Cat — Productivity skill for Claude Code

Reliable agentic coding for professional cat herders

A coding agent can follow a rule in one file and miss it in the next. It can complete only part of a refactor, or apply a project convention in one area and overlook it somewhere else.

Your rules will change as the team learns. When a review uncovers a gap, add the lesson to the project rules and CAT brings it into later sessions. CAT also guides agents to explain their work clearly and avoid unnecessary jargon, making their changes easier to understand in review.

Why CAT

• Proven compliance. CAT’s rules are certified to achieve at least 95% compliance, with 95% confidence.
• Apply rules more consistently across a task. Give the agent the same guidance as it moves between files and sessions, so it is less likely to handle a convention in one place and miss it in another.
• Use rules tested for reliability. CAT certifies an optimized rule only after repeated tests support, with 95% confidence, that agents follow it at least 95% of the time in the cases tested.
• Keep rules with the codebase. Store shared guidance in the repository, where the team can change it alongside the code it affects.
• Make agent changes easier to review. Clear, plain-language explanations help reviewers understand what changed and spot gaps in the current rules.
• Improve the rules as you work. Reviews uncover new cases. Add the lesson as a project rule, and CAT can supply it in later sessions.
• Use the coding harness you prefer. CAT supports both Claude Code and Codex while keeping project guidance in one place.
• Install with confidence. CAT provides native distributions for supported Linux and macOS platforms, including checksum-verified installation and an uninstall path.

How CAT checks its rules

CAT tests and certifies its optimized rules ag