praxen

Installation

Praxen ships as a portable agent skill, packaged for both Claude Code and OpenAI Codex. Both platforms load the same skills/behavior-verifier engine and produce the same JSON / HTML / TXT report format — only the install/packaging differs. On the same inputs, findings should cover the same major themes, but exact counts, grouping, and RAISE maturity scores can vary by model and context (see Understanding Run-to-Run Variability). Claude Code is the most common path; Codex is supported as a first-class agent-skills platform.

Prerequisites

• A coding agent capable of tool use and multi-step instruction-following. Praxen is tested against Claude Code and OpenAI Codex; other coding agents that can read a skill markdown file and call tools (Read, Grep, Glob, Bash, Write) should also work.
• Python 3.9 or newer on the PATH — for the bundled report renderer (render.py), which is standard-library only (nothing to pip install). You almost certainly already have it: 3.9 ships as the macOS Command Line Tools system Python, and on Windows py -3 works. (If python3 isn’t found, the renderer falls back to python.)
• Network access for your coding agent’s LLM provider during analysis. Praxen itself does not phone home, but the LLM calls your coding agent makes during analysis follow whatever provider configuration the agent uses.
• Workspace write permission for the agent. Praxen writes its report files to a ./reports/ directory and runs two bundled Python scripts during a scan, so the agent must be allowed to write within the working directory. Claude Code allows this by default; agents that sandbox file writes need that permission granted (per your agent’s own docs).

That’s the entire dependency surface.

Claude Code

Install from the plugin marketplace. From your terminal:

claude plugin marketplace add open-agent-ai-security/praxen
claude plugin install praxen@open-agent-ai-security
claude plugin list      # confirm: praxen@open-agent-ai-security, enabled, v1.0.0+

The skill registers as behavior-verifier. The in-session equivalents — /plugin marketplace add …, /plugin install …, /plugin list — do exactly the same thing; if you install from within a Claude Code session, run /reload-plugins (or restart) to activate the skill. Prefer the terminal form when scripting: claude plugin … is argument-driven and runs the same way on every interface, whereas in-session slash commands occasionally fall through and get sent as ordinary chat messages.

OpenAI Codex

Codex has its own plugin marketplace, and Praxen installs from the same repo as the Claude Code path. From your terminal:

codex plugin marketplace add open-agent-ai-security/praxen
codex plugin add praxen@open-agent-ai-security
codex plugin list      # confirm: praxen@open-agent-ai-security, installed, enabled, v1.0.0+

This installs and enables the plugin in Codex’s local config; the bundled behavior-verifier skill is then available to every Codex session. Running an analysis is the same as on any agent — see Usage.

Any other agent

No marketplace, no download step — Praxen is just a skill folder in a public repo, so any capable coding agent can fetch and run it from a plain-English instruction. (This also works on Claude Code or Codex if you’d rather skip the marketplace.) In your agent session, say something like:

Clone https://github.com/open-agent-ai-security/praxen and follow its behavior-verifier skill to run a Praxen behavior analysis on [your target]. Use the Worker Remit at [path].

The agent brings down its own copy and runs the skill from skills/behavior-verifier/. Offline? Download the release .zip from the releases page and point your agent at the unzipped folder instead. See Usage for the analysis instructions.

Updating

Every release bumps the version, so a refresh always picks up the latest. Check what you have with claude plugin list.

Claude Code — refresh the catalog, update, then restart (/reload-plugins or relaunch):

claude plugin marketplace update open-agent-ai-security   # refresh the catalog
claude plugin update praxen@open-agent-ai-security         # install the latest

Both steps matter: without the first, plugin update only sees your local (possibly stale) catalog cache.

OpenAI Codex — the same two-step shape:

codex plugin marketplace upgrade open-agent-ai-security
codex plugin add praxen@open-agent-ai-security

Any other agent — git pull the clone (or re-clone), or download a newer release .zip. Praxen is stateless across analyses, so there’s no migration step.

Praxen is a security tool — staying current matters. Update on a regular cadence, or turn on auto-update (below).

Auto-update and fleet config (Claude Code)

Auto-update is per-marketplace and off by default for third-party marketplaces, and Claude Code won’t notify you when a new version exists. To enable it interactively: /plugin → Marketplaces → select open-agent-ai-security → enable auto-update (it then checks at startup). Admins can set it fleet-wide in managed settings.json:

{
  "extraKnownMarketplaces": {
    "open-agent-ai-security": {
      "source": { "source": "github", "repo": "open-agent-ai-security/praxen" },
      "autoUpdate": true
    }
  }
}

(DISABLE_AUTOUPDATER=1 turns off Claude Code’s auto-updates globally; see Claude Code’s settings docs for its exact scope.)

Uninstalling

Claude Code (plugin marketplace):

claude plugin uninstall praxen@open-agent-ai-security
claude plugin marketplace remove open-agent-ai-security

The marketplace is removed by its registered name (open-agent-ai-security, from .claude-plugin/marketplace.json) — which here matches the repo owner used to add it.

OpenAI Codex (marketplace):

codex plugin remove praxen@open-agent-ai-security
codex plugin marketplace remove open-agent-ai-security

Any other agent: delete the cloned (or unzipped) folder. No system state is left behind.

Next steps

• Quickstart — first end-to-end report: have Claude author a remit for the FinBot demo agent, scan it, and read the report
• Writing Worker Remits — authoring the policy document Praxen verifies against
• Usage — the full running-an-analysis reference

This site is open source. Improve this page.