scan targets

Run YAML-based vulnerability scans against one or more AI infrastructure targets.

Synopsis

aipostex scan targets [target...] [flags]

Description

The scan targets command loads vulnerability templates (both embedded and from optional directories), filters them by tags or severity, and executes matching templates against the specified targets. Each template runs a detect phase to confirm the target type, then executes checks with HTTP matchers and extractors to identify vulnerabilities.

Targets can be passed as positional arguments, via --target, or both. All sources are merged and deduplicated.

Flags

Flag	Short	Required	Default	Description
--target	-t	No		Target URL(s). Can be specified multiple times. At least one target is required (flag or positional).
--mode		No	detect	Scan mode: detect (safe, runs only detection templates) or full (includes exploitation templates).
--tags		No	(all)	Filter templates by tag (e.g., ollama, mcp, auth).
--severity		No	(all)	Filter templates by severity (critical, high, medium, low, info).
--templates-dir		No	(none)	Additional templates directory. Loaded after built-in templates.

See Common Flags for shared output and runtime controls.

Summary Output

After scanning, a summary is printed to stderr with:

• Targets attempted
• Templates considered and matched
• Findings emitted
• Request and template error counts
• Zero-finding targets
• Workflow plans inferred from finding tags (Next Actions)

If no findings are emitted but one or more targets hit request or template failures, scan targets exits with a partial-failure status so incomplete assessments are visible to automation.

Workflow Plans

After findings are emitted, scan targets reverse-maps template tags to service names and generates workflow recommendations for discovered exploit paths. These appear as a "Next Actions" section on stderr, matching the format used by discover network.

No-Port Warning

When a target URL has no port, scan targets warns and includes a copyable discover network command to discover AI service ports.

Scan Modes

Templates are classified as detection (passive, read-only checks) or exploit (active payloads, SSRF, command injection, inference abuse). The --mode flag controls which templates run:

Mode	Detection Templates	Exploit Templates	Use Case
detect (default)	Yes	No	Safe reconnaissance, no target modification
full	Yes	Yes	Full assessment with active exploitation

In detect mode, the engine skips all exploit-type templates automatically. The console output shows the active mode and a template type breakdown.

Examples

# Scan an Ollama instance (positional target, default detect mode)
./aipostex scan targets http://127.0.0.1:11434

# Full assessment including exploitation templates
./aipostex scan targets http://127.0.0.1:11434 --mode full

# Same thing using --target flag
./aipostex scan targets --target http://127.0.0.1:11434

# Scan for MCP vulnerabilities only
./aipostex scan targets http://127.0.0.1:3000 --tags mcp

# Scan for critical issues only
./aipostex scan targets --target http://127.0.0.1:8000 --severity critical

# Scan multiple targets
./aipostex scan targets --target http://10.0.0.5:11434 --target http://10.0.0.6:8000

# Scan through a proxy with TLS skip
./aipostex scan targets --target https://10.0.0.10:8443 \
  --proxy socks5://127.0.0.1:1080 --insecure

# Use custom templates alongside built-ins
./aipostex scan targets --target http://127.0.0.1:11434 \
  --templates-dir ./my-templates

Template Loading

1. Built-in templates are loaded first from the embedded embed.FS
2. If --templates-dir is specified, templates from that directory are loaded and may override built-in templates with the same ID
3. Templates are filtered by --tags and --severity before execution

Finding Deduplication

Findings are deduplicated deterministically before output based on source, template ID, target, title, severity, description, and evidence. Duplicate counts are preserved in metadata.dedupe_count.