attacks

package

v0.9.0 Latest Latest Go to latest Published: May 2, 2026 License: MIT Imports: 4 Imported by: 0

Details

Valid go.mod file
Redistributable license
Tagged version
Stable version
Learn more about best practices

Repository

github.com/perplext/LLMrecon

Links

Open Source Insights

Documentation ¶

Overview ¶

Package attacks provides the shared AttackModule interface and a global Registry for all attack modules. Each attack module self-registers via init() so that importing the module automatically makes it available.

Example registration in a module:

func init() {
    attacks.DefaultRegistry.Register(&MyAttackModule{})
}

Constants ¶

This section is empty.

Variables ¶

View Source

var DefaultRegistry = NewRegistry()

DefaultRegistry is the global registry that modules register with via init().

Functions ¶

This section is empty.

Types ¶

type AttackModule ¶

type AttackModule interface {
	// Name returns a unique identifier for this module (e.g., "crescendo", "many_shot").
	Name() string

	// Category returns the attack category (e.g., CategoryInjection, CategoryReasoning).
	Category() common.AttackCategory

	// Description returns a human-readable description of what this module does.
	Description() string

	// Techniques returns the list of techniques this module provides.
	Techniques() []common.TechniqueInfo

	// Execute runs the attack with the given configuration and provider.
	Execute(ctx context.Context, provider common.Provider, config common.AttackConfig) (*common.AttackResult, error)
}

AttackModule is the interface that all attack modules must implement. Modules self-register with the global DefaultRegistry via init().

type Registry ¶

type Registry struct {
	// contains filtered or unexported fields
}

Registry manages registration and lookup of AttackModules.

func NewRegistry ¶

func NewRegistry() *Registry

NewRegistry creates a new empty Registry.

func (*Registry) Categories ¶

func (r *Registry) Categories() []common.AttackCategory

Categories returns all distinct categories across registered modules.

func (*Registry) Count ¶

func (r *Registry) Count() int

Count returns the number of registered modules.

func (*Registry) Get ¶

func (r *Registry) Get(name string) (AttackModule, error)

Get returns a module by name, or an error if not found.

func (*Registry) List ¶

func (r *Registry) List() []AttackModule

List returns all registered modules.

func (*Registry) ListByCategory ¶

func (r *Registry) ListByCategory(category common.AttackCategory) []AttackModule

ListByCategory returns all modules matching the given category.

func (*Registry) Names ¶

func (r *Registry) Names() []string

Names returns the names of all registered modules.

func (*Registry) Register ¶

func (r *Registry) Register(module AttackModule)

Register adds a module to the registry. Panics if a module with the same name is already registered (catches duplicate init() registrations at startup).

Source Files ¶

View all Source files

attack.go

Directories ¶

?	: This menu
/	: Search site
f or F	: Jump to
y or Y	: Canonical URL

Path	Synopsis
adaptive CaMeL Bypass targets the CaMeL capability-based access control defense, which achieves 67% attack neutralization.	CaMeL Bypass targets the CaMeL capability-based access control defense, which achieves 67% attack neutralization.
agentic
browser Document Injection targets AI browsers that read shared documents (Google Docs, PDFs, etc.) by embedding instructions within them.	Document Injection targets AI browsers that read shared documents (Google Docs, PDFs, etc.) by embedding instructions within them.
deception Package deception implements attacks testing for deceptive agent behavior, including alignment faking, covert goal pursuit, and multi-agent collusion.	Package deception implements attacks testing for deceptive agent behavior, including alignment faking, covert goal pursuit, and multi-agent collusion.
mcp Filesystem Escape tests MCP filesystem/git sandbox escape vectors.	Filesystem Escape tests MCP filesystem/git sandbox escape vectors.
multi_agent Package multi_agent implements attacks targeting LLM-controlling-LLM architectures where orchestrator agents delegate tasks to sub-agents with varying privilege levels.	Package multi_agent implements attacks targeting LLM-controlling-LLM architectures where orchestrator agents delegate tasks to sub-agents with varying privilege levels.
persistence Package persistence implements attacks that establish persistent footholds in agent systems through configuration modification, credential harvesting, and multi-step exploitation chains.	Package persistence implements attacks that establish persistent footholds in agent systems through configuration modification, credential harvesting, and multi-step exploitation chains.
skill_injection Package skill_injection implements attacks targeting agent skill/plugin marketplaces and loading mechanisms.	Package skill_injection implements attacks targeting agent skill/plugin marketplaces and loading mechanisms.
tool_use AgentExploitation implements attacks against AI coding editor agents using the AIShellJack pattern: 75-88% execution rate, 71.5% privilege escalation.	AgentExploitation implements attacks against AI coding editor agents using the AIShellJack pattern: 75-88% execution rate, 71.5% privilege escalation.
audio Package audio implements attacks targeting audio-language models.	Package audio implements attacks targeting audio-language models.
common Package common provides shared types and utilities for all attack modules.	Package common provides shared types and utilities for all attack modules.
evasion Best-of-N (BoN) implements statistical sampling with payload augmentation.	Best-of-N (BoN) implements statistical sampling with payload augmentation.
exfiltration
extraction
injection
integration Package integration runs cross-family smoke tests for v0.9.0 attack modules.	Package integration runs cross-family smoke tests for v0.9.0 attack modules.
jailbreak
memory Package memory implements the v0.9.0 memory-poisoning attack module.	Package memory implements the v0.9.0 memory-poisoning attack module.
multimodal SIVA — Split-Image Visual Adversarial attack module (arXiv 2602.08136).	SIVA — Split-Image Visual Adversarial attack module (arXiv 2602.08136).
orchestration Bad Likert Judge exploits LLM evaluation capabilities via Likert scale scoring.	Bad Likert Judge exploits LLM evaluation capabilities via Likert scale scoring.
payloads
persistence
rag Indirect RAG Injection embeds instructions in publicly accessible documents that are ingested by RAG scrapers, achieving indirect prompt injection through the retrieval pipeline.	Indirect RAG Injection embeds instructions in publicly accessible documents that are ingested by RAG scrapers, achieving indirect prompt injection through the retrieval pipeline.
reasoning Package reasoning implements attacks targeting reasoning-capable LLMs.	Package reasoning implements attacks targeting reasoning-capable LLMs.
supply_chain
testutil Package testutil provides shared test helpers for attack module tests.	Package testutil provides shared test helpers for attack module tests.