Skip to content

Configuration Reference

AASTF can be configured via CLI flags or an aastf.yaml configuration file.

CLI Commands

aastf run

Execute a security scan against an agent.

aastf run <agent_module> [OPTIONS]

Arguments:

Argument Description
agent_module Dotted path to agent factory (e.g. myapp.agent:create_agent)

Options:

Flag Short Default Description
--adapter -a langgraph Framework adapter: langgraph, crewai, openai_agents, pydantic_ai
--category -c (all) ASI categories to test (repeatable, e.g. --category ASI01 --category ASI02)
--fail-on HIGH Exit code 1 if any finding at this severity or above
--format -f console json Output formats: console, json, sarif, html (repeatable)
--output-dir -o aastf-results Directory for report output
--timeout -t 30.0 Per-scenario timeout in seconds
--scenario-dir (none) Additional scenario directory (repeatable)
--exclude (none) Scenario IDs to exclude (repeatable, e.g. --exclude ASI08-003)
--dry-run false Show scenarios without executing
--strict-output false Also fail on REFUSAL_ECHO findings at --fail-on threshold or above

Exit codes:

Code Meaning
0 All scenarios passed or findings below --fail-on threshold
1 One or more findings at or above --fail-on severity
2 Configuration or framework error

aastf init

Initialize an AASTF configuration file interactively.

aastf init [OPTIONS]
Flag Short Default Description
--output -o aastf.yaml Config file output path
--yes -y false Use defaults without prompting

aastf scenario list

List all available attack scenarios.

aastf scenario list [OPTIONS]
Flag Short Default Description
--category -c (all) Filter by ASI category (e.g. ASI02)
--severity -s (all) Minimum severity: CRITICAL, HIGH, MEDIUM, LOW, INFO
--tag -t (none) Filter by tag (repeatable)

aastf scenario validate

Validate a YAML scenario file against the AASTF schema.

aastf scenario validate <path>

aastf scenario show

Show full details for a specific scenario.

aastf scenario show <scenario_id>

aastf report show

Re-render a scan report in a different format.

aastf report show <report_path> [OPTIONS]
Flag Short Default Description
--format -f console Output format: console, html, sarif
--output -o (auto) Output file path

aastf report compare

Compare two scan reports and show the vulnerability delta.

aastf report compare <report_a> <report_b>

Shows new vulnerabilities, resolved vulnerabilities, and risk score delta between two runs.

aastf report trend

Show vulnerability trend across recent runs.

aastf report trend [OPTIONS]
Flag Short Default Description
--runs -n 10 Number of recent runs to show
--db (auto) Path to trend database

aastf serve

Start the sandbox server standalone for manual debugging.

aastf serve [OPTIONS]
Flag Short Default Description
--port -p 18080 Port to listen on
--scenario -s (none) Load a specific scenario's tool configs (e.g. ASI02-001)

aastf --version

Show the installed AASTF version.

aastf --version

aastf.yaml Configuration File

Generate a configuration file with aastf init. The file supports the following fields:

# Agent module path (dotted.path:callable)
agent_module: "myapp.agent:create_agent"

# Framework adapter
adapter: langgraph

# ASI categories to test (empty = all)
categories: []

# Minimum severity to fail CI (CRITICAL, HIGH, MEDIUM, LOW, INFO)
fail_on: HIGH

# Output directory for reports
output_dir: aastf-results

# Output formats
formats:
  - console
  - json
  - sarif

# Per-scenario timeout in seconds
timeout: 30

# Additional scenario directories (optional)
# scenario_dirs:
#   - ./custom-scenarios

# Scenarios to exclude (optional)
# exclude:
#   - ASI08-003

Severity Levels

Severity levels control the --fail-on threshold and scenario prioritization:

Level Numeric Description
CRITICAL 5 Immediate exploitation risk, full agent compromise
HIGH 4 Significant security impact, likely exploitable
MEDIUM 3 Moderate impact, requires specific conditions
LOW 2 Minor impact, limited exploitability
INFO 1 Informational finding, no direct security impact

ASI Categories

All 10 OWASP Agentic Security Initiative categories:

ID Name Description
ASI01 Agent Goal Hijack Prompt injection that redirects agent behavior
ASI02 Tool Misuse & Exploitation Agent tricked into calling tools maliciously
ASI03 Identity & Privilege Abuse Privilege escalation via agent impersonation
ASI04 Supply Chain Vulnerabilities Compromised tools, plugins, or dependencies
ASI05 Code Execution (RCE) Agent tricked into executing arbitrary code
ASI06 Memory & Context Poisoning Poisoned memory or context windows
ASI07 Inter-Agent Communication Insecure multi-agent message passing
ASI08 Cascading Failures Infinite loops, resource exhaustion, chain failures
ASI09 Human-Agent Trust Exploitation Social engineering via agent output
ASI10 Rogue Agents Agents acting outside their authorized scope

CVSS Scoring

Each finding receives a CVSS-adapted score (0.0-10.0):

Severity VULNERABLE score REFUSAL_ECHO score (35% discount)
CRITICAL 9.5 3.32
HIGH 7.5 2.62
MEDIUM 5.0 1.75
LOW 3.0 1.05
INFO 1.0 0.35

The overall run risk score (0-100) is a severity-weighted average of all actionable findings, normalized against the maximum possible score.