CLI reference

Usage

openaudit [target] [flags]          # same as: openaudit audit [target]

target is a directory (default .), a file, an http(s):// URL, image:<ref> for a container image, or cloud:<aws|azure|gcp|kubernetes> for a live cloud account (opt-in Prowler, using your ambient credentials). openaudit detects what the target contains (Node, Python, Docker, Terraform, Kubernetes, GitHub Actions, Android/iOS, ...) and runs the auditors that apply.

Flag	Meaning
`-f, --format <list>`	`terminal` (default), `json`, `sarif`, `markdown`, `html`. Repeatable or comma-separated. More than one format requires `--out`.
`-o, --out <dir>`	Write `openaudit-report.<ext>` for each format into `<dir>`.
`--json`	Agent mode: the JSON report on stdout and nothing else. No progress output.
`--fail-on <severity>`	Exit 1 if any finding is at least `critical`/`high`/`medium`/`low`/`info`. Default `none`.
`--only`, `--exclude <patterns>`	Choose auditors by id (`web.headers`), prefix (`web`) or glob (`code.*`).
`--category <list>`	Only run auditors in these categories, e.g. `security,accessibility`.
`--baseline <report.json>`	Suppress findings already in an earlier JSON report, so only new findings count.
`--diff <ref>`	Keep only findings in files changed since the git ref (`<ref>...HEAD`), e.g. `origin/main` on a pull request. Findings with no file are dropped. The report records the ref and the dropped count.
`--offline`	Make no network requests except to the target (skips OSV lookups and similar).
`--browser`, `--max-pages <n>`	Drive a real Chromium through Playwright for URL targets (needs `playwright-core` and a Chromium).
`-H, --header "Name: value"`	Extra header for the target origin only, e.g. an auth cookie for staging. Repeatable.
`--agentic`, `--model <id>`	Run the LLM-driven auditors. Needs `ANTHROPIC_API_KEY` or Claude Code auth.
`--ai`	Attach an AI-prioritised summary to the report. If this fails, openaudit prints a warning and continues.
`--config <path>`, `--no-config`	Use a trusted config file, or ignore the target's `openaudit.config.json`.
`--concurrency <n>`, `-q/--quiet`, `--verbose`

Other commands:

Command	Details
`openaudit list [--json] [--target-kind url]`	Auditors (id, engine, categories, targets, external tool, on/off by default) and playbooks
`openaudit doctor [--json]`	Bun version, external tools on PATH, Playwright/Chromium, AI credentials
`openaudit init [--force]`	Write `openaudit.config.json`
`openaudit schema [report\|config]`	JSON Schema (draft 2020-12) of the report or the config file
`openaudit mcp [--http]`	Start the MCP server on stdio, or over Streamable HTTP with `--http`

Exit codes

Code	Meaning
`0`	Audit completed and no finding reached `--fail-on`
`1`	At least one finding at or above `--fail-on`
`2`	Usage error, invalid config, or the target could not be resolved
`3`	`--fail-on` was set but an auditor errored or timed out, so the gate could not be fully checked

Exit code 3 fails closed, so a target cannot pass CI by crashing the auditor that would have flagged it. Pass --allow-errors to accept partial results. Skipped auditors (missing tool, offline) never change the exit code; their status and reason are recorded in the report.

Agent mode (`--json`)

openaudit . --json --fail-on high > report.json; echo $?
openaudit schema > report.schema.json   # validate the output

stdout carries exactly one JSON document, with keys in a stable order (schemaVersion, tool, target, summary, results[].findings[]). Nothing else is written to stdout, and no progress is shown on stderr.
Findings are sorted by severity. Each one has ruleId, severity, category, location (file/line or url/selector), evidence (redacted), remediation, references and a fingerprint.
To check a fix, re-run the audit with --baseline report.json --fail-on low. Exit code 0 means no new findings appeared.

Help output

Printed by the CLI itself, so it always matches the version the site was built with.

openaudit --help

DESCRIPTION
  Audit codebases, URLs, web apps, containers, IaC and mobile builds for security, accessibility, performance, SEO, privacy and more. Exit codes: 0 ok, 1 findings at/above --fail-on, 2 usage or target error.

USAGE
  openaudit <subcommand> [flags] [<target>]

ARGUMENTS
  target string    Directory, file, http(s) URL or image:<ref> (default: .) (optional)

FLAGS
  --format, -f string      Output format(s): terminal, json, sarif, markdown, html, csv. Repeat or comma-separate; several need --out
  --out, -o string         Write openaudit-report.<ext> per format into this directory
  --only string            Only these auditors: id, prefix or prefix.* (repeatable, comma list)
  --exclude string         Skip these auditors (same patterns as --only)
  --category string        Only auditors in these categories, e.g. security,accessibility
  --fail-on string         Exit 1 if any finding is at least this severe: critical, high, medium, low, info, none
  --offline                No network access except to the target itself
  --browser                Drive a real browser (Playwright) for URL targets
  --max-pages integer      Max pages to crawl for URL targets
  --header, -H string      Extra request header for the target origin, "Name: value" (repeatable)
  --agentic                Enable agentic (LLM) auditors; needs ANTHROPIC_API_KEY or Claude Code auth
  --ai                     Attach AI insights (prioritised summary) to the report
  --model string           Model id for agentic auditors and --ai
  --baseline string        Earlier JSON report; its findings are suppressed (only new findings count)
  --diff string            Only report findings in files changed since this git ref (e.g. origin/main), for pull requests
  --concurrency integer    Auditors run in parallel (default 4)
  --json                   Agent mode: JSON report on stdout, nothing else on stdout or stderr progress
  --quiet, -q              No progress or notices on stderr
  --verbose                Terminal output lists every finding with details
  --config string          Trusted config file. Without it, openaudit.config.json in the target is read with restrictions
  --no-config              Ignore openaudit.config.json in the target
  --allow-errors           With --fail-on, do not fail when an auditor errors or times out

GLOBAL FLAGS
  --help, -h                                          Show help information
  --version, -v                                       Show version information
  --completions <bash|zsh|fish|powershell|sh|pwsh>    Print shell completion script (choices: bash, zsh, fish, powershell, sh, pwsh)

SUBCOMMANDS
  audit     Audit a target (the default command)
  list      List auditors and playbooks
  doctor    Check Bun, external tools, browser and AI credentials
  init      Create openaudit.config.json
  mcp       Start the MCP server: stdio by default (Claude Code, Cursor, Codex, ...), or --http for remote agents
  schema    Print the JSON Schema of the report (or config) so agents can validate output
EXAMPLES
  # Audit the current directory
  openaudit

  # Audit a live site with a real browser
  openaudit https://example.com --browser --max-pages 20

  # Agent/CI mode: JSON on stdout, exit 1 on high+ findings
  openaudit . --json --fail-on high

  # Write reports/openaudit-report.sarif and .html
  openaudit . -f sarif,html -o reports

  # Audit a container image
  openaudit image:nginx:1.27 --only container

  # Fail only on findings that are new since the baseline
  openaudit . --baseline main-report.json --fail-on medium

  # Use openaudit from Claude Code over MCP
  claude mcp add openaudit -- bunx openaudit mcp

openaudit audit --help

DESCRIPTION
  Audit a target (the default command)

USAGE
  openaudit audit [flags] [<target>]

ARGUMENTS
  target string    Directory, file, http(s) URL or image:<ref> (default: .) (optional)

FLAGS
  --format, -f string      Output format(s): terminal, json, sarif, markdown, html, csv. Repeat or comma-separate; several need --out
  --out, -o string         Write openaudit-report.<ext> per format into this directory
  --only string            Only these auditors: id, prefix or prefix.* (repeatable, comma list)
  --exclude string         Skip these auditors (same patterns as --only)
  --category string        Only auditors in these categories, e.g. security,accessibility
  --fail-on string         Exit 1 if any finding is at least this severe: critical, high, medium, low, info, none
  --offline                No network access except to the target itself
  --browser                Drive a real browser (Playwright) for URL targets
  --max-pages integer      Max pages to crawl for URL targets
  --header, -H string      Extra request header for the target origin, "Name: value" (repeatable)
  --agentic                Enable agentic (LLM) auditors; needs ANTHROPIC_API_KEY or Claude Code auth
  --ai                     Attach AI insights (prioritised summary) to the report
  --model string           Model id for agentic auditors and --ai
  --baseline string        Earlier JSON report; its findings are suppressed (only new findings count)
  --diff string            Only report findings in files changed since this git ref (e.g. origin/main), for pull requests
  --concurrency integer    Auditors run in parallel (default 4)
  --json                   Agent mode: JSON report on stdout, nothing else on stdout or stderr progress
  --quiet, -q              No progress or notices on stderr
  --verbose                Terminal output lists every finding with details
  --config string          Trusted config file. Without it, openaudit.config.json in the target is read with restrictions
  --no-config              Ignore openaudit.config.json in the target
  --allow-errors           With --fail-on, do not fail when an auditor errors or times out

GLOBAL FLAGS
  --help, -h                                          Show help information
  --version, -v                                       Show version information
  --completions <bash|zsh|fish|powershell|sh|pwsh>    Print shell completion script (choices: bash, zsh, fish, powershell, sh, pwsh)

EXAMPLES
  # Audit the current directory
  openaudit

  # Audit a live site with a real browser
  openaudit https://example.com --browser --max-pages 20

  # Agent/CI mode: JSON on stdout, exit 1 on high+ findings
  openaudit . --json --fail-on high

  # Write reports/openaudit-report.sarif and .html
  openaudit . -f sarif,html -o reports

  # Audit a container image
  openaudit image:nginx:1.27 --only container

openaudit list --help

DESCRIPTION
  List auditors and playbooks

USAGE
  openaudit list [flags]

FLAGS
  --json                  Machine-readable output
  --target-kind string    Filter: codebase, url, webapp, container, iac, mobile

GLOBAL FLAGS
  --help, -h                                          Show help information
  --version, -v                                       Show version information
  --completions <bash|zsh|fish|powershell|sh|pwsh>    Print shell completion script (choices: bash, zsh, fish, powershell, sh, pwsh)

openaudit doctor --help

DESCRIPTION
  Check Bun, external tools, browser and AI credentials

USAGE
  openaudit doctor [flags]

FLAGS
  --json    Machine-readable output

GLOBAL FLAGS
  --help, -h                                          Show help information
  --version, -v                                       Show version information
  --completions <bash|zsh|fish|powershell|sh|pwsh>    Print shell completion script (choices: bash, zsh, fish, powershell, sh, pwsh)

openaudit init --help

DESCRIPTION
  Create openaudit.config.json

USAGE
  openaudit init [flags]

FLAGS
  --force         Overwrite an existing config
  --dir string    Directory to write into (default: .)

GLOBAL FLAGS
  --help, -h                                          Show help information
  --version, -v                                       Show version information
  --completions <bash|zsh|fish|powershell|sh|pwsh>    Print shell completion script (choices: bash, zsh, fish, powershell, sh, pwsh)

openaudit mcp --help

DESCRIPTION
  Start the MCP server: stdio by default (Claude Code, Cursor, Codex, ...), or --http for remote agents

USAGE
  openaudit mcp [flags]

FLAGS
  --http                     Serve MCP over Streamable HTTP instead of stdio (loopback by default)
  --host string              HTTP bind address (default 127.0.0.1)
  --port integer             HTTP port (default 7337)
  --token string             HTTP bearer token (or OPENAUDIT_MCP_TOKEN); generated on loopback if omitted, required off loopback
  --root string              HTTP: directory that local targets must resolve inside (repeatable)
  --allow-origin string      HTTP: browser Origin allowed to connect (repeatable)
  --allow-host string        HTTP: extra Host header value accepted (repeatable)
  --persist-reports          HTTP: save reports under <root>/.openaudit/reports (off by default over HTTP)
  --allow-cloud              HTTP: accept cloud:<provider> targets (uses the server's credentials)
  --allow-images             HTTP: accept image:<ref> targets on non-loopback binds (registry host is still checked)
  --allow-agentic            HTTP: let clients run agentic auditors (spends the server's ANTHROPIC_API_KEY)
  --allow-private-targets    HTTP: allow URL targets on private/loopback/metadata addresses (non-loopback binds only block them by default)

GLOBAL FLAGS
  --help, -h                                          Show help information
  --version, -v                                       Show version information
  --completions <bash|zsh|fish|powershell|sh|pwsh>    Print shell completion script (choices: bash, zsh, fish, powershell, sh, pwsh)

openaudit schema --help

DESCRIPTION
  Print the JSON Schema of the report (or config) so agents can validate output

USAGE
  openaudit schema [flags] [<name>]

ARGUMENTS
  name choice    Which schema (default: report) (optional)

GLOBAL FLAGS
  --help, -h                                          Show help information
  --version, -v                                       Show version information
  --completions <bash|zsh|fish|powershell|sh|pwsh>    Print shell completion script (choices: bash, zsh, fish, powershell, sh, pwsh)

Usage

Exit codes

Agent mode (--json)

Help output

Agent mode (`--json`)