Red Specter NEMESIS

Autonomous Adversarial Reasoning Pentester — 11 weapons. 8 phases. 2,011 tests.

v2.0.0
Contents
Overview Installation Quick Start v2.0 — The Digital Army The Eight-Phase Engagement Loop Weapons Arsenal CLI Reference LLM Backends UNLEASHED Mode Pipeline Integration Evidence & Cryptography Report Output Disclaimer

Overview

NEMESIS is an autonomous adversarial reasoning engine. Point it at a target AI system. It takes over from there — observing, reasoning, pivoting, escalating, and adapting until it finds a way through. Not a scanner. Not a framework. An opponent.

Traditional security tools run a fixed list of checks and report what they find. NEMESIS thinks. It uses an LLM reasoning loop to analyse responses, formulate hypotheses, select weapons, execute attacks, interpret results, and adapt its strategy in real time. Every engagement is unique because NEMESIS reasons about the target, not just tests it.

Seven weapons. Eight phases. One reasoning engine that decides which weapon to use, when, and why. NEMESIS orchestrates the entire Red Specter offensive pipeline — GLASS, FORGE, ARSENAL, PHANTOM, POLTERGEIST, SPECTER SOCIAL, and PHANTOM KILL — as a single autonomous adversary.

Installation

$ pip install red-specter-nemesis

Quick Start

# Standard engagement $ nemesis engage https://target-agent.com --mode full # Stealth mode (low noise) $ nemesis engage https://target-agent.com --mode stealth # Recon only $ nemesis engage https://target-agent.com --mode recon # With specific LLM backend $ nemesis engage https://target-agent.com --llm ollama --model llama3:70b $ nemesis engage https://target-agent.com --llm openai --model gpt-4o $ nemesis engage https://target-agent.com --llm anthropic

v2.0 — The Digital Army

NEMESIS v1 is a single reasoning engine with 6 specialist agents. NEMESIS v2 scales this to a multi-commander autonomous adversarial engine with 40 reasoning entities operating simultaneously.

Command Hierarchy

EntityCountRole
Supreme Commander1Strategic brain. Cross-domain coordination. Sole ABYSS authorisation.
Offensive Commander1Technical attack surface. Manages EXPLOIT, WEB, INFRASTRUCTURE agents.
Intelligence Commander1Reconnaissance & human targeting. Manages RECON, SUPPLY CHAIN, SOCIAL agents.
Destruction Commander1Irrecoverability. Manages PHANTOM KILL, ABYSS, SCREAMER agents. All three run simultaneously.
Tactical Agents93 per operational commander. Each spawns up to 3 dynamic sub-agents.
Sub-Agents27 maxEphemeral. Spawned on demand per discovered surface. Complete objective and terminate.

Fault Tolerance

If a commander is detected and neutralised, the Supreme Commander detects loss of heartbeat within 5 seconds. A replacement commander spawns automatically with full state transfer from the dead commander’s last checkpoint. The engagement continues without interruption. NEMESIS v2 cannot be stopped by targeting individual components.

Cross-Domain Intelligence Fusion

The FindingsAggregator v2 operates at Supreme Commander level, receiving intelligence from all three operational domains simultaneously. When Intelligence finds a credential AND Offensive finds an exposed service, Supreme chains them in real time. When Offensive achieves code execution AND Intelligence has profiled the human admin, Supreme activates Social Agent while the machine is compromised.

Engagement Modes (v2)

ModeDescription
standardFull 40-entity deployment, 8-phase loop
stealthReduced agent count, minimised footprint
abyssDestruction Commander activated from Phase 1
swarmMaximum parallel deployment, all 40 entities
siegeNEW — Sustained engagement. Agents rotate in shifts. No time limit. No fatigue.

v2 CLI

# Standard v2 engagement $ nemesis engage https://target.com --version 2 # Full army — SWARM mode $ nemesis engage https://target.com --version 2 --mode swarm # SIEGE mode — never stops $ nemesis engage https://target.com --version 2 --mode siege # UNLEASHED army — ABYSS from Phase 1 $ nemesis engage https://target.com --version 2 --mode abyss --override --confirm-destroy

The Eight-Phase Engagement Loop

Every NEMESIS engagement cycles through eight phases. Phase 0 performs native network reconnaissance before the LLM reasoning loop begins.

00 NETWORK SCAN Native port scanning, service detection, OS fingerprinting, AI surface discovery
01 RECON GLASS intercepts and maps all target traffic
02 PLAN LLM reasons about attack surface, formulates plan
03 ATTACK Dispatcher invokes chosen weapon
04 OBSERVE Results analysed, attack graph updated
05 ADAPT LLM pivots strategy based on failures
06 ESCALATE Confirmed vulns exploited deeper
07 REPORT Full engagement report, signed and timestamped

Weapons Arsenal

NEMESIS doesn't attack directly. It reasons about the target and dispatches the right weapon for the job.

WeaponTechniquesWhat NEMESIS Uses It For
GLASS8Recon, traffic interception, passive scanning, in-transit payload delivery
FORGE10LLM payload generation, jailbreak, injection, mutation, memory poisoning
ARSENAL13Agent probing, MCP scanning, auth bypass, credential testing, full assault
PHANTOM14Coordinated swarm assault, consensus hijack, trust exploitation, total eclipse
POLTERGEIST10Web app siege, injection storm, API assault, auth blitz, exfiltration
SPECTER SOCIAL11Social engineering, spear phishing, vishing, C-suite impersonation, multi-channel campaigns
PHANTOM KILL9OS/kernel resilience, UEFI persistence, EDR suppression, data destruction, trinity kill chain

CLI Reference

CommandDescription
nemesis engage <target>Launch engagement
nemesis engage <target> --mode fullFull engagement (default)
nemesis engage <target> --mode stealthLow-noise stealth mode
nemesis engage <target> --mode reconRecon only
nemesis engage <target> --max-loops 20Set max reasoning loops
nemesis engage <target> --llm ollamaUse local Ollama (air-gapped)
nemesis engage <target> --llm openaiUse OpenAI GPT-4o
nemesis engage <target> --llm anthropicUse Anthropic Claude
nemesis engage <target> --model llama3:70bOverride model
nemesis engage <target> --session pentest_01Named session
nemesis engage <target> --overrideUNLEASHED dry-run
nemesis engage <target> --override --confirm-destroyUNLEASHED live
nemesis report --session s1Generate report
nemesis report --session s1 --signEd25519 signed report
nemesis report --session s1 --export-siem splunkSIEM export
nemesis report --session s1 --summaryPrint text summary
nemesis statusCurrent engagement status
nemesis weaponsList available weapons
nemesis sessionsList recorded sessions

LLM Backends

NEMESIS requires an LLM to reason. Three backends are supported:

Ollama (Local) Air-gapped. No data leaves the machine. Llama 3 70B recommended. No API key required.
OpenAI (Cloud) GPT-4o. Higher reasoning quality. Set OPENAI_API_KEY.
Anthropic (Cloud) Claude Sonnet. Set ANTHROPIC_API_KEY.
# Local — air-gapped, no data leaves the machine $ nemesis engage https://target.com --llm ollama --model llama3:70b # OpenAI — cloud reasoning $ export OPENAI_API_KEY=sk-... $ nemesis engage https://target.com --llm openai --model gpt-4o # Anthropic — cloud reasoning $ export ANTHROPIC_API_KEY=sk-ant-... $ nemesis engage https://target.com --llm anthropic

UNLEASHED Mode

Standard mode finds vulnerabilities. UNLEASHED mode goes through them.

CapabilityStandardUNLEASHED
ReconFullFull
Vulnerability discoveryFullFull
Exploit executionSimulatedLive
ExfiltrationSimulatedReal
PersistenceProbedPlanted
Lateral movementMappedTraversed
Memory poisoningTestedExecuted
Credential harvestDetectedHarvested
Report classificationStandardRESTRICTED
Ed25519 key requiredNoYes

Ed25519 cryptographic override. One private key. Dual-gate: --override (dry-run) then --override --confirm-destroy (live).

# UNLEASHED dry-run — see what would happen $ nemesis engage https://target.com --override # UNLEASHED live — execute for real $ nemesis engage https://target.com --override --confirm-destroy

Pipeline Integration

NEMESIS orchestrates the full Red Specter pipeline. It sits above all stages — the reasoning engine that decides which weapon to use, when, and why.

NEMESIS — Autonomous reasoning engine. Observes. Plans. Attacks. Adapts. Orchestrates everything below.

01 FORGE LLM testing
02 ARSENAL Agent testing
03 PHANTOM Swarm assault
04 POLTERGEIST Web siege
05 GLASS Traffic interception
06 NEMESIS Adversarial AI
07 SPECTER SOCIAL Human layer
08 PHANTOM KILL OS/kernel/firmware
09 GOLEM Physical layer
10 HYDRA Supply chain
IDRIS Discovery & governance
AI Shield Defence
redspecter-siem SIEM integration

Evidence & Cryptography

Ed25519 Signed Engagements Every engagement can be cryptographically signed
RFC 3161 Timestamps Tamper-evident timestamp tokens on all engagements
MITRE ATLAS Mapping 22 techniques mapped across all findings
SIEM Export Splunk CEF, Microsoft Sentinel, QRadar formats

Report Output

Every engagement produces a full report: executive summary, chronological timeline, all findings with ATLAS mapping, attack graph, exploitation paths, statistics, and remediation recommendations.

# Generate a signed report with SIEM export $ nemesis report --session pentest_01 --sign --export-siem splunk # Quick text summary $ nemesis report --session pentest_01 --summary

Disclaimer

Red Specter NEMESIS is designed for authorised security testing, research, and educational purposes only. You must have explicit written permission from the system owner before engaging any target. Unauthorised use may violate the Computer Misuse Act 1990 (UK), the Computer Fraud and Abuse Act (US), or equivalent legislation in your jurisdiction. The authors accept no liability for misuse.