Claude-Beyond-Code

A curated library of Output Styles that transform Claude Code into specialized agents for non‑coding roles while keeping its local powers (read/write files, run scripts, TODOs).

Why Output Styles

• Output Styles replace the default system prompt and turn Claude into a different agent with its own rules and formatting.
• Pair them with Slash Commands, Subagents, and Hooks to ship ready-to-use workflows.

Included Styles (MVP)

• 🖥️ html-terminal: Claude outputs retro terminal‑style HTML only with strict formatting and saving rules.
• 📝 docs-writer: Technical documentation specialist with templates (README, ADR) and writing rules.
• 🚨 sre-incident-scribe: Incident log keeper with chronological updates and structured postmortems.

Quickstart

1. Project-level installation (recommended)

cp -r path/to/claude-beyond-code/.claude .

2. Activate a style

/output-style html-terminal
/output-style docs-writer
/output-style sre-incident-scribe

3. Use style commands

/docs:readme
/sre:triage "Database connection failures"
/sre:update "Applied hotfix v2.1.1"

Subagents (optional, MVP-aligned)

Project subagents live in .claude/agents/. Claude can delegate to them automatically when appropriate, or you can invoke them explicitly (see /agents UI or ask "Use the X subagent …").

• 🔎 log-slicer (for SRE)

  • Tools: Read, Grep, Glob
  • When to use: Segment and summarize large logs during incidents. Returns structured findings, evidence, metrics, and next steps.
  • Example: "Use the log-slicer subagent to analyze logs in logs/ for 'timeout' patterns".

• 📎 snippet-extractor (for Docs)

  • Tools: Read, Grep
  • When to use: Extract code snippets with context (path/language) to enrich READMEs/ADRs.
  • Example: "Have snippet-extractor pull usage examples for the Foo API from src/".

• 💾 html-saver (for HTML Terminal)

  • Tools: Write
  • When to use: Persist HTML produced by html-terminal using terminal_output_[name]_[YYYYMMDD_HHMMSS].html.
  • Example: "Ask html-saver to persist the last HTML response as a preview file".

Note: These subagents are optional and scoped with minimal permissions to keep the MVP safe.

Agent Skills (NEW)

Agent Skills are modular, auto-invoked capabilities that extend Claude Code. Unlike Output Styles (manual activation) or Subagents (delegation), Skills are automatically loaded by Claude when your request matches their description.

Included Skills

🎯 commit-message-generator (Basic - Level 1)

Single-file skill for generating conventional commit messages.

• Auto-activates when: You mention commits, git commit, or staged changes
• Structure: Single SKILL.md file
• Example: "Write a commit message for these auth changes"
• See: .claude/examples/commit-message-generator-example.md

📋 documentation-structure-validator (Intermediate - Level 2)

Multi-file skill with progressive disclosure for validating documentation.

• Auto-activates when: You mention docs validation, README check, or structure review
• Structure: SKILL.md + templates.md + validation-rules.md
• Features: Validates READMEs, ADRs, API docs against templates
• Tools: Read, Grep, Glob (read-only for safety)
• Example: "Check if our README is complete"
• See: .claude/examples/documentation-structure-validator-example.md

🔍 incident-log-analyzer (Pro - Level 3)

Pro-level skill with code execution for analyzing production logs.

• Auto-activates when: You mention log analysis, incidents, errors, or troubleshooting
• Structure: SKILL.md + Python scripts for parsing and pattern detection
• Features:
  • Parse logs in various formats (JSON, text, structured)
  • Detect error patterns and clusters
  • Generate timeline visualizations
  • Root cause analysis with evidence
• Scripts: parse_logs.py, pattern_detector.py, timeline_visualizer.py
• Example: "Analyze the logs for database errors in the last 2 hours"
• See: .claude/examples/incident-log-analyzer-example.md

Skills vs Output Styles vs Subagents

Feature	Output Styles	Subagents	Agent Skills
Activation	Manual (/output-style)	Delegated or explicit	Automatic by Claude
Scope	Complete system prompt	Focused task	Modular capability
When to use	Transform agent role	Delegate specific task	Extend capabilities
Example	"Be an SRE scribe"	"Use log-slicer"	"Analyze these logs"

Skill Levels (Progressive Disclosure)

Based on Anthropic's Agent Skills design:

• Level 1 - Basic: Single SKILL.md file, all context loaded at once
• Level 2 - Intermediate: Multiple files, Claude loads selectively as needed
• Level 3 - Pro: Includes executable scripts, deterministic code execution without context loading

Installation

Skills are automatically discovered by Claude from:

~/.claude/skills/        # Personal skills (all projects)
.claude/skills/          # Project skills (shared via git)

Project installation (included):

cp -r path/to/claude-beyond-code/.claude .

Usage

Simply describe what you need—Claude will activate the relevant Skill:

# Commit message generation
"Write a commit message for my staged changes"

# Documentation validation  
"Check if the README follows best practices"

# Log analysis
"Analyze the application logs for errors"

Creating Your Own Skills

See .claude/skills/ for examples at each level. Basic template:

---
name: your-skill-name
description: What it does and when to use it. Be specific and include trigger words.
---

# Your Skill Name

## Instructions
Step-by-step guidance for Claude.

## Examples
Concrete usage examples.

For detailed guidance, see the official Claude Skills documentation.

Repository Structure

Actual (MVP + Skills)

.claude/
├── output-styles/
│   ├── html-terminal.md
│   ├── docs-writer.md
│   └── sre-incident-scribe.md
├── agents/
│   ├── log-slicer.md
│   ├── snippet-extractor.md
│   └── html-saver.md
├── skills/                          # ← NEW: Agent Skills
│   ├── commit-message-generator/    # Level 1: Basic
│   │   └── SKILL.md
│   ├── documentation-structure-validator/  # Level 2: Intermediate
│   │   ├── SKILL.md
│   │   ├── templates.md
│   │   └── validation-rules.md
│   └── incident-log-analyzer/       # Level 3: Pro
│       ├── SKILL.md
│       └── scripts/
│           ├── parse_logs.py
│           ├── pattern_detector.py
│           └── timeline_visualizer.py
├── commands/
│   ├── docs-readme.md
│   ├── sre-triage.md
│   └── sre-update.md
├── examples/
│   ├── docs-writer-example.md
│   ├── sre-scribe-example.md
│   ├── log-slicer-example.md
│   ├── snippet-extractor-example.md
│   ├── html-saver-example.md
│   ├── commit-message-generator-example.md      # ← NEW
│   ├── documentation-structure-validator-example.md  # ← NEW
│   └── incident-log-analyzer-example.md         # ← NEW
└── recipes/
    └── sre-quick-playbook.md

Reference (extended from skeleton.md plan)

/styles         # Styles by category with README, examples, compatibility notes
/commands       # Slash command templates per style
/agents         # Optional subagents for focused delegation by task
/hooks          # Validation/safety hooks (e.g., block writes in advisory-only styles)
/recipes        # Playbooks (style + commands + agents + hooks)
/evals          # Deterministic checks (schema markdown, unified diff, etc.)
/examples       # End-to-end demos per style
.claude/output-styles  # Project-level style templates, ready to copy
.claude/commands       # Project-level command templates
.claude/examples       # Project-level examples
.claude/recipes        # Project-level recipes

Principles

• Keep it simple and safe (least privileges).
• Clarity over complexity.
• Styles transform behavior; commands and subagents operationalize it.

Short Roadmap

• Minimal hooks: block Write in advisory-only styles; diff validator for future security styles.
• Workflow recipes (incident write-up, quick ADR).
• Lightweight evals and CI to validate formats.

Optional SEO System (Agent-first)

The SEO pack lives in systems/seo/ to keep .claude/ clean. It's opt-in and agent-first (subagents over slash commands). Install via terminal:

• Project-level copy (recommended)

chmod +x systems/seo/install.sh
./systems/seo/install.sh --project --copy --all

• User-level symlinks (profile-wide)

./systems/seo/install.sh --user --symlink --agents --styles

• Minimal agent-first (agents + styles only)

./systems/seo/install.sh --project --copy --agents --styles

Usage after install:

• Switch style: /output-style seo-strategist
• Subagents: /agents brief-architect, /agents keyword-clusterer, /agents optimizer, /agents refresh-auditor, /agents cluster-planner, /agents gap-analyst, /agents serp-analyst, /agents entity-clusterer, /agents internal-linker
• Commands are optional wrappers if you also install systems/seo/commands/

Safety: all agents are advisory by default (Markdown/diffs first). Existing hooks can block writes unless explicitly approved.

Uninstall: remove copies/symlinks under .claude/.

Contributing

Open an issue with the proposed style/subagent, a clear purpose, and two usage examples. See CONTRIBUTING.md.