in Phase 1
• reference/failure-pattern-library.md — 20 documented agent failure patterns across scope, assumption, regression, and verification, each mapped to the rule shape that fixes it
• LICENSE.md — usage terms

How a Claude Skill works

This is a Claude Skill — a single SKILL.md file Claude reads as instructions for a specific kind of work. Drop it in your Claude Project, .claude/skills/ folder, or wherever Claude can read it. Claude triggers the skill automatically when your prompt matches its description, then runs the methodology end to end.

Author

Built by Ukiyo Productions — a strategy-led branding and growth agency. Part of a 100+ skill marketplace built for founders, operators, consultants, and agencies.