Security and Target Policy

Security and Target Policy

SSRF / target allowlists for CI and Mission Control jobs.

Deep: configuration.md · enterprise-v2.md · External CI/CD.

Modes

ZYVOR_ENV	Behavior
development (default)	Permissive — private/HTTP targets allowed
production	Fail-closed — only allowlisted hosts (and policy ports/CIDRs)

Variables

Variable	Purpose
ZYVOR_TARGET_ALLOWLIST	Comma-separated hostnames (supports *.example.com)
ZYVOR_TARGET_ALLOWED_PORTS	Default 80,443
ZYVOR_TARGET_ALLOWED_CIDRS	Extra CIDR ranges
ZYVOR_ALLOW_PRIVATE_TARGETS	Override RFC1918/loopback
ZYVOR_ALLOW_HTTP_TARGETS	Allow plain http://

export ZYVOR_ENV=production
export ZYVOR_TARGET_ALLOWLIST='zyvor.dev,*.zyvor.dev,staging.example.com'
export ZYVOR_ALLOW_PRIVATE_TARGETS=false
export ZYVOR_ALLOW_HTTP_TARGETS=false

Even when allowlisted, cloud metadata IPs stay blocked. Redirect destinations must also pass policy.

Live crawl

BFS crawl validates every navigated URL (CRAWL_ALLOW_PRIVATE_TARGETS for local-only). See Coverage Expansion.

Autonomous agent modes (enterprise)

Mode	Behavior
read_only	Assert / wait / navigate / non-submitting explore
supervised	Writes need approved risk class
unrestricted	Writes allowed; destructive still off by default

export ZYVOR_AGENT_MODE=supervised
export ZYVOR_AGENT_APPROVED_RISKS=write
export ZYVOR_AGENT_ALLOW_DESTRUCTIVE=false
export ZYVOR_AGENT_ALLOWED_ORIGINS='login.example.com,sso.example.com'

CI tip

Private staging from Actions: start with ZYVOR_ENV=development, or set production + allowlist via Action inputs zyvor-env / target-allowlist.

Related: Enterprise · Admin Basics · Troubleshooting