elementalsouls/Claude-BugHunter

What it is

Claude-BugHunter packages bug-hunting and external red-team tradecraft as installable Claude Code skills. After installation, describing the target in plain English causes the matching skill to load automatically, without calling a command by name. The bundle advertises 83 skills and 15 slash commands, organized into four layers: methodology and red-team mindset, 58 per-class hunt-* web vulnerability skills drawn from disclosed HackerOne reports, 10 enterprise platform attack skills covering M365/Entra, Okta, vCenter, SSL-VPN appliances and cloud IAM, and triage, reporting and evidence-hygiene skills.

The scope is deliberately the external attack surface: bug bounty programs, web application pentests, and initial-access work against internet-facing enterprise assets. Internal Active Directory attacks, C2 frameworks, post-exploitation and evasion are explicitly excluded. It targets authorized testers, not general developers.

Installation

The recommended route is the Claude Code plugin marketplace:

/plugin marketplace add elementalsouls/Claude-BugHunter
/plugin install claude-bughunter@elementalsouls

This loads the 83 skills and 15 commands under the claude-bughunter: namespace. For a pinned clone or systems without the plugin system, clone the repository and run the copy installer from the project root: bash scripts/install.sh on macOS/Linux, or pwsh ./scripts/install.ps1 on Windows. The copy installer also wires the hunt engagement scaffolder, which is not available through the plugin path. The separate cbh CLI installs with pipx install git+https://github.com/elementalsouls/Claude-BugHunter.

Skills are plain SKILL.md files, and scripts/install.sh --all (or -All in PowerShell) copies them to the skill directories of detected harnesses, including OpenCode, OpenAI Codex CLI, Hermes Agent and Google AntiGravity; the Burp MCP server can be wired in with --burp-mcp (-BurpMcp). The slash commands and /hunt engine remain Claude-Code-only.

Working through an engagement

The documented workflow is six phases: recon, map and rank, hunt, validate, report. /hunt creates an engagement folder and tracks state, while the engine/ directory maps a target's attack surface to the relevant skills. Before anything is submitted, a 7-Question Gate in the triage skill checks scope and accepted-impact status. A worked example and decision tree are in the usage guide.

Constraints

Use is limited to assets you own or have written authorization to assess. The skills surface scope and impact questions automatically, but they do not verify authorization themselves. Anthropic's cyber safeguards can block vulnerability-exploitation requests even in authorized engagements; the README points to the free Cyber Verification Program as the supported route for legitimate dual-use work. Separately, higher-risk cyber requests on Opus 5 may fall back to Opus 4.8 rather than being refused. External tooling such as Burp, ProjectDiscovery tools and wordlists is listed as part of the setup, but installing it is not covered by this bundle.

Code files are MIT-licensed; documentation, wordlists, regex catalogs and rubrics are CC BY 4.0 with attribution. Eight skills and the slash commands are vendored from shuvonsec/claude-bug-bounty, and the repo ships a NOTICE with upstream attribution.

HotRepo's take

The only recorded snapshot is 2026-11-14, when it placed 7th on the daily Python list after gaining 127 stars that day, bringing the repo to 5,794 stars and 890 forks. There is no all-languages daily rank or longer time series, so this shows one strong day on the Python list rather than a sustained trend. The bundle's value depends on whether the curated per-class patterns and enterprise chains actually match the reader's engagements; the 681-report claim and 433 cited reports are the author's counts and were not independently checked here.