SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsclaude-code-pro-pack

claude-code-pro-pack

claude-code-pro-pack is a collection of behavioral rules and skill templates for AI coding agents, suitable for Claude Code, Codex, Cursor, and Hermes Agent. At its core is the CLAUDE.md or AGENTS.md twelve-rule behavior file, which claims to reduce AI coding errors from about 40% to about 3%, with a total size of about 700 tokens, and it takes effect simply by placing it in the project root directory. The pack is built on real agent failure patterns from more than 30 codebases, supplementing agent orchestration issues not covered by the original 4 rules, such as token spirals, silent partial failures, dual-mode contamination, and duplicate function drift. The project uses the MIT license and includes the cc-audit auditing tool and example skill templates.

Open website

About this tool

Purpose

claude-code-pro-pack is used to provide AI coding agents (such as Claude Code, Codex, Cursor, and Hermes Agent) with a set of behavioral rules, prompts, and reusable skill templates to reduce common errors during the coding process. It is built on real agent failure patterns from more than 30 codebases, with the goal of making agents more reliable when coding through a streamlined behavior constraint file.

Main capabilities

  • Provides a CLAUDE.md or AGENTS.md behavior file containing 12 rules, with a total size of about 700 tokens.
  • Claims to reduce AI coding errors from about 40% to about 3%; the original 4-rule version reduced errors from about 40% to about 11%, and this pack adds 8 additional rules covering agent orchestration failures such as token spirals, silent partial failures, dual-mode contamination, and duplicate function drift.
  • The 12 rules include: think before coding; simplicity first; surgical changes; goal-driven execution; do not make the model do non-linguistic work; hard token budget; expose conflicts rather than averaging them; read before writing; testing with correctness as the gate; long-running operations require checkpoints; convention over novelty; fail visibly rather than silently.
  • Includes cc-audit: a single-file Python checker that can score any CLAUDE.md / AGENTS.md against the 12-rule baseline and can be used in CI.
  • Provides example skill templates that can be copied to the .claude/skills/ or skills/ directory.
  • MIT license, can be forked, modified, redistributed, and used for company guidelines.

Usage

  1. Copy CLAUDE.md (use AGENTS.md for Codex) to the project root directory.
  2. Uncomment the ## Project specifics block, add your tech stack, test runner, and any "do not touch X" rules, keeping it to no more than 50 lines.
  3. Commit, and the agent will read the file on its next run.
  4. Optional: copy 2–3 skills from examples/ to .claude/skills/ or skills/. The documentation provides the full rationale for each rule, see docs/why-12-rules.md.

Applicable scenarios

Suitable for teams or individuals using AI coding agents such as Claude Code, Codex, Cursor, and Hermes Agent for project development, especially suitable for scenarios where you want to constrain agent behavior at extremely low token cost, reduce common coding errors, and check the quality of behavior files in CI.

Limitations

The project description points out that after CLAUDE.md exceeds about 200 lines, compliance drops sharply and rules are easily buried, so this pack stays at 12 rules plus minimal boilerplate. The pack itself does not include a framework. Other details such as the specific supported platform versions, installation commands, pricing, and effectiveness verification data are not yet provided.

Related tools

Browse category