How do I install the Create Rule skill?

Run npx skillmds@latest add richfrem/create-rule in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Create Rule safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: reads secrets. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Create Rule?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Create Rule free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Create Rule?

richfrem (@richfrem) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Create Rule by richfrem — Agent Skill

Name: Create Rule
Rating: 4.666666666666667 (3 reviews)
Author: richfrem

Create Rule

Scaffolds a lean, invariant-driven agent rule following universal best practices. Enforces hard constraints, zero incident post-mortems, zero dates, zero usernames, high information density, and strict separation between rules (constraints) and skills (procedures).

richfrem 1b08699 4 files · 4.1 KB Updated Sep 10, 2026

File contents

create-rule: Universal Agent Rule Scaffolder

Creates lean, invariant-driven markdown rules (.agent/rules/<rule-name>.md or plugins/<plugin>/rules/<rule-name>.md) adhering to strict agentic engineering standards.

The 5 Invariants of Agent Rules

Rules are loaded into context continuously or matched across file interactions. Every line consumes token budget. Every rule must satisfy:

Strict Invariants, Not Narrative: State hard constraints (MUST, NEVER, ALWAYS), forbidden actions, and deterministic verification commands.
Zero Historical Fluff or Post-Mortems:
- ❌ NO session dates (2026-08-18), incident stories ("Task 7 failed because..."), commit hashes, or session diaries.
- ❌ NO user directory references (/Users/..., C:\...).
- ❌ NO ADR historical changelog citations.
Universally Applicable: Rules must be fully portable and valid when installed into any external downstream repository. No repo-specific domain files (ta-sweep-results.json, portfolio.json).
High Information Density:
- Target 30–70 lines. If a rule exceeds 80 lines, strip background exposition or split concerns.
Rule vs. Skill Distinction:
- Rules define hard constraints, invariants, and guardrails (passive policy).
- Skills define executable, multi-step procedures and interactive interviews (active execution).

Standard Rule Template

---
trigger: always_on | on_match
description: Concise one-sentence summary of the constraint and why it exists.
globs: ["**/*"]
---

# Rule Title

## The Law

> **Core invariant stated in 1-2 sentences.** Every agent action touching [Scope]
> MUST comply with this invariant before execution.

## Invariants & Forbidden Actions

1. **[Invariant 1]**: Concrete MUST / NEVER constraint.
2. **[Invariant 2]**: Specific forbidden action and failure condition.
3. **[Invariant 3]**: Deterministic verification requirement.

## Verification & Recovery

```bash
# Deterministic verification command
<verification-command>

If verification fails: .


---

## Workflow

### 1. Discovery & Intent
Identify:
- **Rule Name**: lowercase-hyphen slug (e.g., `destructive-action-guard`).
- **Core Invariant**: What is forbidden? What must be verified?
- **Scope/Trigger**: Is it `always_on` or scoped to file patterns via `globs`?
- **Target Location**: Canonical source in `plugins/<plugin>/rules/` or local repo `.agent/rules/`.

### 2. Validation Gate
Before writing the rule, verify:
- [ ] Are all dates (`YYYY-MM-DD`) and incident names omitted?
- [ ] Are all personal usernames and absolute machine paths omitted?
- [ ] Is the entire text under 80 lines?
- [ ] Are constraints expressed as actionable invariants with verification commands?

### 3. Execution & Symlink Registration
- Write canonical rule to `plugins/<plugin>/rules/<rule-name>.md`.
- Symlink to `.agent/rules/<rule-name>.md` via `symlink_manager.py`.
- If modifying rule mirrors across `CLAUDE.md`, `AGENTS.md`, `GEMINI.md`, and `.github/copilot-instructions.md`, sync via `python3 plugins/cli-agents/scripts/sync_instruction_files.py`.

---name: create-ruledescription: Scaffolds a lean, invariant-driven agent rule following universal best practices. Enforces hard constraints, zero incident post-mortems, zero dates, zero usernames, high information density, and strict separation between rules (constraints) and skills (procedures).---# create-rule: Universal Agent Rule ScaffolderCreates lean, invariant-driven markdown rules (`.agent/rules/<rule-name>.md` or `plugins/<plugin>/rules/<rule-name>.md`) adhering to strict agentic engineering standards.---## The 5 Invariants of Agent RulesRules are loaded into context continuously or matched across file interactions. Every line consumes token budget. Every rule must satisfy:1. **Strict Invariants, Not Narrative**: State hard constraints (`MUST`, `NEVER`, `ALWAYS`), forbidden actions, and deterministic verification commands.2. **Zero Historical Fluff or Post-Mortems**: - ❌ NO session dates (`2026-08-18`), incident stories ("Task 7 failed because..."), commit hashes, or session diaries. - ❌ NO user directory references (`/Users/...`, `C:\...`). - ❌ NO ADR historical changelog citations.3. **Universally Applicable**: Rules must be fully portable and valid when installed into *any* external downstream repository. No repo-specific domain files (`ta-sweep-results.json`, `portfolio.json`).4. **High Information Density**: - Target **30–70 lines**. If a rule exceeds 80 lines, strip background exposition or split concerns.5. **Rule vs. Skill Distinction**: - **Rules** define *hard constraints, invariants, and guardrails* (passive policy). - **Skills** define *executable, multi-step procedures and interactive interviews* (active execution).---## Standard Rule Template```markdown---trigger: always_on | on_matchdescription: Concise one-sentence summary of the constraint and why it exists.globs: ["**/*"]---# Rule Title## The Law> **Core invariant stated in 1-2 sentences.** Every agent action touching [Scope]> MUST comply with this invariant before execution.## Invariants & Forbidden Actions1. **[Invariant 1]**: Concrete MUST / NEVER constraint.2. **[Invariant 2]**: Specific forbidden action and failure condition.3. **[Invariant 3]**: Deterministic verification requirement.## Verification & Recovery```bash# Deterministic verification command<verification-command>```- If verification fails: <immediate rollback or corrective action>.```---## Workflow### 1. Discovery & IntentIdentify:- **Rule Name**: lowercase-hyphen slug (e.g., `destructive-action-guard`).- **Core Invariant**: What is forbidden? What must be verified?- **Scope/Trigger**: Is it `always_on` or scoped to file patterns via `globs`?- **Target Location**: Canonical source in `plugins/<plugin>/rules/` or local repo `.agent/rules/`.### 2. Validation GateBefore writing the rule, verify:- [ ] Are all dates (`YYYY-MM-DD`) and incident names omitted?- [ ] Are all personal usernames and absolute machine paths omitted?- [ ] Is the entire text under 80 lines?- [ ] Are constraints expressed as actionable invariants with verification commands?### 3. Execution & Symlink Registration- Write canonical rule to `plugins/<plugin>/rules/<rule-name>.md`.- Symlink to `.agent/rules/<rule-name>.md` via `symlink_manager.py`.- If modifying rule mirrors across `CLAUDE.md`, `AGENTS.md`, `GEMINI.md`, and `.github/copilot-instructions.md`, sync via `python3 plugins/cli-agents/scripts/sync_instruction_files.py`.

Frequently asked questions