Code Abyss

Persona-driven configuration system for Claude Code, Codex CLI, Gemini CLI, and OpenClaw

Code Abyss installs a switchable persona + output style + engineering skill system into your AI coding CLI. One command configures persona rules, proactive execution guidance, output styles, 26 domain skills, and 5 verification tools across Claude Code, Codex CLI, Gemini CLI, and OpenClaw.

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/
npx code-abyss --target openclaw -y     # One-line install to ~/.openclaw/

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`

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`
OpenClaw	`~/.openclaw/openclaw.json` + `<workspace>/AGENTS.md` + `<workspace>/SOUL.md`	`~/.openclaw/skills/`	`SOUL.md`

Codex config now ships explicit presets: full_auto (workspace-write + on-request) and full_access (danger-full-access + on-request). Use codex -p full_access when you really need full filesystem access without relying on removed UI presets.

Personas

5 switchable personas, each with a distinct character, interaction style, and shared proactive assistance bias:

Slug	Name	Style
`abyss` (default)	邪修红尘仙	Direct, security-first, proactive close-the-loop
`scholar`	文言小生	Classical, methodical, proactive review notes
`elder-sister`	知性大姐姐	Warm, insightful, proactive guardrails
`junior-sister`	古怪精灵小师妹	Playful, sharp, proactive follow-through
`iron-dad`	铁壁暖阳	Firm, protective, proactive safety net

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

26 skills across 15 domains, driven by SKILL.md frontmatter as single source of truth.

User Invocation

Core skills are now routed automatically by context and are not exposed as slash commands by default. The runtime is tuned to proactively finish the closest safe loop: inspect, implement, verify, then report. Verification tools remain executable directly from the repository when needed:

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
Office Documents	Word, PDF, PowerPoint, Excel, OOXML, forms, spreadsheet automation
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    (optional)     ├── 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  (optional)
├── skills/          (domain skills)
├── settings.json    (config)
└── .sage-uninstall.js
~/.openclaw/                      <workspace>/
├── openclaw.json   (optional)    ├── AGENTS.md       (rules / routing)
├── skills/         (shared)      └── SOUL.md         (persona + style)
└── .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|openclaw> [-y]
npx code-abyss --target claude --style <slug> --persona <slug> -y

# Uninstall
npx code-abyss --uninstall <claude|codex|gemini|openclaw>

# 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 libs
packs/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

Pack management:

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.

Required frontmatter:

---
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
---

Generation chain:

Registry scans and validates all skills/**/SKILL.md
Only skills with user-invocable: true generate commands (current core set defaults to none)
Claude: renders ~/.claude/commands/*.md only when invocable skills exist
Codex: installs to ~/.codex/skills/, discovered directly, with proactive execution guidance from generated AGENTS.md + instruction.md
Gemini: renders ~/.gemini/commands/*.toml only when invocable skills exist, with proactive guidance in generated GEMINI.md
OpenClaw: installs shared skills into ~/.openclaw/skills/, and writes runtime rules/persona into workspace AGENTS.md + SOUL.md
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
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 Claude / Codex / Gemini / OpenClaw

Uninstall

npx code-abyss --uninstall claude
npx code-abyss --uninstall codex
npx code-abyss --uninstall gemini
npx code-abyss --uninstall openclaw

Backup script alternative:

node ~/.claude/.sage-uninstall.js
node ~/.codex/.sage-uninstall.js
node ~/.gemini/.sage-uninstall.js
node ~/.openclaw/.sage-uninstall.js

Restores backed-up configuration and removes all installed files.

License

MIT

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Code Abyss

Quick Start

What It Does

Personas

Output Styles

Skills

User Invocation

Domain Knowledge (auto-loaded by context)

Install Layout

CLI Reference

Pack System

Skill Registry

Development

Uninstall

License

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

Code Abyss

Quick Start

What It Does

Personas

Output Styles

Skills

User Invocation

Domain Knowledge (auto-loaded by context)

Install Layout

CLI Reference

Pack System

Skill Registry

Development

Uninstall

License