HackSkills — Offensive-Security Skills for AI Agents
A structured skill library that turns AI agents into practical hackers across 14 security domains with 101 deep-topic skills.
Evidence shows the repository is a knowledge base with no executable code or permission declarations, so all trust criteria score 0. No evidence of least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution.
Self-consistency: Directory structure and skill naming are consistent, and skills listed in README match directories, score 2. Dependency availability: Build depends on Yak engine but no version pinning or checksum, score 1. Failure messages: Workflows have error checks, but skill content lacks failure handling guidance, score 1.
Audience and scenarios: Clearly targets security researchers and penetration testers, covering multiple domains, score 2. Capability boundaries: No explicit limitations or applicable conditions, score 1. Trigger precision: Skill descriptions are vague, no precise trigger conditions, score 1. Environment fit: No OS or tool compatibility notes, score 1.
Information architecture: Clear directory structure with master and category entries, score 2. Install notes: Provides npx command but lacks detailed installation steps, score 1. Naming stability: Semantic naming but no naming convention documented, score 2. Examples and FAQ: Examples present but no FAQ, score 1. Known limitations: Not mentioned, score 0. License: MIT license, score 2. Versioning and changelog: Version numbers but no changelog, score 1. Maintenance responsibility: No explicit maintainer, score 1.
Output usability: Skills are markdown but no integration guidance, score 1. Marginal value: Provides extensive security knowledge, practical value, score 2. Cost benefit: Simple installation but no runtime cost assessment, score 1.
Claim traceability: README lists knowledge sources but no specific links, score 1. Cross-source corroboration: No independent verification, score 1. Fact-inference separation: No clear distinction between facts and inferences, score 1.
- Repository contains offensive security content, intended for authorized testing only; users must ensure legal compliance.
- No explicit permission management or user confirmation mechanisms; Agent integration must handle these.
- Depends on external Yak engine without version pinning or checksum, posing supply chain risk.
What does this agent do, and when should you use it?
HackSkills is a comprehensive, agent-oriented knowledge base for offensive security, offering structured, installable, and searchable skills for penetration testing, bug bounty, CTF, and authorized security research. The repository follows a layered routing design: a master entry (hack), six category entries (recon-for-sec, api-sec, auth-sec, injection-checking, file-access-vuln, business-logic-vuln), and 101 deep-topic skills spanning 14 security domains including web, API, authentication, OS privilege escalation, Active Directory, mobile, binary exploitation, reverse engineering, cryptography, blockchain, AI/ML security, and forensics. Each skill is a SKILL.md file in a uniform skills/{semantic-identifier}/SKILL.md structure, designed for on-demand loading. Content is distilled from public resources like PayloadsAllTheThings and HackTricks, emphasizing methodology and auditability. The project distributes via a static website (skills.hackbenchmark.com), the GitHub repo itself, and an encrypted offline ZIP (password public) to bypass AV scanning. Installation is via npx skills add or direct curl of SKILL.md files.
HackSkills provides agents with executable security testing playbooks. After installing, an agent loads the master entry hack to route based on the target, then switches to category entries like recon-for-sec for asset discovery, api-sec for API testing, or injection-checking for injection flaws. From there, deep skills such as sqli-sql-injection detail attack patterns, WAF bypass matrices, and detection techniques; ssrf-server-side-request-forgery covers cloud metadata exploits and DNS rebinding; active-directory-kerberos-attacks provides Kerberoasting and golden ticket methods. Each skill includes scenario files (SCENARIOS.md) for practical examples. The agent reads the SKILL.md to generate test cases, craft payloads, and validate findings, using time-based proofs (e.g., SLEEP) to avoid destructive actions. The online site offers fuzzy search and filtering, while the ZIP enables offline use. Outputs are test reports and PoCs suitable for authorized engagements.
- A penetration tester in an authorized engagement uses recon-for-sec for asset discovery, then sqli-sql-injection to confirm SQL injection points.
- A bug bounty hunter analyzing APIs applies api-sec skills to test for BOLA and JWT abuse.
- A red teamer moves laterally in a Windows domain using active-directory-kerberos-attacks and windows-privilege-escalation.
- A CTF competitor solves binary exploitation challenges with stack-overflow-and-rop and rsa-attack-techniques for crypto tasks.
- A blockchain auditor uses smart-contract-vulnerabilities to identify reentrancy and overflow in Solidity contracts.
- An AI security engineer evaluates LLM applications with llm-prompt-injection to detect prompt injection and RAG poisoning.
What are this agent's strengths and limitations?
- Structured three-layer routing (master, category, deep skills) enables efficient navigation and on-demand loading, avoiding information overload.
- Comprehensive domain coverage: 14 security domains and 101 skills, including web, API, auth, privesc, AD, mobile, pwn, RE, crypto, blockchain, AI security, and forensics.
- Distilled from authoritative sources (PayloadsAllTheThings, HackTricks, CTF-Wiki) into actionable methodologies, filtering out noise.
- Multiple delivery modes: website, GitHub source, and encrypted ZIP, supporting online search and offline air-gapped use.
- Adheres to non-destructive PoC policy, replacing data-destroying operations with time-delay proofs, aligned with ethical testing.
- Installation depends on npx and Node.js, which may not be available on all agent platforms.
- All content is in English, requiring translation or comprehension for non-English speakers.
- Many skills assume the ability to execute external tools (e.g., Frida, Volatility), requiring agent integration or manual execution.
- The offline ZIP password is a public constant, which may raise security concerns despite being for AV bypass only.
- The large number of skills (101) may consume significant storage or token budget if fully loaded.
How do you install or deploy this agent?
Requires Node.js for npx. Install with: npx skills add yaklang/hack-skills. Alternatively, fetch the master file directly: curl -fsSL https://raw.githubusercontent.com/yaklang/hack-skills/main/skills/hack/SKILL.md. For offline use, clone the repo: git clone https://github.com/yaklang/hack-skills.git. Encrypted ZIP: curl -fsSLO https://oss-qn.yaklang.com/hack-skills/latest/hack-skills.zip, then extract with 7z x -phack-skills hack-skills.zip (password is the public 'hack-skills').
How do you use this agent?
After install, start with the hack master entry to get routing guidance. Based on the target, select a category entry (e.g., api-sec for API endpoints). Then drill down into specific skills as needed, such as loading sqli-sql-injection to test SQL injection according to its method matrix and bypass techniques. Use the accompanying SCENARIOS.md files for practical examples. All skills are Markdown; the agent reads them and follows the instructions to perform tests, craft proofs, and adjust based on responses.
FAQ
Is HackSkills free to use?
Can I use it for unauthorized targets?
Does the content include harmful payloads?
How can I integrate these skills into my custom agent?
npx skills add if your platform supports it.