@openai/codex-security

@openai/codex-security

Open-source TypeScript SDK and CLI for running Codex Security scans. The ESM-only package includes TypeScript declarations, the codex-security executable, and the matching Codex runtime.

[!NOTE] This package follows semantic versioning. Its public API may change between minor versions before 1.0.0.

Install

npm install @openai/codex-security
npx @openai/codex-security --version

The package supports macOS, Linux, and Windows and requires Node.js 22 or later. Scanning and exporting findings also require Python 3.10 or later. If you use Python 3.10, install the tomli package. Select another interpreter with --python, pythonPath, or PYTHON when needed.

When a newer version is available, the CLI shows the update command for your installation method. Set CODEX_SECURITY_NO_UPDATE_NOTICE=1 to hide the notice. Notices are also disabled in CI and when stderr is not a terminal.

Run a scan from TypeScript

Sign in with npx @openai/codex-security login or set OPENAI_API_KEY or CODEX_API_KEY. Then create a client and scan a repository you own or have permission to assess:

import { CodexSecurity } from "@openai/codex-security";

const security = new CodexSecurity();

try {
  const result = await security.run("/path/to/repository", {
    outputDir: "/path/outside/repository/results",
  });

  console.log(result.reportPath);
  console.log(result.findings.findings.length);
} finally {
  await security.close();
}

The SDK supports repository, path, committed-diff, and working-tree targets. Use security.preflight() to validate local inputs, onWorkerStatus and onReconnect to observe long-running scans, and an AbortSignal to cancel a scan.

Results can contain source excerpts, vulnerability details, and reproduction steps. Keep result directories and saved reports outside the repository and limit access to authorized reviewers.

Authentication

For local use, sign in with ChatGPT:

npx @openai/codex-security login
npx @openai/codex-security scan .

On a remote or headless machine, use device authentication:

npx @openai/codex-security login --device-auth

For CI, set OPENAI_API_KEY or CODEX_API_KEY. To store an API key instead, pass it on stdin:

printenv OPENAI_API_KEY | npx @openai/codex-security login --with-api-key

On Windows, set the API key in PowerShell:

$env:OPENAI_API_KEY = "<your-api-key>"
npx @openai/codex-security scan C:\code\repository

Check or remove the stored sign-in with npx @openai/codex-security login status and npx @openai/codex-security logout. Codex Security reuses an existing file-based Codex sign-in. If Codex stores credentials in the system keyring, run npx @openai/codex-security login once before scanning.

An environment API key takes precedence over a stored sign-in by default. When both a stored ChatGPT sign-in and an environment API key are available, an interactive scan asks which credential to use. JSON output, dry runs, CI, and other noninteractive scans never prompt and retain automatic API-key precedence. Select the credential source explicitly with --auth:

npx @openai/codex-security scan . --auth chatgpt
npx @openai/codex-security scan . --auth api-key

--auth chatgpt uses the stored sign-in and ignores OPENAI_API_KEY and CODEX_API_KEY. --auth api-key requires one of those environment variables. Omit --auth, or pass --auth auto, to preserve automatic API-key precedence for existing CI and unattended scans. The SDK accepts the same selection as security.run(repository, { auth: "chatgpt" }) and security.preflight(repository, { auth: "chatgpt" }).

To make the stored ChatGPT sign-in the automatic default instead, unset any configured API-key variables:

unset OPENAI_API_KEY CODEX_API_KEY

The interactive choice applies only to the current scan and is not persisted.

When an environment key is configured, ChatGPT login and codex-security login status identify the effective scan credential source without printing its value, including when no stored sign-in exists.

CLI

npx @openai/codex-security scan /path/to/repository
npx @openai/codex-security scan /path/to/repository --model gpt-5.6-terra
npx @openai/codex-security scan /path/to/repository --model gpt-5.6-terra --effort high
npx @openai/codex-security scan /path/to/repository --path src --path tests
npx @openai/codex-security scan /path/to/repository --knowledge-base /path/to/threat-models --knowledge-base /path/to/architecture.pdf
npx @openai/codex-security scan /path/to/repository --diff origin/main --json
npx @openai/codex-security scan /path/to/repository --output-dir /path/outside/repository/results
npx @openai/codex-security scan /path/to/repository --output-dir /path/outside/repository/results --archive-existing
npx @openai/codex-security scan /path/to/repository --dry-run
npx @openai/codex-security scan /path/to/repository --fail-on-severity high
npx @openai/codex-security scan /path/to/repository --max-cost 5
npx @openai/codex-security install-hook
npx @openai/codex-security bulk-scan
npx @openai/codex-security bulk-scan --model gpt-5.6-terra --effort high
npx @openai/codex-security bulk-scan repositories.csv --output-dir /path/outside/repositories/security-scans --workers 4
npx @openai/codex-security scans list /path/to/repository
npx @openai/codex-security scans list --scan-root /path/outside/repository/results
npx @openai/codex-security scans show SCAN_ID
npx @openai/codex-security scans rerun SCAN_ID
npx @openai/codex-security scans match PREVIOUS_SCAN_ID CURRENT_SCAN_ID
npx @openai/codex-security scans match --all
npx @openai/codex-security scans compare PREVIOUS_SCAN_ID CURRENT_SCAN_ID
npx @openai/codex-security findings false-positive OCCURRENCE_ID --reason "The route already checks permissions"
npx @openai/codex-security export /path/outside/repository/results --export-format sarif --output /path/outside/repository/results.sarif
npx @openai/codex-security export /path/outside/repository/results --export-format csv --output /path/outside/repository/findings.csv
npx @openai/codex-security export /path/outside/repository/results --export-format json --output /path/outside/repository/findings.json
npx @openai/codex-security validate /path/outside/repository/findings.json "Possible SQL injection in src/query.ts:42"
npx @openai/codex-security validate "Possible SQL injection" --effort high
npx @openai/codex-security patch /path/outside/repository/findings.json "Missing authorization check in src/routes.ts:18"
npx @openai/codex-security patch "Missing authorization check" --effort high

Run npx @openai/codex-security --version for the installed CLI version or npx @openai/codex-security info --json for the package, bundled plugin, Codex runtime, default model, reasoning effort, and first-scan command. A scan with --dry-run also reports its effective model and reasoning effort, including --codex overrides, without starting Codex or contacting the network.

install-hook scans staged and unstaged changes before each commit. It respects core.hooksPath, does not replace an existing hook, and blocks high-severity findings or failed scans. Set --fail-on-severity to change the threshold.

--path scopes a scan to one or more paths, --diff scans committed changes, and --working-tree scans staged and unstaged changes. Deep scans support repository and path targets. The output directory must be outside the scanned directory and any enclosing Git worktree. When SARIF is produced, it is written to <scan-dir>/exports/results.sarif.

Repeat --knowledge-base PATH for multiple files or directories. Directories are searched recursively for Markdown, text, PDF, and Word (.docx) files.

On macOS/Linux, an existing output directory must be private to the current user (chmod 700).

If the output directory already contains results, add --archive-existing. The CLI moves them to <output-dir>.previous-<timestamp>-<id> and starts the scan in a new, empty directory at the original path. Add --dry-run to see the destination without moving files.

Scans are report-only by default. Use --fail-on-severity in CI to exit 1 when a completed scan contains a finding at or above the selected severity. Incomplete coverage and CLI/runtime errors exit 2 so they cannot be mistaken for a passing policy. Incomplete scans still write the available human or JSON result to stdout and a coverage warning to stderr, including in report-only mode.

Scans use gpt-5.6-sol with extra-high reasoning effort by default. OpenAI is the implied provider. Use --model gpt-5.6-terra to switch models and --effort minimal|low|medium|high|xhigh to set reasoning effort. Repeat --codex KEY=VALUE for other Codex settings; existing --codex 'model_reasoning_effort="high"' overrides remain supported.

Scan progress identifies the requested paths and reports actual ranking, file-review, validation, and attack-path phases as they become available. Completion summarizes findings, severity, coverage, elapsed time, available token and worker counts, estimated cost, the results directory, and the next useful command. Progress and summaries use stderr; structured scan results remain on stdout.

Each scan records its model, tokens, and estimated cost in its JSON result, scan history, and bulk-scan receipt. Estimates use standard API token prices, including cached input and cache writes; fees and surcharges are not included.

Use --max-cost USD to stop a scan, including its delegated workers, when its running cost exceeds the limit. Partial results are preserved. Requests already in progress can finish above the limit.

Run npx @openai/codex-security scan --help or npx @openai/codex-security bulk-scan --help for the complete CLI references.

Sign in with gh auth login, then run npx @openai/codex-security bulk-scan to discover GitHub repositories pushed in the last 90 days. Archived repositories and forks are excluded. Search the repository list, select the repositories to scan, and confirm before scanning. Private checkouts reuse your GitHub CLI sign-in without changing your global Git configuration. The selected repositories are saved to <output-dir>/repositories.csv for review or resumption.

To use an existing repository list or run in CI, pass a CSV with required id, repository, and revision columns. Revisions must be full commit hashes; optional scope and mode columns narrow individual scans:

id,repository,revision,scope,mode
service,https://github.com/acme/service.git,0123456789abcdef0123456789abcdef01234567,src,standard

--workers limits concurrent scans and --max-attempts retries failures. Results remain under --output-dir; rerun the same command to resume.

Scan history and reruns

npx @openai/codex-security scans list lists scans for the current repository. Pass a repository path to inspect another checkout, --scan-root DIR to list scans whose artifacts are under a particular root. scans show SCAN_ID includes the scan configuration, results, coverage, and artifact locations.

Every scan history command accepts a full scan ID or a unique prefix of at least eight characters.

Scan history uses the existing Codex Security workbench database at $CODEX_HOME/state/plugins/codex-security/workbench.sqlite3. Set CODEX_SECURITY_STATE_DIR to place the database elsewhere. Scan credentials are never stored in the scan configuration.

The scan sandbox permits writes to the selected state directory so SQLite can maintain its database and journal files. If the host itself cannot write to the default directory, select a writable directory outside the scanned repository:

export CODEX_SECURITY_STATE_DIR=/path/to/writable/codex-security-state

Use findings false-positive OCCURRENCE_ID --reason TEXT to mark a finding as a false positive and explain why. Later scans dismiss a matching finding only when the same reason still applies.

scans rerun SCAN_ID repeats the original configuration against the current checkout so a fixed vulnerability can be checked again.

scans match BEFORE_SCAN_ID AFTER_SCAN_ID links findings with the same root cause; scans match --all matches all completed scans of the current repository, including other worktrees and clones. Saved matches appear in scans show and are reused unless --force is passed. Scans without sealed artifacts are skipped.

scans compare BEFORE_SCAN_ID AFTER_SCAN_ID reads saved matches and reports findings as new, persisting, reopened, resolved, or unknown. Missing findings are not treated as resolved when the later scan is incomplete or does not cover their original scope.

The CLI uses Incur for agent-friendly discovery and structured output. Inspect the command manifest with --llms, inspect a command schema with scan --schema --format json, register the CLI as an MCP server with mcp add, sync agent skills with skills add, or generate shell completions with completions bash|zsh|fish. Scan results support --format toon|json|yaml|jsonl and --full-output. Use info --json for SDK and bundled-plugin metadata. MCP exposes only this read-only metadata command; scans, bulk repository scans, authentication, exports, validation, and patching remain CLI-only because the MCP transport cannot cancel active scans.

For CI, save machine-readable output outside the checked-out repository and apply a severity policy. Incomplete coverage and runtime errors still exit nonzero:

SCAN_ROOT="$(mktemp -d)"
npx @openai/codex-security scan . \
  --diff origin/main \
  --output-dir "$SCAN_ROOT/results" \
  --json \
  --fail-on-severity high > "$SCAN_ROOT/findings.json"

JSON scans never use interactive terminal controls, even when stderr is a TTY. The validate, patch, login, and logout commands reject --json because they do not produce structured CLI output. Sign-in commands remain interactive. CSV exports cannot be written to stdout while JSON output is requested.

Use export to create CSV, JSON, or SARIF from a completed, sealed scan without starting Codex or loading credentials. JSON preserves the sealed findings document. CSV uses the portable findings columns, marks findings as open, and does not include local workbench triage state. The exporter validates the seal before writing, accepts --output - for stdout, and can use --source-root /path/to/repository with SARIF to add source-line fingerprints. Run npx @openai/codex-security export --help for all export options.

Use validate to run the bundled validation skill on candidate findings and patch to run the bundled fix-finding skill on security issues. Each positional input can be either a file, whose contents are read into the request, or literal text. Both commands operate on the current directory, use the scan model and reasoning defaults, ignore unrelated user configuration and plugins, and print the final response without the underlying Codex event stream. Override the model with --codex 'model="gpt-5.6-sol"' and the reasoning effort with --effort high or --codex 'model_reasoning_effort="high"'. Inputs are limited to 64 items and 1 MiB total.

Canonical scan documents are limited to 16 MiB for the manifest, 128 MiB for findings, and 32 MiB for coverage. Oversized scans are rejected before sealing.

Exit codes are 0 for a completed report-only scan or a passing policy, 1 for a completed policy violation, 2 for invalid input, incomplete coverage, or a runtime/export error, 130 for interruption, and 143 for termination.

Use --dry-run or await security.preflight(...) to validate the repository, target, mode, output location, and Codex overrides without initializing the runtime or loading credentials. Dry runs do not inspect the plugin or probe its Python interpreter. The preflight result includes the selected authentication method and, for an environment API key, its variable name. Authentication and model access remain unverified until a real scan starts.

Scan progress identifies the selected credential source before Codex starts. Terminals and noninteractive CI logs also show how to retry with --auth chatgpt when an environment API key overrides the stored sign-in. Progress remains on stderr so JSON output stays machine readable. Network failures and rate limits remain retryable; definitive authentication and model authorization failures stop immediately.

Documentation and security

• CLI quickstart
• TypeScript SDK guide
• GitHub issues for bugs and feature requests
• Security policy for private vulnerability reporting and safe operation