Code Abyss: Persona and Engineering Skills for Coding Agents
Give supported coding agents a consistent, configurable voice and a library of on-demand engineering skills.
- Source repo
- telagod/code-abyss
- Stars
- ★ 244
- Last updated
- 2mo ago
- License
- MIT
- Primary language
- JavaScript
- FA score
- 66/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A Claude Code team wants to install a shared persona, output style, and engineering skill set.
- Not a fit if
- Teams that need direct installation into an unlisted coding agent
- Users who expect the separate abyss code-graph CLI to be included
- Source review
- 66/100 · Some gaps
What does this agent do, and when should you use it?
Code Abyss is a command-line installer for Claude Code, Codex CLI, Gemini CLI, and OpenClaw. It layers personas, output styles, and a discipline kernel, alongside 30 domain skills and four native security skill suites. The installer writes platform-specific configuration, skills, and styles into the chosen agent environment; personas and styles can be combined, while a router invokes kernel bundles on demand. Installation enables a character Stop-hook by default for Claude Code and Codex, with an opt-out. Code graph features come from the separate abyss Rust CLI, which must be installed independently.
Run npx code-abyss -t <target> -y to install personas, skills, output styles, and platform configuration for a selected agent; supported targets include claude, codex, gemini, and openclaw. Choose combinations with --persona <slug> and --style <slug>. The bundled abyss persona works offline; other personas are fetched from GitHub on first use and cached. The installed content includes engineering and security skills, a router-invoked discipline kernel, and files for the selected platform. Use code-abyss compose to recombine configuration, code-abyss doctor for health checks and migration hints, and --uninstall <target> to remove an installation and restore backups. The indexing-code skill documents how to call the separate abyss CLI; code-abyss does not install that CLI.
- A Claude Code team wants to install a shared persona, output style, and engineering skill set.
- A Codex CLI developer wants on-demand engineering judgment and security knowledge with a default character enforcement hook.
- A team using Gemini CLI or OpenClaw wants to configure those coding agents with the same persona and skill content.
- A security engineer needs domain skills spanning application security, cloud and supply chain, detection and response, and security architecture.
- A developer who wants to turn repeated workflows into reusable skills can use
cultivating-skillswith its safety scan and tiered publishing flow.
How do you install or deploy this agent?
Requires Node.js and a target coding agent. Choose a platform and run its install command:
npx code-abyss -t claude -yReplace claude with codex, gemini, or openclaw as needed. Claude Code can also install it as a plugin:
claude plugin install code-abyssCharacter enforcement is enabled by default for Claude Code and Codex; use --no-enforcement to opt out. The README does not specify a Node.js version or the account requirements of each target agent.
How do you use this agent?
Run the interactive installer to choose a target, persona, and style:
npx code-abyssOr specify the target, persona slug, and style directly:
npx code-abyss -t claude --persona elder-sister --style abyss-cultivator -yList styles, check the installation, or recombine configuration:
npx code-abyss --list-styles
npx code-abyss doctor
npx code-abyss compose -t claude --persona <slug> --style <slug>For code graph features, install the separate abyss CLI and then run commands such as abyss attach claude; its installation channels are documented by the separate abyss project.
What are this agent's strengths and limitations?
- Supports Claude Code, Codex CLI, Gemini CLI, and OpenClaw with platform-specific configuration.
- Combines six personas with six output styles; the README reports 36 validated combinations.
- Includes 30 domain skills and nine on-demand discipline kernel bundles covering engineering and security.
- Tracks installed files in a manifest; uninstall restores prior configuration and preserves custom skills.
- The bundled
abysspersona works offline.
- First use of personas other than
abyssrequires fetching them from GitHub, so that step needs network access. - The code graph CLI is a separate product to install; v5 no longer downloads it or injects graph hooks for Claude, Codex, or Gemini.
- The README identifies v5.0.0 as
rc.1, a release candidate. - Users need to understand their target agent's configuration; the README does not state a Node.js version or agent account requirements.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Code Abyss: Persona and Engineering Skills for Coding Agents This agent | 66 · Some gaps | CLI | ★ 244 | 2mo ago | JavaScript | Codex · Claude Code |
| Agent Skills Library | 67 · Some gaps | Agent plugin / skillFree + model costs | ★ 240 | 1mo ago | Python | Codex · Claude Code |
| EOC — Everything OpenAI Codex Workflow OS | 75 · Good | Agent plugin / skillFree + model costs | ★ 85 | 4mo ago | JavaScript | Codex |
| Loop Library: Reusable AI Agent Workflows | 61 · Some gaps | Agent plugin / skillFree | ★ 3.2k | 3mo ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The installer targets user-level configuration, documents uninstall and backup restoration, and explains that remote personas are fetched from GitHub on first use and cached. It also identifies the abyss CLI as a separate install, giving some data-flow clarity. Rollback receives a high score because restoration is described concretely. The examples use -y without explaining which changes require individual confirmation. Dependency security evidence is limited to a single runtime dependency and CI import checks, with no vulnerability audit shown. The author and repository are identifiable, but publisher identity is unknown per the prompt and no clear maintenance commitment is supplied.
README, package.json, and test files do not agree fully on release state: the README shows v5.0.0-rc.1, the package is rc.2, and tests refer to v5.8. The README describes multi-platform CI, uninstall smoke checks, and contract validation, but these source excerpts do not establish that they ran successfully. Dependency ranges and the Node requirement are explicit. Failure handling is only partly evidenced: persona validation falls back to a neutral persona and malformed Claude configuration is not overwritten, while general install failure messages are not shown in enough detail.
The project clearly targets Claude Code, Codex CLI, Gemini CLI, and OpenClaw, and lists security, development, office, and hardware scenarios. The Persona Voice Card constrains personas to voice fields, and the README separates code-graph capabilities into the external abyss CLI; these boundaries are explained, though skill-trigger rules and out-of-scope cases remain limited. Target-specific commands, artifacts, and CI smoke-check definitions provide concrete environment-fit evidence.
The README is organized around features, installation, skills, the spec, and contribution; installation, upgrades, uninstall, and examples are well covered. The MIT license text is complete. Version notes are extensive, but conflicting version markers in the supplied materials reduce naming stability. The README identifies limits such as the offline core persona, remote persona caching, a separate CLI, and the cost of optional evaluations; FAQs are not prominent. Contribution routes are clear, but maintenance ownership or response expectations are not stated.
The product combines personas, output styles, skills, and cross-CLI configuration to support reusable customized workflows. The structured persona card, installer, and uninstall restoration improve usability. The claimed value is supported mainly by the README; this static review has no execution evidence for realized outcomes. The large skill library and external CLI dependency also add configuration and upkeep costs, so the score is moderate.
The README names files and interfaces for major features, the external CLI split, the persona spec, and install commands. package.json, CI configuration, and test excerpts corroborate some contracts and regression guards. Most capability, precision, and scale figures in the README remain project claims without independent corroboration in the supplied material. CI and test files show defined verification procedures only; this review does not claim they were executed.
- Version markers differ across the README, package metadata, and test material; confirm the intended release and matching documentation before following install or migration guidance.
- Remote personas are fetched from and cached from GitHub, while code-graph features require the separately installed abyss CLI; check network access and external components before deployment.
- CI and tests are presented only as repository-defined checks; they were not executed as part of this static review.
FAQ
Which coding agents does it support?
Does installation require network access?
npx. Personas other than the bundled abyss persona are fetched from GitHub on first use and cached.Is code graph analysis included?
indexing-code documents its calling convention.