PentestCode is a terminal-based security testing agent, built as a hard fork of OpenCode with the code-editing focus removed and replaced by offensive-security tooling. You point it at an authorized target, and it runs the relevant command-line tools, reads their output, updates its model of the network, and decides the next step. It ships as a single self-contained binary for Linux and macOS on x64 and arm64.
The design uses a team of agents rather than one prompt. A lead agent plans the engagement and dispatches specialist subagents in parallel, 13 in all, covering reconnaissance, scanning, enumeration, exploitation, identity, infrastructure, web application, post-exploitation, exploit development, a false-positive critic, and reporting, plus hidden agents for context compression and session management. The structure follows the strategist-coordinator model from HPTSA research.
All agents read from and write to one structured engagement state that records hosts and services, vulnerabilities with an evidence chain and confidence score, credentials, access levels, an entity-relationship graph, an Active Directory domain model, network segments, and cost-based attack paths. The state persists across sessions, and a human-readable findings file logs each result with a timestamp. It connects to 20 or more model providers through the AI SDK.
Features
- Multi-agent team: a lead agent dispatches 13 specialist subagents, each with its own prompt and tool permissions
- Shared engagement state: one structured store of hosts, vulnerabilities, credentials, access, relationships, and AD model that all agents read and write
- Persistent sessions: state survives closing the terminal and the agent resumes where it stopped
- Parser tools: 18 built-in tools that parse output from nmap, Nuclei, NetExec, Gobuster, SharpHound, and sqlmap into engagement state
- Attack-path finding: cost-based routing through the relationship graph with k-shortest paths
- Knowledge skills: 19 markdown knowledge packs (phase checklists, per-service notes, playbooks) loaded on demand, extensible by adding a SKILL.md
- Run modes: auto, free, and guided modes combined with pause behavior to control how much autonomy the agent has
- Reporting: slash-command dashboards for hosts, vulnerabilities, and credentials, plus Markdown and JSON report generation
- Provider choice: works with 20 or more model providers through the AI SDK, including Anthropic, OpenAI, Google, Azure, AWS Bedrock, and Ollama
