claude-security

Scan codebases and diffs for vulnerabilities and generate verified patch files.

1|Updated Jul 6, 2015
One-click install
npx skills add https://github.com/ksolomon/dotfiles --skill claude-security-ksolomon
Or copy as Structured Prompt for Agentβ–Ό
Please help me install this Agent Skill.
Skill: claude-security
Source: https://github.com/ksolomon/dotfiles/tree/main/AI/.claude/plugins/marketplaces/claude-plugins-official/plugins/claude-security/skills/claude-security
Command: npx skills add https://github.com/ksolomon/dotfiles --skill claude-security-ksolomon

SYSTEM DOCUMENTATION & REQUIREMENTS

πŸ’‘ This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Finding real vulnerabilities in a codebase requires deep review that manual audits and noisy SAST tools often miss, and turning findings into safe fixes is slow. This Skill runs a team of agents that maps the architecture, hunts for vulnerabilities, verifies every finding through an adversarial panel, and produces reviewable patch files. ## Core Features & Use Cases - Codebase scanning: Scan the whole repository or a scoped directory at low, medium, high, or max effort, with findings verified by a three-lens panel before reaching the report. - Change scanning: Scan a branch's diff, a pull request, or a single commit so small changes come back in minutes. - Patch suggestions: Turn confirmed findings into targeted .patch files, each developed in a scratch checkout, reviewed by an independent verifier, and validated with git apply --check β€” nothing is ever committed or pushed automatically. - Use Case: Before merging a feature branch, ask for a scan of the branch's changes; receive a timestamped report directory with a human-readable results file, machine-readable JSONL for CI gates, and optionally verified patch files you apply yourself. ## Quick Start Ask Claude to scan this repository for security vulnerabilities at medium effort and confirm when prompted.

Frequently Asked Questions about claude-security

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I scan my codebase for vulnerabilities with Claude?β–Ό

Invoke the claude-security skill and choose "Scan codebase" from the menu, or ask directly to scan the repository. You pick the scope and effort tier, confirm the run once, and receive a timestamped report directory with verified findings.

How to scan only the changes in a git branch or pull request?β–Ό

Choose "Scan changes" or pass a commit sha or base ref. The skill diffs your branch against its base (upstream, origin/main, or similar), sizes the change, and runs a proportionate scan β€” small diffs get a fast single-researcher pass that is still panel-verified.

Does the security scan apply fixes automatically?β–Ό

No. The suggest-patches job writes `.patch` files into the report's `patches/` folder after independent verification, but nothing is applied, committed, pushed, or opened as a pull request. You review each patch and apply it yourself with `git apply`.

Can I scan a pull request that is not checked out locally?β–Ό

Yes, when the GitHub CLI is granted and authenticated. The skill lists your open pull requests via `gh pr list`, and if the head branch is not local it offers to fetch it with your approval before scanning the range against its base.

Why did the scan refuse to start without a confirmation?β–Ό

Every scan requires an explicit "Yes" to a fixed cost confirmation unless your request already acknowledged the time or token cost in words. This guard prevents expensive unattended runs from starting without consent.

What are the limitations of agent-based security scanning?β–Ό

Scans are nondeterministic and derive findings from reading code only β€” no tests are executed and no exploits are fired. The skill complements SAST, dependency scanning, and code review rather than replacing them, and regular runs build coverage over time.