Writing Rules

Writing Rules Skill

Overview

Unified system for extracting writing rules AND managing the extensible rule type registry. Handles both rule extraction operations and registry management operations.

For usage examples and workflows, see CLAUDE.md.

Operations

Extraction Operations

style - Extract style guidelines (voice, tone, word choice)

• Entry: writing-rules-skill style --type <mode> --auto-discover
• Modes: primary, technical-docs, blog, etc.
• Output: rules/style/<descriptive-name>.md

structure - Extract structure templates (document organization, section flow)

• Entry: writing-rules-skill structure --type <mode> --auto-discover
• Modes: tutorial, landing-page, api-reference, etc.
• Output: rules/structure/<descriptive-name>.md

persona - Extract target personas (audience roles, pain points, preferences)

• Entry: writing-rules-skill persona --audience-type <type> --auto-discover
• Types: technical, business, all, etc.
• Output: rules/personas/<descriptive-name>.md

publisher - Extract publisher profile (company identity, messaging, positioning)

• Entry: writing-rules-skill publisher --auto-discover
• Output: rules/publisher/publisher-profile.md (single file)

[custom-type] - Extract custom rule types (if configured in registry)

• Entry: writing-rules-skill <custom-type> --type <mode> --auto-discover
• Output: rules/<custom-type>/<descriptive-name>.md

Update Operations (Feedback-Driven)

style --update - Update existing style rule based on recent feedback

• Entry: writing-rules-skill style --type <type> --update
• Requires: Existing style rule + recent tone/style feedback
• Output: Updated style rule with backup
• Subskill: subskills/update-style.md

structure --update - Update existing structure template based on recent feedback

• Entry: writing-rules-skill structure --type <type> --update
• Requires: Existing structure template + recent structure/organization feedback
• Output: Updated structure template with backup
• Subskill: subskills/update-structure.md

persona --update - Update existing persona based on recent feedback

• Entry: writing-rules-skill persona --audience-type <type> --update
• Requires: Existing persona + recent info/length/comprehension feedback
• Output: Updated persona with backup
• Subskill: subskills/update-persona.md

Management Operations

list - List all rule types (built-in + custom) with extraction status

• Entry: writing-rules-skill list

show - Show details about a specific rule type

• Entry: writing-rules-skill show <type>

add - Add new custom rule type (interactive wizard with conflict detection)

• Entry: writing-rules-skill add

validate - Validate registry configuration and file system consistency

• Entry: writing-rules-skill validate

generate-subskill - Generate extraction subskill for custom type

• Entry: writing-rules-skill generate-subskill <custom-type>

onboard - Onboarding wizard for new teams

• Entry: writing-rules-skill onboard

Technical Details

Routing Logic

1. Parse arguments - Determine operation type and parameters
2. Load context:
   • Registry: Read rules/rules-config.yaml
   • Project context (if available)
   • Source content status
   • Existing rules
   • Feedback data (for update operations)
3. Route to subskill:
   • Management ops → subskills/manage-<operation>.md
   • Extraction ops → subskills/extract-<type>.md
   • Update ops (--update flag) → subskills/update-<type>.md
4. Pass context - Provide all necessary info to subskill

Preview Mode (Iterative Extraction)

When invoked from extract-rules subskill:

Pattern: Analyze → Preview → Approve → Extract

1. Analyze available content - Group indexed documents by type, domain, date
2. Preview sample documents - Show 3-5 examples with coverage stats
3. Get approval - User can approve, refine (use different docs), or skip
4. Execute extraction - Run with --auto-discover on approved document set
5. Review results - Show extracted rule file and key characteristics

Why preview?

• Ensures right documents analyzed
• Shows coverage before expensive AI analysis
• Allows refinement if wrong documents selected

Registry Configuration

Location: rules/rules-config.yaml

Defines:

• Which rule types are enabled (built-in + custom)
• What each type extracts and governs
• Directory locations for each type
• Extraction settings (discovery modes, sample size)

Built-in types (always available):

• style, structure, persona, publisher

Custom types (team-configurable):

• verticals, use-cases, channels, journey-stages, etc.

Extraction Modes

Incremental (default - recommended):

• Compares with existing rules
• Creates new files only if distinct patterns found
• Safe, additive approach

Overwrite (nuclear option):

• Deletes existing rules in category
• Performs fresh analysis
• Use when rules are outdated

Integration with Other Skills

Called by:

• extract-rules subskill - Routes extraction requests here with preview mode
• check-onboarding subskill - For extracting core rules (publisher, primary voice, personas)
• feedback-skill patterns subskill - Recommends update operations when patterns emerge
• Direct invocation for rule extraction or updates

Delegates to:

• Extraction subskills - Type-specific extraction logic in subskills/extract-<type>.md
• Update subskills - Feedback-driven updates in subskills/update-<type>.md
• Management subskills - Registry operations in subskills/manage-<operation>.md

Works with:

• project-management-skill - Rules are referenced in project.md
• content-writing-skill - Rules are applied during content creation
• feedback-skill - Provides feedback data to drive rule updates

Key Principles

1. Registry as source of truth - All rule types defined in rules-config.yaml
2. Extensibility - Custom types via registry configuration
3. Preview before extraction - Always show sample documents for approval
4. Incremental by default - Don't overwrite existing rules unless explicitly requested
5. Quality over quantity - Need 3-5 documents minimum for reliable extraction
6. Type-specific extraction - Each type has specialized extraction logic
7. Feedback-driven updates - Use real user feedback to improve rules over time
8. User control - Always show diff preview and get approval before updating rules

For usage examples and workflows, see CLAUDE.md. For registry configuration details, see README.md in this skill directory.