Skip to main content
code-abyss - AI Agents | SkillsLLM
Home AI Agents code-abyss 为 Claude Code / Codex CLI 注入邪修人格、4种可切换输出风格与 56 篇攻防工程秘典
ai-assistant blue-team claude-code cli codex-cli
# Add to your Claude Code skills
git clone https://github.com/telagod/code-abyssCode Abyss
Persona-driven configuration system for Claude Code, Codex CLI, and Gemini CLI
中文文档
Code Abyss installs a switchable persona + output style + engineering skill system into your AI coding CLI. One command configures persona rules, output styles, 21 domain skills, 6 slash commands, and 5 verification tools across Claude Code, Codex CLI, and Gemini CLI.
Quick Start
npx code-abyss # Interactive menu
npx code-abyss --target claude -y # One-line install to ~/.claude/
npx code-abyss --target codex -y # One-line install to ~/.codex/
npx code-abyss --target gemini -y # One-line install to ~/.gemini/
What It Does
Code Abyss is a three-layer configuration system:
| Layer | What | Where |
|-------|------|-------|
| Persona | Character identity, rules, execution chains | config/personas/*.md + config/CLAUDE.md |
| Output Style | Tone, formatting, response structure | output-styles/*.md + index.json |
| Skills | Domain knowledge + executable verification tools | skills/**/*.md + scripts/*.js |
Sign in with GitHub to leave a comment.
No comments yet. Be the first to share your thoughts!
Related Skills Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.
configuration
prompt-engineering
red-team
security
The installer generates target-specific artifacts for each CLI:
| Target | Config | Skills | Style |
|--------|--------|--------|-------|
| Claude | ~/.claude/CLAUDE.md + settings.json | ~/.claude/commands/*.md + ~/.claude/skills/ | settings.json.outputStyle |
| Codex | ~/.codex/config.toml + AGENTS.md | ~/.codex/skills/ | ~/.codex/AGENTS.md |
| Gemini | ~/.gemini/GEMINI.md + settings.json | ~/.gemini/commands/*.toml + ~/.gemini/skills/ | GEMINI.md |
Personas 5 switchable personas, each with a distinct character and interaction style:
| Slug | Name | Style |
|------|------|-------|
| abyss (default) | 邪修红尘仙 | Direct, security-first, no-nonsense |
| scholar | 文言小生 | Classical, methodical, scholarly |
| elder-sister | 知性大姐姐 | Warm, insightful, guiding |
| junior-sister | 古怪精灵小师妹 | Playful, sharp, energetic |
| iron-dad | 铁壁暖阳 | Firm, protective, structured |
Switch persona during install:
npx code-abyss --target claude --persona elder-sister -y
Output Styles 5 output styles control tone and response formatting:
| Slug | Name | Best For |
|------|------|----------|
| abyss-cultivator (default) | 宿命深渊 | Immersive, high-tension delivery |
| scholar-classic | 墨渊书阁 | Formal, structured analysis |
| elder-sister-gentle | 星霜雅筑 | Gentle, step-by-step guidance |
| junior-sister-spark | 灵犀洞天 | Fast-paced, playful interaction |
| iron-dad-warm | 钢铁柔情 | Decisive, warm mentoring |
Switch style during install:
npx code-abyss --target claude --style scholar-classic -y
npx code-abyss --list-styles # List all available styles
Skills 21 skills across 14 domains, driven by SKILL.md frontmatter as single source of truth.
Slash Commands (user-invocable) | Command | Function |
|---------|----------|
| /verify-security | Scan code for security vulnerabilities and dangerous patterns |
| /verify-module | Check directory structure and documentation completeness |
| /verify-change | Analyze git changes, detect doc sync issues |
| /verify-quality | Detect complexity, naming, code quality issues |
| /gen-docs | Generate README.md and DESIGN.md scaffolds |
| /frontend-design | UI aesthetics, component patterns, UX guidance |
Domain Knowledge (auto-loaded by context) | Domain | Coverage |
|--------|----------|
| Security | Penetration testing, code audit, defense engineering, threat intelligence, vulnerability research |
| Architecture | API design, cloud-native, security architecture, messaging, caching |
| Development | Python, TypeScript, Go, Rust, Java, C++, Shell |
| DevOps | Git workflow, testing, database, observability, performance, cost optimization |
| Frontend | Component patterns, state management, UI aesthetics, 4 design system variants |
| Mobile | iOS/SwiftUI, Android/Compose, React Native, Flutter |
| AI | Agent development, LLM security, RAG systems, prompt engineering |
| Data Engineering | Pipeline orchestration, stream processing, data quality |
| Infrastructure | Kubernetes, GitOps, IaC (Terraform/Pulumi/CDK) |
| Orchestration | Multi-agent task decomposition and parallel coordination |
Install Layout ~/.claude/ ~/.codex/
├── CLAUDE.md (persona) ├── AGENTS.md (persona + style)
├── output-styles/ (style files) ├── instruction.md (core instructions)
├── commands/*.md (slash cmds) ├── skills/ (domain skills)
├── skills/ (domain skills)├── bin/lib/ (runtime libs)
├── bin/lib/ (runtime libs) ├── config.toml (recommended config)
├── settings.json (config) └── .sage-uninstall.js
└── .sage-uninstall.js
~/.gemini/
├── GEMINI.md (persona + style)
├── commands/*.toml (commands)
├── skills/ (domain skills)
├── settings.json (config)
└── .sage-uninstall.js
All installed files are tracked in .sage-backup/manifest.json. Uninstall restores previous state.
CLI Reference # Install
npx code-abyss --target <claude|codex|gemini> [-y]
npx code-abyss --target claude --style <slug> --persona <slug> -y
# Uninstall
npx code-abyss --uninstall <claude|codex|gemini>
# Info
npx code-abyss --list-styles
npx code-abyss --help
# Verification tools (run directly)
node skills/tools/verify-security/scripts/security_scanner.js <path>
node skills/tools/verify-module/scripts/module_scanner.js <path>
node skills/tools/verify-change/scripts/change_analyzer.js --mode staged
node skills/tools/verify-quality/scripts/quality_checker.js <path>
node skills/tools/gen-docs/scripts/doc_generator.js <path>
Pack System Code Abyss supports installable packs for extending functionality per target CLI.
packs/abyss/manifest.json — core pack: persona, styles, skills, runtime libspacks/gstack/manifest.json — optional pinned upstream gstack runtime (installed only when declared in packs.lock).code-abyss/packs.lock.json — project-level pack declarations with required/optional/sources node bin/packs.js bootstrap # Initialize packs.lock
node bin/packs.js bootstrap --apply-docs # Write pack docs into README/CONTRIBUTING
node bin/packs.js diff # Show lock vs template drift
node bin/packs.js vendor-pull <pack> # Pull upstream into .code-abyss/vendor/
node bin/packs.js vendor-sync --check # CI gate: verify vendor integrity
node bin/packs.js report summary # View install reports
node bin/packs.js uninstall <pack> # Remove pack artifacts
Skill Registry skills/**/SKILL.md frontmatter is the single source of truth. The shared registry (bin/lib/skill-registry.js) normalizes metadata and feeds it to the installer and runtime.
---
name: verify-quality # kebab-case, unique
description: Code quality gate
user-invocable: true # false = knowledge-only
allowed-tools: Bash, Read, Glob # optional, default: Read
argument-hint: <path> # optional
---
Registry scans and validates all skills/**/SKILL.md
Filters user-invocable: true for command generation
Claude: renders ~/.claude/commands/*.md
Codex: installs to ~/.codex/skills/, discovered directly
Gemini: renders ~/.gemini/commands/*.toml
Scripted skills execute via skills/run_skill.js (lock + spawn + exit code passthrough)
Knowledge skills load SKILL.md content directly
Development npm test # Jest test suite (218 tests)
npm run verify:skills # Validate SKILL.md frontmatter contracts
node bin/install.js --help # Installer CLI help
CI runs on Node 18/20/22 across Linux, macOS, and Windows:
Unit tests + skill contract validation
4 verification tools (security, module, quality, change)
Smoke install/uninstall for all 3 targets on all 3 platforms
Uninstall npx code-abyss --uninstall claude
npx code-abyss --uninstall codex
npx code-abyss --uninstall gemini
Backup script alternative:
node ~/.claude/.sage-uninstall.js
node ~/.codex/.sage-uninstall.js
node ~/.gemini/.sage-uninstall.js
Restores backed-up configuration and removes all installed files.
License 57,200
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
AI Agents ai-agents anthropic
The agent that grows with you
AI Agents ai ai-agent
LLM inference in C/C++
AI Agents ggml
An open-source AI agent that brings the power of Gemini directly into your terminal.
AI Agents ai ai-agents
An AI SKILL that provide design intelligence for building professional UI/UX multiple platforms
CLI Tools ai-skills antigravity
