Autonomous OODA-loop pentesting agent. Hermes brain, Kali + Playwright tools over MCP, ATT&CK as the Orient language.
Python ai-agentsautonomous-agentskali-linuxllm-agentsoffensive-securityooda-looppentestingplaywrightsecurity
# =============================================================================
# docker-compose.target.yml — DELIBERATELY VULNERABLE web target for testing.
# =============================================================================
# OPTIONAL override for test scenarios of the `web-pentest` skill. It is NOT part
# of the base stack (docker-compose.yml): bring it up explicitly alongside it to have a
# target to practice against without going out to the internet or touching real systems.
#
# docker compose -f docker-compose.yml -f docker-compose.target.yml up -d
#
# The target lives on the SAME internal compose network, so Hermes/Kali reach it by
# its DNS name `juice-shop:3000`. NO port is published to the host: the target only
# exists inside the docker network, scoping the exercise to the controlled environment.
#
# OWASP Juice Shop: a deliberately insecure web application (OWASP project, MIT), the
# de facto standard for practicing web pentesting (XSS, SQLi, auth, etc.).
services:
juice-shop:
image: bkimminich/juice-shop
# NO ports section: the target is NOT published to the host; it is only reachable over the
# internal docker network (Hermes sees it as http://juice-shop:3000). Keeps the exercise
# confined to the controlled environment.
restart: unless-stopped
# Healthcheck: the root responds 200 when the app is ready. The image ships node but
# not guaranteed curl/wget, so node (always present) is used for the probe.
healthcheck:
test: ["CMD", "node", "-e", "require('http').get('http://127.0.0.1:3000/',r=>process.exit(r.statusCode===200?0:1)).on('error',()=>process.exit(1))"]
interval: 10s
timeout: 5s
retries: 12
start_period: 60s