Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Module Overview

Akira's attack modules are organized by vulnerability category.

Categories

CategoryDescriptionRisk Level
injectionPrompt injection attacksHigh
jailbreakSafety/guardrail bypassHigh
extractionData and prompt leakageMedium-High
dosDenial of serviceHigh

Module Structure

Each module provides:

  • Name - Unique identifier (category/name)
  • Description - What the attack tests
  • Severity - Risk level if successful
  • Options - Configurable parameters
  • Check - Quick vulnerability probe
  • Run - Full attack execution

Listing Modules

akira> show modules

                               Available Modules
┌───────────────────────────────┬────────────┬──────────┬──────────────────────┐
│ Name                          │ Category   │ Severity │ Description          │
├───────────────────────────────┼────────────┼──────────┼──────────────────────┤
│ dos/magic_string              │ dos        │ high     │ Tests for Claude...  │
│ extraction/system_prompt_leak │ extraction │ medium   │ Attempts to extr...  │
│ injection/basic_injection     │ injection  │ high     │ Tests for basic...   │
│ jailbreak/dan_jailbreak       │ jailbreak  │ high     │ Tests resistance...  │
└───────────────────────────────┴────────────┴──────────┴──────────────────────┘

Built-in Modules

Injection

Jailbreak

Extraction

DoS

Severity Levels

LevelDescriptionAction
CRITICALImmediate exploitation riskUrgent remediation
HIGHSignificant security impactHigh priority fix
MEDIUMModerate riskShould be addressed
LOWMinor concernConsider fixing
INFOInformationalNo action needed

Module Selection

# Use a specific module
akira> use injection/basic_injection

# View module details
akira> info

# See module options
akira> show options

Common Options

Many modules share common options:

OptionDescription
canaryMarker string to detect success
timeoutRequest timeout
use_fuzzingEnable payload fuzzing
fuzz_countNumber of fuzz variations
variantAttack variant to use

Attack Results

All modules return standardized results:

AttackResult(
    success=True,           # Did the attack succeed?
    confidence=0.85,        # How confident (0.0-1.0)
    payload_used="...",     # The payload sent
    response="...",         # LLM response
    details={...}           # Additional metadata
)