Dev Breakfast · 2026-11-16
Today's headline: Claude-BugHunter Offers a Vulnerability Hunting Skill Pack.
Claude-BugHunter Offers a Vulnerability Hunting Skill Pack
elementalsouls has made Claude-BugHunter available on GitHub, providing vulnerability hunting and external red-team workflows for Claude Code. The README lists 83 skills, 15 slash commands, and 681 patterns from disclosed reports covering 24 core vulnerability classes. However, the repository description still says 82 skills, so the two counts do not match. The credited author is Sachin Sharma; this is a third-party skill pack, not a built-in feature released by Anthropic.
On November 15, we covered parallel task status notifications in Claude Code Projects: the coordinator proactively reports completions, blockers, and pending decisions, filling a gap in progress feedback. Today's project addresses a different gap: organizing Claude Code tasks around security research methods. Installed as a plugin or by copying files, it loads relevant skills by testing topic and is separate from the earlier Projects update. The README divides the work into four layers: methodology, Web application testing, enterprise perimeter system testing, and report delivery. Of these, 58 hunt-* skills were compiled from disclosed HackerOne reports.
This is closer to claude-seo, which we covered on October 26. That project organizes SEO audits into 26 sub-skills and 19 specialized Agents; Claude-BugHunter instead packages detection patterns, bypass checklists, vulnerability chain templates, and reporting standards into skill files. Both add domain-specific workflows to general-purpose coding tools rather than release new models. Claude-BugHunter uses the SKILL.md format, and its README also lists compatible tools such as OpenCode and OpenAI Codex CLI, so its content is not tied exclusively to Claude Code.
The headline scale should not be taken as evidence of vulnerability discovery performance. The figure 681 refers to compiled report patterns, not vulnerabilities newly discovered by the tool; only 433 are marked as individually cited and auditable. The README claims validation through authorized red-team engagements and training platforms, and names DVWA and OWASP Juice Shop, among others, but provides no comparable discovery rates or false-positive rates. The conversation showing “47 hosts, 12 live targets” is also explicitly labeled as illustrative and cannot serve as evidence of real-world results. “Like a senior researcher after installation” is the project's promotional claim; the figures above do not substantiate it.
The repository currently offers two installation paths, but their scope differs: the plugin loads namespaced skills and commands without copying files into ~/.claude/; the terminal cbh runner requires separate installation, while task directory scaffolding comes with the repository clone. Copy-based installation writes skills and commands into the user's Claude configuration directory. The project also specifies personal information redaction, standardized evidence handling, and testing scope constraints, but the README has yet to provide quantitative validation of whether these workflow conventions can be executed consistently.
Sources:
This article selects one story for an in-depth look from 45 items on X / Hacker News / GitHub Trending over the past 24 hours. The content was generated with LLM assistance and fact-checked; background context draws partly on this site's previous coverage. Original source links are included at the end; cross-check before making important decisions.
Like this brief? Get it by email
Daily AI coding picks at 8:00, plus a hands-on field-notes issue every Saturday. Written in Chinese.
This page is auto-generated by LLM aggregation; please cross-check with original sources.