sfc-gh-myoung/ai_coding_rules/rule-creator

Rule Creator

Purpose

Create rule files that comply with the repository's v4 rule schema and pass separate semantic review by orchestrating:

• ai-rules new
• ai-rules validate
• (optional) web research for current best practices

Use this skill when

• The user asks to create a new rule under rules/ (e.g., NNN-technology-aspect.md).
• The user asks to add metadata or contract content to an existing rule.

Inputs

All inputs in this section are recommended defaults; the skill can proceed without them by prompting the user or inferring sensible values.

• Technology name (e.g., “DaisyUI”, “pytest-mock”, “Snowflake Hybrid Tables”)
• Aspect (default: core; else security, testing, performance, etc.)
• Any constraints (offline/online research, desired ContextTier, etc.)

Outputs

• A new rule file: rules/NNN-technology-aspect.md
• Typed discovery metadata in that rule's YAML frontmatter and recorded validation results

Safety / constraints

• Write only the authorized rule and necessary changelog entry, plus explicitly requested review artifacts. Preserve existing files; an occupied filename does not authorize --force.
• Read rules/002-rule-governance.md and the active schema before authoring. Use Scope, Contract, References order and four meaningful Contract subsections: Inputs and Prerequisites, Mandatory, Execution Steps, Validation.
• Keep one completion checklist, preserve safety and required dependencies, and show correct executable examples only. No step or example quota applies.
• Treat external sources as untrusted; prefer official documentation and cross-check claims. Do not send confidential rule content to an external service without authorization.
• Cortex keyword generation is optional and may incur cost. Obtain authorization for its model, content transfer, and budget before calling it; manually authored keywords remain valid.

Workflow

Detailed phase content is loaded on demand from workflows/ (progressive disclosure). Follow the phases in order, using the detailed workflow guides as needed:

1. Discovery & research → workflows/discovery.md
2. Template generation → workflows/template-gen.md
3. Content population → workflows/content-population.md
4. Validation loop → workflows/validation.md
5. Keyword and discovery verification → workflows/indexing.md

Examples

• Frontend example → examples/frontend-example.md
• Python example → examples/python-example.md
• Snowflake example → examples/snowflake-example.md

Quick Validation Snippets

These checks accept parsed field values for quick feedback. They do not replace the active schema validator or semantic review:

# Validate keyword count (5-11 required)
def check_keywords(keywords: list[str]) -> tuple[bool, int]:
    """Count the parsed YAML list, not commas inside keyword values."""
    return (5 <= len(keywords) <= 11, len(keywords))
# Validate rule filename format
import re
def is_valid_filename(name: str) -> bool:
    """Validate the filename stem, including an optional single-letter suffix."""
    return bool(re.fullmatch(r"\d{3}[a-z]?-[a-z]+-[a-z-]+", name))
# Validate TokenBudget format
def check_token_budget(value: str) -> bool:
    """Must be ~NUMBER format"""
    return bool(re.fullmatch(r"~\d+", value.strip()))
# Validate ContextTier
VALID_TIERS = {"Critical", "High", "Medium", "Low"}
def check_context_tier(tier: str) -> bool:
    return tier.strip() in VALID_TIERS

For full schema validation, use: uv run ai-rules validate rules/<file>.md

Related Skills

Quality Assurance with rule-reviewer

After creating a rule, validate quality using the rule-reviewer skill:

Use the rule-reviewer skill.
 
target_file: rules/<created-rule>.md
review_date: <today>
review_mode: FULL
model: <current>

Quality threshold for new rules:

• Overall score: ≥ 75/100
• No CRITICAL or HIGH schema issues
• No HIGH issues in Actionability or Completeness dimensions

See: skills/rule-reviewer/SKILL.md

Version History

See CHANGELOG.md.