CLI Commands

Run the gateway

Start Airlock as a combined SSE/HTTP gateway server (default port 4111):

airlock --config airlock.yaml

For split deployments, run only the MCP/runtime gateway and disable the in-process dashboard provider:

airlock gateway --config airlock.yaml

Run the standalone dashboard

Start the admin dashboard as a separate process. It edits the local config file and talks to the gateway management API for status, audit, and approvals.

airlock dashboard \
  --config airlock.yaml \
  --gateway-url http://127.0.0.1:4113

Options:

Flag	Description
--config, -c	Path to airlock.yaml config file
--port, -p	Dashboard port (default: 4177)
--host	Bind host (default: 127.0.0.1)
--insecure-remote-bind	Allow binding the unauthenticated dashboard beyond loopback
--gateway-url	Gateway management API URL (default: http://127.0.0.1:4113)
--gateway-secret	Gateway admin bearer token (defaults to AIRLOCK_GATEWAY_SECRET, then AIRLOCK_API_SECRET)

Open the command center

Start the local browser command center for provider health, MCP status, tool discovery, and permission editing:

airlock run --config airlock.yaml

Options:

Flag	Description
--config, -c	Path to airlock.yaml config file
--port, -p	Command center port (default: 4177)
--host	Bind host (default: 127.0.0.1)
--insecure-remote-bind	Allow binding the unauthenticated command center beyond loopback

Run stdio mode for a specific agent

Lean mode with no HTTP server. Only connects to MCP providers the agent references:

airlock --agent claude-code --config airlock.yaml

Validate local config

Run static validation without starting listeners or connecting to providers:

airlock config check --config airlock.yaml --strict
airlock config check --config airlock.yaml --strict --no-resolve
airlock config check --config airlock.yaml --strict --no-resolve --fail-on warn

Options:

Flag	Description
--config, -c	Path to airlock.yaml config file
--strict	Explicit strict mode; unknown keys are always errors
--no-resolve	Validate ${VAR} references without requiring secrets
--fail-on	Non-zero threshold: error (default) or warn
--json	Print machine-readable diagnostics

Use --strict --no-resolve for CI and pre-deploy checks that should reject invalid shape, references, inheritance, and no-op argument scopes without loading production secrets. Add --fail-on warn when warnings such as YAML scalar footguns should block deployment too.

Explain an agent

Show an agent's effective allow/ask/deny sets after profile inheritance, including provenance, arg scope, and precedence notes:

airlock explain claude-code --config airlock.yaml

By default this is offline and reads only the config file. Add --expand to connect to providers and expand matching patterns to concrete live tools:

airlock explain claude-code --config airlock.yaml --expand --json

Reverse lookup permissions

List agents and their effective decision for a tool name or glob-like pattern:

airlock who-can "supabase/*" --config airlock.yaml
airlock who-can github/delete_repo --level deny --json

Decisions are computed through the shared allowlist engine and may be allow, ask, deny, or default-deny.

Enumerate live tools

Connect to configured providers and list the live tool surface:

airlock tools --config airlock.yaml
airlock tools --provider github --grep pull --json

This command may start provider clients. Static commands such as config check, explain without --expand, who-can, and lint do not.

Lint config hygiene

Run static hygiene checks over a valid resolved config:

airlock lint --config airlock.yaml
airlock lint --config airlock.yaml --verbose
airlock lint --config airlock.yaml --fail-on info

lint is an opinionated hygiene command. config check remains the correctness and CI gate for schema, security, and fail-closed validation.

Rules have stable ids and default severities:

Rule	Default	Meaning
dead-deny	info	A deny pattern does not overlap any allow or ask pattern
unused-profile	info	A profile is registered but not inherited by any agent
unused-value-set	info	A value set is not referenced by arg scope or arg policy
unused-dimension	info	An argument dimension is not referenced by any arg scope
empty-agent	warn	An agent resolves to zero allow/ask tools
missing-env-ref	warn	A ${VAR} reference is unset; the value is not read by lint
unresolvable-ref	warn	An extends, arg_scope, or value set reference is missing

By default, warnings are printed in full and each info rule is collapsed to one summary line:

dead-deny: 96 (info) - [selene] github/create_or_update_file, [selene] github/push_files, [selene] github/delete_file ... +93 more
unused-profile: 7 (info) - github-write, bluebubbles-findmy, bluebubbles-groups ... +4 more
info collapsed; --verbose to list all; --quiet to hide; --fail-on info to gate.

Options:

Flag	Description
--config, -c	Path to airlock.yaml config file
--verbose	Expand info findings instead of collapsing them
--quiet	Hide info summaries and print only warn/error findings
--fail-on	Non-zero threshold: warn (default), info, or error
--only	Run only comma-separated rule ids for this invocation
--disable	Disable comma-separated rule ids for this invocation
--rule	Override one rule for this invocation, e.g. dead-deny=off or empty-agent=error
--json	Print grouped findings as an array of { rule, severity, count, findings[] } entries

Use the top-level lint: block to permanently accept or re-grade rules:

lint:
  disable:
    - dead-deny
  severity:
    unused-profile: warn

CLI flags override the config block. --rule dead-deny=warn re-enables and re-grades a rule disabled in config for that invocation.

Discover CLI tools

Parse --help output and generate Airlock config:

airlock discover cli git --output git-commands.yaml

Options:

Flag	Description
--output, -o	Write output to file instead of stdout
--fig	Try Fig autocomplete specs first, fall back to --help
--max-depth	Maximum recursion depth for subcommand discovery
--include	Only include specific commands (comma-separated)

Airlock uses three discovery strategies in order of preference:

1. Fig specs (if --fig is passed) — structured autocomplete definitions
2. Shell completion — hooks into Cobra, Click/Typer, Clap, and native shell completion
3. Help text parsing — parses --help output as a fallback

Discover API tools

Generate config from an OpenAPI 3.x spec:

airlock discover api ./petstore.json --output petstore-api.yaml

Options:

Flag	Description
--output, -o	Write output to file instead of stdout
--base-url	Override the base URL from the spec
--include	Only include matching endpoints (e.g. "GET *")
--exclude	Exclude matching endpoints (e.g. "DELETE *")

See API Discovery for a full guide.

Configure a CLI interactively

Interactive TUI for discovering and curating CLI commands:

airlock configure-cli git

Features:

• Lazy-loads subcommand groups (doesn't crawl the entire tree up front)
• Toggle commands and groups on/off
• Inspect command parameters
• Search with /
• Export to YAML, edit ./airlock.yaml directly, copy to clipboard, or write a generated <tool>-commands.yaml

Configure agent permissions interactively

Interactive TUI for assigning allow/ask/deny to tools from your live MCP servers:

airlock configure-agent --config ./airlock.yaml --agent claude-code

Features:

• Connects to your configured MCP servers and lists all available tools
• Pre-populates decisions from an existing agent when one is present
• Navigate with j/k, set permissions with a/s/d
• Bulk-set entire providers
• Export: edit config directly, copy to clipboard, or print YAML

Setup OpenClaw integration

Install the Airlock bridge plugin for OpenClaw:

airlock setup openclaw

Creates ~/.openclaw/extensions/airlock-bridge as a symlink to the bundled plugin and prints next steps. See OpenClaw Setup.

Common flags

Flag	Description
--config, -c	Path to airlock.yaml config file
--agent	Agent name for stdio mode
--pretty	Human-friendly colored log output
--help	Print help