OWASP Codex Skill

A reusable Codex skill for application security reviews based on the OWASP Top 10.

This skill helps Codex act as an independent, read-only security-review track for codebases, pull requests, APIs, configuration, and dependency changes. It uses OWASP Top 10 as the taxonomy while requiring concrete code evidence, affected flows, impact, and remediation for every confirmed finding.

Optional open-source scanners can complement the review, but the skill never installs or runs them implicitly. Source uploads, remote rules, database downloads, network access, telemetry, and live-target scanning require explicit approval.

What This Skill Does

owasp-codex-skill guides Codex through security reviews focused on:

• Authentication and session handling
• Authorization, access control, IDOR, BOLA, and tenant isolation
• Input handling and injection risks
• Cryptographic failures and secret handling
• Security misconfiguration
• Supply chain and dependency risks
• Software and data integrity failures
• Logging, alerting, and audit gaps
• Failure-path and exceptional-condition handling
• Insecure design and business logic weaknesses

The skill is designed for code review. It does not run intrusive tests, exploit live systems, brute force credentials, exfiltrate secrets, or perform active scanning unless a user explicitly authorizes that work in a separate task.

OWASP Version

The default review frame is OWASP Top 10:2025.

The skill also includes a high-level mapping for OWASP Top 10:2021 so reviewers can adapt findings when a project, client, or organization still requires the 2021 taxonomy.

Official OWASP Top 10 project:

https://owasp.org/www-project-top-ten/

OWASP Top 10:2025 release page:

https://owasp.org/Top10/2025/

Repository Structure

owasp-codex-skill/
├── bin/
│   └── owasp-codex-skill.js
├── package.json
├── SKILL.md
├── agents/
│   └── openai.yaml
├── test/
│   └── cli.test.js
└── references/
    ├── open-source-tooling.md
    ├── owasp-top-10-review-map.md
    └── report-template.md

SKILL.md

The main Codex skill file. It contains the trigger metadata and the core review workflow:

• Establish review scope
• Trace security-sensitive flows
• Map observations to OWASP categories
• Verify issues before reporting
• Report findings with evidence and remediation

agents/openai.yaml

UI-facing metadata for environments that display skill names, descriptions, and default prompts.

bin/owasp-codex-skill.js

The npm CLI used to install the skill into your Codex skills directory.

references/owasp-top-10-review-map.md

A compact review map for OWASP Top 10:2025 categories, including:

• What to look for
• Relevant code areas
• Common review targets
• Remediation cues
• Legacy OWASP Top 10:2021 mapping

references/report-template.md

A reusable security review report structure for final output, including:

• Findings ordered by severity
• OWASP category
• Evidence
• Affected flow
• Impact
• Exploit scenario
• Remediation
• Confidence
• Coverage and residual risk

references/open-source-tooling.md

The local-first policy and allowlist for optional OpenGrep, Gitleaks, OSV-Scanner, Trivy, Checkov, and Syft integrations. Scanner output remains unverified until Codex traces and confirms the affected flow.

Installation

Install the package globally:

npm install -g owasp-codex-skill

Then install the Codex skill:

owasp-codex-skill install

By default, the installer copies the skill into:

~/.codex/skills/owasp-codex-skill

If CODEX_HOME is set, the installer uses:

$CODEX_HOME/skills/owasp-codex-skill

To replace an existing installation:

owasp-codex-skill install --force

To install into a custom directory:

owasp-codex-skill install --target /path/to/skills/owasp-codex-skill

You can print the default installation path with:

owasp-codex-skill path

After installation, restart or reload your Codex session if your environment does not automatically discover newly added skills.

Note: this command works after the package is published to npm.

Install from Git

mkdir -p ~/.codex/skills
git clone https://github.com/LirielC/owasp-codex-skill.git ~/.codex/skills/owasp-codex-skill

If you use a custom CODEX_HOME, install it under that location instead:

mkdir -p "$CODEX_HOME/skills"
git clone https://github.com/LirielC/owasp-codex-skill.git "$CODEX_HOME/skills/owasp-codex-skill"

Usage

Ask Codex to use the skill explicitly:

Use $owasp-codex-skill to review this codebase for OWASP Top 10 security risks.

Other useful prompts:

Use $owasp-codex-skill to review this pull request for authentication, authorization, and injection risks.

Use $owasp-codex-skill to audit the API routes and produce a security findings report with severity, evidence, and remediation.

Use $owasp-codex-skill to review this service against OWASP Top 10:2021 instead of 2025.

Expected Output

The skill is designed to produce concise, evidence-backed findings. A typical finding should include:

• Severity
• OWASP category
• File and line evidence
• Affected flow
• Impact
• Exploit scenario
• Remediation
• Confidence level

If no confirmed issues are found, Codex should state that clearly and include the review scope plus residual risks.

Review Philosophy

This skill prioritizes exploitability over pattern matching.

A dangerous API, framework setting, or dependency is not automatically a vulnerability. A valid finding should explain how attacker-controlled input, missing authorization, unsafe configuration, or another concrete condition reaches a security-sensitive outcome.

The skill also encourages checking for compensating controls before reporting an issue, including:

• Shared middleware
• Authorization policies
• Framework-level validation
• Database constraints
• Infrastructure restrictions
• Existing tests

Severity Model

The skill uses a pragmatic severity model:

• Critical: unauthenticated remote compromise, broad data exposure, credential or session takeover, supply-chain execution, or production secret exposure
• High: privilege escalation, cross-user or cross-tenant access, meaningful injection impact, weak reset or session design, or SSRF to sensitive internal resources
• Medium: constrained security bypass, limited sensitive information disclosure, unsafe defaults, or missing auditability for important actions
• Low: hardening gaps, defense-in-depth improvements, low-impact leakage, or incomplete validation with effective downstream controls

Severity should be adjusted based on deployment context, attacker role, data sensitivity, and available runtime evidence.

Scope and Limitations

This skill is not a replacement for:

• A full penetration test
• Dynamic application security testing
• Manual threat modeling with system owners
• Production infrastructure review
• Dependency advisory tooling
• Legal or compliance advice

It is a reusable review workflow for Codex. The quality of results depends on repository access, runtime context, available configuration, test coverage, and the specificity of the user request.

Development

Test the npm CLI locally:

npm test

Check the packaged skill files without installing or executing scanners:

npm run validate

Install the local package globally during development:

npm install -g .
owasp-codex-skill install --force

Check the package contents before publishing:

npm pack --dry-run

Publish to npm:

npm login
npm publish --access public

Validate the skill structure with the Codex skill validator (adjust the path for your Codex installation):

python ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py /path/to/owasp-codex-skill

Contributing

Contributions should keep the skill focused, reusable, and concise.

Good contributions include:

• Better OWASP category review prompts
• Clearer remediation guidance
• Improved report structure
• More accurate mappings between OWASP versions
• Reduced ambiguity in the review workflow

Avoid adding broad documentation that Codex does not need while performing a review. The skill should remain compact enough to load quickly and specific enough to guide useful security analysis.

License

Apache License 2.0. Optional tools, rule packs, advisory databases, and generated artifacts retain their own licenses.