Codex CLI · OpenCode

Security testing from Codex or OpenCode, without uploading your repository.

One command installs a portable Agent Skill into whichever agent you use. The assessment runs on your machine, proves what it can against your running application, and writes reports you keep. No slash command needed, and no second model key.

npx myrqen@latest
Invocation
myrqen auto
Account
not required
Uploaded
nothing

Codex has its own security product. Read this first.

Codex Security launched in March 2026 and is free for ChatGPT Pro and Enterprise. It ingests a repository's full commit log, reasons about how the code changed over time, and can open remediation pull requests. Finding a vulnerability introduced two years ago and never noticed is something Myrqen does not attempt.

The trades are real in both directions.

Where it runs

Codex Security

OpenAI's service. Your repository is ingested.

Myrqen

Your machine. Nothing is uploaded but the report, and only if you ask.

What it reads

Codex Security

The full commit history.

Myrqen

The working tree in front of it.

How a finding is settled

Codex Security

Reasoning, then an automated fix pull request.

Myrqen

Exercised against your running application under two identities and an admin, with the outcome recorded as a verification state.

Maturity

Codex Security

Research preview.

Myrqen

Released, with published accuracy, including the bad number.

Languages

Codex Security

Broad.

Myrqen

JavaScript and TypeScript; Python for FastAPI and Flask.

Portable to another agent

Codex Security

No.

Myrqen

Yes. Thirteen agents, one skill bundle.

They are not substitutes. If your constraint is the repository must not leave the machine, or you want a finding demonstrated rather than argued, that is what this is for.

Install.

Nothing is written before the plan is printed and accepted. --dry-run prints it and stops, --agent codex narrows it to one agent, and myrqen unlink removes only Myrqen's own files.

install

$ npx myrqen@latest

Detected coding agents
  ● Codex CLI         Confirmed
  ● OpenCode          Confirmed

Ready to link Myrqen
Codex CLI · this project
  CREATE  .codex/skills/myrqen/SKILL.md
OpenCode · this project
  CREATE  .config/opencode/skills/myrqen/SKILL.md

Neither Codex nor OpenCode has a slash-command mechanism, so there is no /myrqen here and none is faked. The universal invocation is the documented path, and it behaves identically to the slash command elsewhere.

Start your app. Then ask the agent to run it.

The skill bundle tells the agent how to work the phases. The CLI owns the contract: it assigns identity, deduplicates, redacts, caps severity by policy, and decides which verification state a finding may carry. An agent cannot promote its own finding to verified.

from Codex or OpenCode

myrqen auto

Depth is a policy profile. It never changes what is authorized.

DepthWhat changes
autoResolves from the size and shape of the project. The right answer for almost everybody.
lowA fast pass over the routes and the obvious surface.
highThe default depth for a real review.
xhighAdds identity differentials and longer chains.
ultraEverything, and it will take its time.

Switching agents does not mean switching tools.

Thirteen integrations ship, each written in that vendor's own documented format. A security feature built into one agent is a reason not to change agent, and this market changes every few months. The skill bundle is the behavioural source of truth, and it is the same file everywhere.

  • Claude Codeskill
  • Codex CLIskill
  • OpenCodeskill
  • Kiroskill
  • Cursorrules file
  • Windsurfrules file
  • Clinerules file
  • Roo Coderules file
  • Continuerules file
  • Aiderconventions
  • GitHub Copilotinstructions
  • Gemini CLIGEMINI.md
  • AGENTS.mdshared file

Every agent, and the exact file written for each.

What it will not do.

  • Find everything. Recall is 0.250 on a corpus nobody had tuned against, and the same whether or not a finding is allowed anywhere in the range the case's author marked rather than within eight lines of the line they named. It is the static pass alone, scored one file at a time with no agent involved, against our own gate of 0.90. The method and every number.
  • Read Java, Ruby, Go, or a Python framework other than FastAPI and Flask.
  • Follow every path data can take. In JavaScript and TypeScript the static pass follows a value across modules only where a call resolves. Events, queues and dynamic dispatch are handed to the agent.
  • Anything destructive, at any depth. There is no opt-in flag.
  • Touch a host you did not name. No wildcard grants exist.
  • Scan production. It exercises the application running on your machine.

Common questions.

Does Myrqen work with Codex?
Yes. myrqen link installs the skill bundle into .codex/skills/myrqen/, and the assessment is driven with myrqen auto from Codex's shell. Codex has no slash-command mechanism, so the CLI invocation is the documented path there.
Does Myrqen work with OpenCode?
Yes, the same way: the bundle installs into .config/opencode/skills/myrqen/ and the assessment runs as myrqen auto. OpenCode also reads .claude/skills and .agents/skills, so one install can satisfy several agents, the installer deduplicates by destination path.
How is this different from Codex Security?
Codex Security ingests your repository and its full commit history into OpenAI's service and reasons about how the code changed over time, which is a genuinely different and useful angle. Myrqen runs entirely on your machine, uploads nothing, and exercises candidates against your running application. OpenAI's own material describes Codex Security as a research preview that is slower, compute-intensive, and not a replacement for static analysis where high recall is required.
Do I need a separate API key?
No. Myrqen has no model of its own and never asks for a model provider credential. The reasoning comes from the agent subscription you already pay for.
Which languages does it support?
JavaScript and TypeScript across Node, Bun, Deno and edge runtimes, and Python for FastAPI and Flask applications. Django and other Python frameworks are parsed, but their routes are not modelled, so data-flow checks do not start from them. Java, Ruby, Go, template languages and infrastructure definitions are not parsed at all.
Can I run it without any agent at all?
Yes. myrqen auto from a plain terminal runs the static pass, writes all four report formats, and prints the coverage notes. Without an agent nobody exercises the candidates against the running application, so every one stays needs_review, which the report says rather than implying they were proven.
How accurate is it?
Precision is 1.000 and the false-positive rate on deliberately correct code is 0.000 on holdout-3 — but only after the fixes its blind run prompted, which burned it as a measurement. On holdout-7, a later corpus scored blind, the false-positive rate on deliberately correct code was 0.125 and 3 of four gates failed. Precision itself is not computable on it: precision is derived from exact class-name matches, and thirty independent authors named the defects in their own vocabulary. The safe-case false-positive rate is the precision signal that survives. Recall is 0.250 on a corpus nobody had tuned against, and the same whether or not a finding is allowed anywhere in the range the case's author marked rather than within eight lines of the line they named. It is the static pass alone, scored one file at a time with no agent involved, against our own gate of 0.90. A report with no findings is not evidence that an application is secure.
Is Myrqen free, and is it open source?
Local scanning, local reports in all four formats, and every finding at every severity are free and unlimited, with no account. It is not open source: the source is published under a source-available licence that grants reading and running, not redistribution.