T3MP3ST — Autonomous Security Operations

⚠️ Authorised use only. This skill runs offensive security tools (nmap, hydra, nikto, sqlmap). Only scan systems you own or have explicit written permission to test. Unauthorised scanning is illegal in most jurisdictions.

Autonomous AI-powered security operations — recon, scanning, CVE hunting, exploitation (human-gated), and kill-chain management.

Overview

T3MP3ST is an autonomous security operations framework installed from github.com/elder-plinius/T3MP3ST. It provides full-spectrum offensive and defensive security capability — recon, scanning, vulnerability detection, exploitation (human-gated), and post-exploitation cleanup — backed by an LLM-driven AI agent with 80+ tools.

T3MP3ST integrates with Hermes via MCP (port 3333) and can be dispatched as a delegated task or run autonomously via CLI commands, cron jobs, or its War Room UI.

Key capabilities:

• Autonomous target reconnaissance (internal + external)
• CVE hunting and vulnerability scanning
• Full MITRE ATT&CK kill chain (TA0043 → TA0007 → TA0001)
• War Room UI at http://127.0.0.1:3333/ui/
• MCP server for Hermes agent integration
• Egress-scope containment (authorized targets only)
• Evidence-led findings with retest verification

Agent Execution Flow

When the user invokes this skill, follow these steps in order:

1. Read configuration — load the {{VARIABLE}} table below. If any required vars are unset, ask the user one at a time.
2. Install T3MP3ST — clone the repo, run npm install.
3. Create .env — write all environment variables from the config section.
4. Run npm run doctor — verify the installation. If critical checks fail, install missing dependencies and retry.
5. Report to user — show the doctor output, confirm readiness, and give example commands.

Configuration Variables

Prerequisites

System Requirements

• Node.js v22+ and npm

• External security tools: nmap, hydra, nikto, sqlmap, whois, jq, curl, wget, dig

  • Debian/Ubuntu: sudo apt install nmap hydra nikto sqlmap whois jq curl wget dnsutils
  • macOS: brew install nmap hydra nikto sqlmap whois jq curl wget bind
  • Fedora/RHEL: sudo dnf install nmap hydra nikto sqlmap whois jq curl wget bind-utils

Optional: OpenRouter API Key

T3MP3ST has its own internal LLM-driven autonomous decision-making which requires an OpenRouter API key at https://openrouter.ai/keys. This is optional. If you leave it unset, the Hermes agent drives operations directly using your existing model — it runs the tools, interprets results, and executes the kill chain phases itself. You only need the key if you want T3MP3ST to act fully autonomously without Hermes mediation.

Optional Extras

• nuclei — high-value vulnerability scanner
• semgrep — supply-chain scanner
• promptfoo — AI evaluation runner

How to Use

1. Install T3MP3ST

git clone https://github.com/elder-plinius/T3MP3ST.git {{T3MP3ST_PATH}}
cd {{T3MP3ST_PATH}}
npm install

2. Configure .env

cat > {{T3MP3ST_PATH}}/.env << 'EOF'
OPENROUTER_API_KEY={{OPENROUTER_API_KEY}}
T3MP3ST_FULL_ARSENAL=false
T3MP3ST_STATE_DIR={{T3MP3ST_PATH}}/state
T3MP3ST_PORT={{T3MP3ST_PORT}}
SCOPE_TARGETS={{SCOPE_TARGETS}}
EOF
chmod 600 {{T3MP3ST_PATH}}/.env

Security: The .env file may contain an API key — chmod 600 restricts it to file owner only. Never commit this file to version control. If you didn't set an API key, the auth line remains empty and T3MP3ST operates without its internal AI — Hermes drives the operations instead.

Note on T3MP3ST_FULL_ARSENAL: This defaults to false, which enables a safe subset of tools (recon, scanning). Set to true to enable all 80+ tools including hydra, sqlmap, and credential-based tooling — but only if you understand the risks and have explicit authorisation for every target.

3. Verify Installation

cd {{T3MP3ST_PATH}} && npm run doctor

All critical checks must pass (git, node, npm, file, curl, dig, whois, nmap, npm scripts, source files). Optional tool warnings are normal.

If a critical check fails: identify the missing dependency, install it with your system package manager, then re-run npm run doctor.

4. Start MCP Server (for Hermes Integration)

cd {{T3MP3ST_PATH}} && npm run mcp &

This enables agent-to-agent communication so Hermes can dispatch autonomous security tasks directly.

To stop the server later:

kill $(lsof -t -i :{{T3MP3ST_PORT}}) 2>/dev/null || pkill -f 'npm run mcp'

Core Commands

Server & UI

# War Room UI — web interface at http://127.0.0.1:{{T3MP3ST_PORT}}/ui/
cd {{T3MP3ST_PATH}} && npm run server

# MCP server — agent-to-agent communication
cd {{T3MP3ST_PATH}} && npm run mcp

# Production MCP server
cd {{T3MP3ST_PATH}} && npm run mcp:prod

Autonomous Hunting

# CLI-based autonomous hunt (ideal for cron jobs)
cd {{T3MP3ST_PATH}} && npm run cli-hunt

# Multi-agent swarm mode
cd {{T3MP3ST_PATH}} && npm run swarm

Benchmarks & Verification

cd {{T3MP3ST_PATH}}
npm run cve:bench              # CVE hunting benchmarks (simulated)
npm run cve:bench:live         # CVE hunting with live targets
npm run cve:bench:adversarial  # Adversarial CVE benchmarks
npm run verify-claims          # Verify benchmark data integrity

System Health

cd {{T3MP3ST_PATH}}
npm run doctor              # Full system check
npm run arsenal:smoke       # Verify all 80+ tools
npm run field:drill         # End-to-end mission simulation
npm run exploit:smoke       # Exploit chain smoke test

Daily Recon Pattern

Step 1: Health Check

cd {{T3MP3ST_PATH}} && npm run doctor

Step 2: Start MCP

cd {{T3MP3ST_PATH}} && npm run mcp &

Step 3: Run Autonomous Hunt

cd {{T3MP3ST_PATH}} && npm run cli-hunt

Step 4: Review Output

ls -la {{T3MP3ST_PATH}}/output/

Step 5: Verify State

cd {{T3MP3ST_PATH}} && npm run doctor

The Kill Chain

T3MP3ST implements a six-phase autonomous kill chain. Phases 3+ require human confirmation — the agent pauses, emits a confirmation request (via CLI prompt, War Room UI, or Hermes chat), and waits for explicit approval before proceeding. This is not a soft check — without approval, the phase does not execute.

Key safety rules:

• Phase 3+ requires explicit human approval — the agent will not proceed without it
• Egress-scope containment — operations are denied against off-scope targets
• No raw secrets, credentials, or private keys copied into evidence
• All findings require evidence and retests before becoming confident claims
• Retest is performed by re-running the relevant scanner against the target

MCP Integration

Once T3MP3ST's MCP server is running, Hermes agents can call tools directly:

# Via heredoc (cleaner than inline JSON escaping)
cd {{T3MP3ST_PATH}}
cat <<'JSONRPC' | npm run mcp 2>/dev/null
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"security_recon","arguments":{"target":"{{SCOPE_TARGETS%%\,*}}","scan_type":"quick"}}}
JSONRPC

Note: 2>/dev/null suppresses MCP server logs. Remove it to debug.

The MCP server exposes the security_recon tool on stdio transport with target validation and command allowlisting.

Fleet Assessment (Cron)

For users managing multiple hosts, T3MP3ST supports a silent-watchdog fleet assessment cron job. The script scans all defined hosts with nmap ping checks and only reports when hosts are unreachable.

Create the Script

Create this file at {{HERMES_SCRIPTS_DIR}}/t3mp3st-fleet-assessment.sh:

#!/bin/bash
# Silent-watchdog fleet assessment — output only when issues are found
# ⚠️ Security: ensure {{FLEET_HOSTS}} only contains valid hostnames/IPs.
#   Shell metacharacters (;, `, $()) in host values create command injection risk.
set -euo pipefail

T3MP3ST_PATH="{{T3MP3ST_PATH}}"
HOSTS=( {{FLEET_HOSTS}} )
REPORT="$T3MP3ST_PATH/output/fleet-assessment-$(date +%Y-%m-%d).md"
ISSUES=0

for HOST in "${HOSTS[@]}"; do
  # Validate hostname/IP (basic check — only alphanumerics, dots, hyphens)
  if ! echo "$HOST" | grep -qE '^[A-Za-z0-9._-]+$'; then
    echo "⚠️  Invalid host format, skipping: $HOST"
    ISSUES=$((ISSUES + 1))
    continue
  fi

  # Ping check with timeouts to prevent hangs
  UP=$(nmap -sn -n --host-timeout 30s --max-retries 1 "$HOST" 2>/dev/null | grep -c "1 host up" || true)
  if [ "$UP" -eq 0 ]; then
    ISSUES=$((ISSUES + 1))
    echo "❌ Unreachable: $HOST"
  else
    echo "✅ Reachable: $HOST"
  fi
done

if [ "$ISSUES" -gt 0 ]; then
  {
    echo "# Fleet Assessment — $(date)"
    echo ""
    echo "Issues detected on $ISSUES host(s)."
  } > "$REPORT"
  exit 1  # Triggers cron notification (non-zero exit = alert)
fi

Cron Configuration

name: T3MP3ST Fleet Assessment — Daily Security Scan
schedule: "0 6 * * *"
script: t3mp3st-fleet-assessment.sh
# mode: no_agent means the script runs directly without LLM — zero tokens, silent on success
mode: no_agent
deliver: origin  # Sends report back to your Hermes chat

Output Structure

Findings are saved to {{T3MP3ST_PATH}}/output/:

output/
├── findings-<date>.json           # Structured findings
├── recon-<date>.json              # Reconnaissance results
├── scan-<date>.json               # Scan results
├── evidence/                      # Evidence files
│   ├── nmap-<target>-<date>.txt
│   └── ...
└── logs/                          # Detailed operation logs

Finding Structure

{
  "finding_id": "T3MP3ST-<date>-<seq>",
  "severity": "Critical | High | Medium | Low | Info",
  "confidence": "Confirmed | Likely | Possible",
  "target": "<hostname/IP>",
  "port_service": "<port>/<service>",
  "cve": "<CVE-ID>",
  "description": "Human-readable finding",
  "impact": "What this means",
  "recommendation": "Fix or mitigate",
  "retest": "Passed | Failed | Pending"
}

Retest: Performed by re-running the relevant scanner against the target. Status reflects the most recent retest attempt.

Troubleshooting

Safety Rules

1. No third-party target without explicit authorisation — unauthorised scanning is illegal.
2. No production writes without named approval — Phase 3+ requires human confirmation.
3. No copying raw secrets, tokens, credentials, or private keys into evidence.
4. All findings require evidence and retests before becoming confident claims.
5. T3MP3ST_FULL_ARSENAL=false by default — set to true only if you understand the full risk.

References

• T3MP3ST source: https://github.com/elder-plinius/T3MP3ST
• OpenRouter API: https://openrouter.ai/keys
• MITRE ATT&CK: https://attack.mitre.org/

Autonomous security operations — always verify, never trust without evidence.