AI NewsOpen sourceAnnouncement
Claude-BugHunter turns Claude Code into a bug bounty hunter
Sachin Sharma's Claude-BugHunter adds 83 bug-hunting skills and 15 slash commands to Claude Code.
Image: GitHub
Why it mattersA bounty hunter who already uses Claude Code can install one bundle and get a scoped workflow with validation gates. They no longer have to hand-prompt every engagement.
Bug bounty hunters who already use Claude Code have had to re-explain the same hunting workflow every engagement. Sachin Sharma's Claude-BugHunter packages that workflow as a skill pack. The agent arrives knowing the job. The bundle holds 83 skills, 15 slash commands and 681 disclosed-report patterns. The repository reached rank 19 on Trendshift's daily list. Sharma created it in May 2026 and it now has 6,377 stars.
The author calls it a Claude Code skill pack for bug hunting and external red-team work. The 83 skills sit in five groups. 58 cover web application vulnerability classes. 10 cover enterprise platform attacks. 6 cover reporting and validation. 5 cover recon and OSINT. 4 cover methodology and mindset. The 681 patterns run across 24 core vulnerability classes, and 433 of them are individually cited to disclosed reports.
What it does differently
The bundle enforces a six-phase workflow: recon, then map and rank, then hunt, then validate, then report. The author says the phases are non-linear, so a hunter can revisit any step. Four of the fifteen slash commands sit at the top of that list: /hunt, /recon and /report. The README lists the rest.
The README also names the kinds of offensive work that are out of scope on purpose. A hunter will not pick this up for the wrong engagement. Out of scope: internal Active Directory attacks, command-and-control frameworks, post-exploit persistence, evasion techniques, iOS or hardware or ICS work, and binary exploitation.
A triage-validation skill carries a seven-question gate the author says must pass before a finding is submitted. Two of its questions check the asset is in the program's scope. Two check the finding matches the program's accepted impact. Sharma says the engagement engine enforces scope in code. Skills load from plain-English descriptions of the target. A hunter does not have to invoke each skill by hand.
Install and licence
The bundle installs as a Claude Code plugin from the marketplace. Skills are namespaced under claude-bughunter:. A copy install also works. Clone the repository and run the install script for the host operating system. An --all flag covers Claude Code, OpenCode, OpenAI's Codex CLI, Hermes Agent and Google AntiGravity in one pass.
The code is MIT licensed. The content is CC BY 4.0. The author defines the content as skills, methodology, wordlists, rubrics and documentation. Commercial reuse requires attribution to "Sachin Sharma / Claude-BugHunter" with a link to the repository.
The README states that skills must only be used against assets the hunter owns or has written permission to test. Users must also enroll in Anthropic's free Cyber Verification Program for authorised offensive work. Claude Code's runtime blocks vulnerability exploitation by default.
The repository names a foundational dependency on shuvonsec's earlier claude-bug-bounty project. It was last pushed on 15 November 2026 and is not archived.
A security researcher who bills by the hour can now measure one thing. Does a pre-written skill pack and a scope gate close reports faster than their own prompt library did? The answer depends on which vulnerability classes they already cover well. The bundle is most useful to a hunter who works web applications and the named enterprise platforms. That is where the catalogue sits.
Source
- Repository: elementalsouls/Claude-BugHunter
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
