#2525th of 25 in Security

pentest-ai-agents

Claude Code subagents for pentesting, from recon to reporting

Stars
2.5k
License
MIT
Last commit
Aug 2026
Last release
Aug 2026
Language
Shell
Stars, last 29 weeks+2,215
25 Apr 2026: 243 stars2 May 2026: 791 stars9 May 2026: 982 stars16 May 2026: 1,351 stars23 May 2026: 1,452 stars30 May 2026: 1,614 stars6 Jun 2026: 1,723 stars13 Jun 2026: 1,771 stars20 Jun 2026: 1,811 stars27 Jun 2026: 1,852 stars4 Jul 2026: 1,923 stars11 Jul 2026: 1,962 stars18 Jul 2026: 1,995 stars25 Jul 2026: 2,021 stars1 Aug 2026: 2,049 stars8 Aug 2026: 2,081 stars15 Aug 2026: 2,118 stars22 Aug 2026: 2,135 stars29 Aug 2026: 2,167 stars5 Sept 2026: 2,193 stars12 Sept 2026: 2,221 stars19 Sept 2026: 2,246 stars26 Sept 2026: 2,278 stars3 Oct 2026: 2,307 stars10 Oct 2026: 2,328 stars17 Oct 2026: 2,352 stars24 Oct 2026: 2,388 stars31 Oct 2026: 2,407 stars7 Nov 2026: 2,438 stars14 Nov 2026: 2,458 stars

243 on 25 Apr 20262,458 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

A set of Claude Code subagent definition files, each covering one pentest area such as recon, web, Active Directory, cloud, mobile, C2, reverse engineering, detection engineering and reporting. Claude routes a task to the matching agent. Execution-capable Tier 2 agents require a declared scope and a scope-guard block. It installs by shell script into ~/.claude/agents or as a Claude Code plugin, and adds a SQLite findings database and slash commands.

Who it is for: Pentesters and red teamers using Claude Code on authorized engagements

Strengths

  • Covers recon, exploitation, post-exploitation, detection and reporting with distinct agents
  • Tier 2 agents need a declared scope; CI checks the scope-guard block is present
  • Installs by copying files or as a Claude Code plugin; no servers or Python deps
  • Offensive agents pair techniques with detection content; includes SQLite findings DB

Weaknesses

  • Requires Claude Code; agents are prompt files, not a standalone runtime
  • Underlying tools (nmap, nuclei, sqlmap, etc.) must be installed separately
  • README says both 50 and 52 agents, so the count is inconsistent
  • Includes C2, evasion and payload agents that are easy to misuse outside authorized work

What it needs

  • no GPU
  • Docker
  • Needs Claude Code, nmap, nuclei, sqlmap and other CLI pentest tools (optional, via install.sh --tools), SQLite
  • Models: Claude

What changed

  1. v3.4.0 ↗ (opens in a new tab)Security fix

    “poc-validator proves a bug is real before it reaches a report. Nothing covered the other end: someone says it's fixed, and that claim goes untested. That's how the same vulnerability gets reported tw…”

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in Security

See all 25
Also in Security
RankProjectScore
1shannonAutonomous pentester that exploits web apps and APIs using source analysis52k starsAGPL-3.092 out of 100
2ifixaiPython CLI that audits AI agents for policy, access and reliability failures46.3k starsApache-2.086 out of 100
3StrixAutonomous multi-agent pentesting CLI that validates vulnerabilities with working PoCs76.4k stars+7,005 (+10%) in 30 daysApache-2.085 out of 100
#4ai-infra-guardRed-teaming platform for AI infra CVEs, MCP servers, agent skills and jailbreaks7.5k starsApache-2.080 out of 100
#5failproofaiPolicy enforcement and run tracing for coding agent CLIs and gateways9.6k stars+3,964 (+70%) in 30 dayscustom license79 out of 100