agnix CLI
agnix lints CLAUDE.md, AGENTS.md, SKILL.md, hooks and MCP configs with 702 rules, autofix and editor extensions for VS Code and JetBrains
- 457 ★ on the repository
- Last commit
- Apache-2.0
Minimal access
Why this level
- The --fix-unsafe autofix tier rewrites agent config files on disk with no separate confirmation per edit
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/agnixDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The --fix-unsafe autofix tier rewrites agent config files on disk with no separate confirmation per edit
This is third-party code. Review the repository files before installing.
What it does
agnix checks agent configuration files: CLAUDE.md, AGENTS.md, SKILL.md, hook settings and MCP server configs for Claude Code, Codex CLI, OpenCode, Cursor, Copilot and others. Running npx agnix . walks the project and prints findings as path:line:col with an error or warning level, a reason and a fixable tag where an autofix exists; the --fix, --fix-safe and --fix-unsafe flags apply fixes at different risk levels. The tool installs through npm, Homebrew, pip with a no-install uvx agnix . run, or Cargo for the Rust build its core is written in. The project's own example flags a skill named Review-Code as an error because the format requires lowercase letters and hyphens only, with a ready fix to code-review. There is a separate GitHub Action package that fails CI on found errors, plus editor extensions for VS Code, JetBrains, Neovim and Zed that surface findings right inside the file. The rule set, 702 entries as of the README, is built from official agent specs, academic research and real breakage cases, and the developer's own blog post cites Vercel research showing a skill with the wrong syntax can go uncalled by the model entirely. The agent-sh.github.io/agnix/playground browser page lets anyone paste a config and see diagnostics with no install.
Who it is for. For developers who maintain CLAUDE.md, AGENTS.md and several SKILL.md files in one repository and want format mistakes caught before an agent silently ignores the file.
Good fit when
- A project has several SKILL.md files and needs a check that names and descriptions match the format an agent will actually pick up
- CLAUDE.md has grown generic lines like "be helpful and accurate," and those need catching automatically for removal
- A project supports Claude Code, Cursor and Copilot at once, and their configs should pass one shared rule set in CI
- You want agnix findings highlighted right in the editor: extensions exist for VS Code, JetBrains, Neovim and Zed
- You want to check one config quickly with no install: paste it into agnix's browser playground and see the diagnostics
Not a fit when
- You need the instructions checked for meaning, not file format: agnix catches syntax and structure, not whether a rule you wrote makes sense
- The project's only agent is nonstandard and outside the list of supported formats: a rule for it simply will not fire
- There is no Node.js, Python, Homebrew or Cargo on the machine and none of those install paths work: the README describes no other way to get the binary
Example request
Check every CLAUDE.md, AGENTS.md and SKILL.md in this repository with agnix and apply the safe autofixesLimitations
Rules cover the formats the project has already described: CLAUDE.md, AGENTS.md, SKILL.md, hooks and MCP configs for Claude Code, Codex CLI, OpenCode, Cursor, Copilot and the rest listed in the README, not an arbitrary agent file. Autofix splits into --fix-safe and --fix-unsafe tiers, meaning the developer treats some fixes as risky enough to leave off by default. The GitHub Action and the JetBrains and VS Code editor extensions ship separately from the main package and need their own install step.
How to disable. Remove the agnix dependency: npm uninstall -g agnix, brew uninstall agnix or pip uninstall agnix depending on how it was installed. The GitHub Action step comes out of the workflow file, and the editor extension is disabled or removed through that editor's own extension manager.
FAQ
How do I run agnix for Claude Code without installing it?
Run npx agnix . at the project root. It checks CLAUDE.md, AGENTS.md, SKILL.md and the other supported configs and prints findings at error or warning level.
How do I apply autofixes?
The --fix or --fix-safe flag applies safe edits; --fix-unsafe adds riskier ones that the developer leaves off by default.
Why isn't my SKILL.md firing for the agent even though the file exists?
A common cause is a bad name or description format. agnix's own example flags a skill named Review-Code as an error because the format needs lowercase letters and hyphens, suggesting the fix code-review.
Can configs be checked in CI?
Yes, the project ships a separate GitHub Action package that fails the build on errors found in CLAUDE.md, AGENTS.md, SKILL.md and other agent configs.
Can I try agnix with no install at all?
Yes, the agent-sh.github.io/agnix/playground page lets you paste a config straight into the browser and see diagnostics with no npx and no local install.
Discussion
No questions or reviews yet.
Related
CLI toolsThe author's personal set: shared agent rules and dozens of skills for GitHub, the browser, macOS, messaging and releases, plus a sync script
agr is a package manager for agent skills: installs them from any Git repo into Claude Code, Cursor and Codex by a versioned manifest, like npm or uv
A Claude Code and Codex plugin with skills for code research, tests, refactoring and independent review of finished changes
A CLI that generates and maintains AI context files, CLAUDE.md, Cursor rules, AGENTS.md and Copilot instructions, keeping them accurate as code changes
A Rust SDK for coding agents and the ready-made abstract CLI: REPL, 34 tools, MCP, graph memory, Anthropic and OpenAI, JSON output for CI
CodeGraph (Rust)
codegraph-rust
A Rust code graph for agents: four MCP tools show what an edit will touch, how the architecture fits together and where the fragile spots are
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| agnix (this entry) | yes | 457 | |
| MDA: one source for SKILL.md and AGENTS.md | no | 621 | the mda CLI compiles one .mda file into ready SKILL.md |
| ClaudeForge | no | 430 | a Claude Code plugin that generates and maintains CLAUDE.md |
README in short
The README opens with a statistic that 66 percent of developers call an "almost right" result their biggest AI frustration, and links to Vercel research on skills a model never invokes because of a syntax mistake. It continues with a sample finding for CLAUDE.md and SKILL.md, an install-method table covering npm, Homebrew, pip and Cargo, a list of editor extensions and a ready GitHub Action step.