claude_md_files

Markdown rules and discipline skills for Claude Code, bringing a structured lifecycle approach to project development.

What this is

A drop-in set of CLAUDE.md rules and lazy-loaded skills that give Claude Code a consistent, lifecycle-aware way of working on a project. The repo provides:

• An always-loaded core of three rule files — safety, honesty, and project-specific facts — that govern every session.
• A skill index under .claude/skills/ covering product, business, marketing, engineering, quality characteristics, process, measurement, lifecycle, and formal-process disciplines. Skills load on demand when their triggers fire, so the context window stays small.
• A clear precedence model: safety and honesty rules override skills; project rules override skills; skills are subordinate guidance, not commands.

Why this exists

A Claude Code session without an entry point freelances — sometimes well, sometimes badly. This repo makes the discipline set discoverable and explicit, keeps the always-loaded core small, and lets a team change disciplines without touching every other file. Every session starts from the same baseline.

Repository layout

.
├── CLAUDE.md              # Entry point: imports always-loaded files and indexes skills
├── safety-rules.md        # Absolute rules for destructive / irreversible actions
├── honesty-rules.md       # Absolute rules on what Claude may claim and verify
├── project-rules.md       # Project-specific facts (stack, conventions, CI, ops)
├── LICENSE
├── .gitignore
└── .claude/
    └── skills/
        ├── coding-rules/SKILL.md
        ├── design-discipline/SKILL.md
        ├── testing-discipline/SKILL.md
        ├── quality-reliability/SKILL.md
        ├── … (50+ skills across product, marketing, engineering, quality, process)
        └── standard-iso-25010/SKILL.md

Quick start

1. Copy the four root files (CLAUDE.md, safety-rules.md, honesty-rules.md, project-rules.md) and the .claude/ directory into the root of your project.
2. Open project-rules.md and fill in the sections for your stack: language, framework, folder structure, naming, package manager, CI commands, environment, data stores, branching, observability. Comment out (<!-- -->) sections that do not apply.
3. Open Claude Code in your project. The core files load automatically; skills load when their triggers fire (or invoke them by name).
4. Tune individual skills by opening their SKILL.md and commenting out Layer 2 rules you do not want, or adding project-specific rules below the defaults.

Rule precedence

When two rules appear to conflict, higher priority wins:

1. safety-rules.md — absolute. Destructive / irreversible actions.
2. honesty-rules.md — absolute. What Claude may claim and how it verifies.
3. project-rules.md — project facts.
4. Skill files under .claude/skills/.
5. Claude's defaults.

Safety and honesty are co-equal at the top — they cover different concerns and both apply where both are relevant.

Skill categories

The full index lives in CLAUDE.md. At a glance:

• Product and business — product discovery, market analysis, business model, go-to-market, strategic planning, and supporting frameworks (Lean Startup, JTBD, OKRs, BCG, Five Forces, Business Model Canvas, Value Proposition Canvas).
• Marketing — brand, demand generation, content strategy, performance marketing, marketing operations, customer advocacy, AARRR, AIDA.
• Engineering — requirements, design, coding rules, testing.
• Quality characteristics — reliability, maintainability, testability, performance efficiency, security, usability, portability, compatibility, functional suitability.
• Process and documentation — developer guideline, user manual, internal documentation, error handling and observability.
• Measurement and feedback — measurement program discipline, early-warning indicators.
• Lifecycle and formal process — lifecycle models, formal artifacts, formal reviews, plus skeletons for ISO/IEC 25010, DO-178C, IEC 62304, ISO 26262, MIL-STD-498, IEC 61508, MISRA C.
• Safety — destructive-action-safety companion to safety-rules.md.

Note on standards and frameworks: Skill files for industry standards (standard-*) and named business frameworks (framework-*) are skeletons only. The substantive content of those standards and frameworks is copyrighted; teams fill in details from their own licensed copies of the source material.

Skill structure

Every skill follows the same template:

• Why this exists — the concrete problem this skill prevents or solves.
• Precedence note — reminder that the skill is subordinate to safety-rules.md, honesty-rules.md, and project-rules.md.
• Layer 1 — Lenses — always-on attention questions (Requirement, Design, Code, Test). Not optional.
• Layer 2 — Rules — opt-in / opt-out single-line rules. Comment out with <!-- -->; add project-specific rules below.
• Do / Don't list and a verification checklist.

Customizing per project

Two places to tune:

1. project-rules.md — language, framework, folder structure, naming, CI commands, environment, data stores.
2. The Layer 2 (Rules) section in each skill — comment out lines you do not adopt; add lines below for project-specific rules.

Contributing

Issues and pull requests are welcome. When proposing a new skill or modifying an existing one:

• Keep skills language- and framework-agnostic. Project-specific choices belong in project-rules.md, not in skills.
• Follow the existing skill template (Why / Precedence / Layer 1 Lenses / Layer 2 Rules / Do-Don't / verification).
• Do not paste copyrighted standard or framework content into skill files; keep them as skeletons.
• Respect the precedence: nothing in a skill can override safety-rules.md or honesty-rules.md.

License

MIT — see LICENSE.