# Rule Creation

> Wire new rules, policies, and behavioral constraints into the correct governance file with an explicit enforcement assessment.

- Skill: `levineam/rule-creation` (Agent Skill)
- Install (CLI): `npx skillmds@latest add levineam/rule-creation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/levineam/rule-creation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: levineam (https://skillmd.com/u/levineam)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/levineam/rule-creation

---


# Rule Creation

Use this skill when a user declares, proposes, or asks to formalize a behavioral
rule. Do not leave the rule as a chat promise. Wire it into the system of record,
then report what changed.

## Contract

The workflow is complete only when:

- the rule has a name, type, and target governance file
- enforcement need has been assessed
- high-stakes or multi-step rules get an enforcement mechanism when the runtime
  supports one
- the rule is written before the response claims it will be followed
- the final response names the file and enforcement status

## Rule shape

```md
## Rule Name (HARD|soft)

Plain-language requirement.

**Enforced by:** `path/to/workflow-or-script` (when applicable)
```

## Routing

| Rule type | Typical destination |
|---|---|
| Always-loaded behavior | `AGENTS.md` or equivalent root instructions |
| Tool/CLI policy | `TOOLS.md` or equivalent tooling guide |
| Non-negotiable safety | `CRITICAL-RULES.md` or root instructions |
| Persona/voice | `SOUL.md`, `IDENTITY.md`, or persona docs |
| Project-only rule | Project issue, project brief, or project docs |

## Enforcement assessment

Create enforcement when any condition is true:

- the rule affects external messages, publishing, money, secrets, production, or
  destructive actions
- a similar rule has been violated before
- the rule requires a multi-step ordered gate
- a script/workflow/tool restriction can make violation materially harder

Choose the lightest effective mechanism:

- workflow gate for multi-step approval or sequencing
- lint script for one validation check
- tool restriction when an action should never be called directly
- doc-only when judgment is enough and risk is low

## Final report

```md
Rule wired: <name>
Location: <file>
Type: HARD|soft
Enforcement: <workflow/script/tool restriction/doc-only/unavailable>
```