# Cybersecurity Skills: 818 security playbooks for AI assistants (mukul975/Anthropic-Cybersecurity-Skills) — skill card for AI agents

> AgentGid skill card. Checked on 2026-10-31 against commit 54a7988 (version 1.3.0).
> Every fact comes from the pack's own files at that commit. Numbers marked "as reported" are the author's.
> Human page: https://agentgid.com/skills/mukul975/anthropic-cybersecurity-skills/

## Explain it to a non-programmer
A library of 818 security playbooks for your AI assistant: investigating alerts, analysing malware, checking cloud setups, hunting for intruders, and also attack techniques used in authorised testing. Your assistant picks the playbook that matches your question and follows its steps. Almost every playbook comes with a small Python helper script the assistant can run.

## Who it is for
- You work in security (incident response, threat hunting, forensics, cloud or app security) and use an AI assistant
- You want your assistant to follow a step-by-step procedure, with tool names and checks, instead of guessing
- You run authorised penetration tests or red-team work and want written procedures for it
- Not for: You are not a security professional: many playbooks run real security tools against real systems
- Not for: You want only defensive content: the library also contains attack procedures

## What it is
- Kind: knowledge, scripts
- License: Apache-2.0
- Repository: https://github.com/mukul975/Anthropic-Cybersecurity-Skills

## What it does
- 818 skill folders, each with one SKILL.md: a description with "Use when" triggers, then When to Use, Prerequisites, Workflow and Verification
- Each skill is tagged with security frameworks (MITRE ATT&CK, NIST CSF 2.0, and where relevant ATLAS, D3FEND, NIST AI RMF, MITRE F3)
- Largest areas by the author's count: Cloud Security 66, SOC Operations 63, Threat Hunting 58, Threat Intelligence 52, Web Application Security 46, Network Security 43, Digital Forensics 41, Identity & Access 40, Malware Analysis 39, Red Teaming 35
- Smaller areas: containers, OT/ICS, API security, incident response, vulnerability management, penetration testing, DevSecOps, zero trust, endpoint, phishing defence, cryptography, AI security, mobile, ransomware, compliance, supply chain and more (34 areas in all)
- 1,098 Python and PowerShell helper scripts (scripts/agent.py and scripts/process.py in nearly every skill), plus reference notes and report templates
- Only the name and description load at first; the full playbook loads when it matches your request

## Skills
| Skill | Call | Changes code? | In plain words |
|---|---|---|---|
| performing-memory-forensics-with-volatility3 | plain language ("Analyse this memory dump ...") | no | Looks through a saved copy of a computer's memory for signs of malware or an intruder |
| triaging-security-incident | plain language ("Triage this alert ...") | no | Sorts a new security alert by type and urgency and says who should handle it |
| performing-malware-hash-enrichment-with-virustotal | plain language ("Check this hash on VirusTotal") | no | Looks up a suspicious file's fingerprint at VirusTotal to see whether it is known malware |
| detecting-business-email-compromise | plain language ("Is this email an impersonation?") | no | Helps spot emails that pretend to be your boss or a supplier to get money sent |
| auditing-mcp-servers-for-tool-poisoning | plain language ("Audit this MCP server") | no | Checks a tool server for your AI assistant for hidden instructions and unsafe access before you add it |
| building-patch-tuesday-response-process | plain language ("Help me plan this month's Microsoft patches") | no | Sets up a monthly routine for testing and rolling out Microsoft security updates |

### performing-memory-forensics-with-volatility3
- Asks for: A memory dump file (raw, ELF or crash dump)
- Asks for: The operating system symbol files
- Asks for: Disk space for the output
- Gives: Running and hidden processes
- Gives: Network connections
- Gives: Signs of injected code
- Gives: Findings mapped to ATT&CK
- Say: "Analyse this memory dump for signs of credential theft"

### triaging-security-incident
- Asks for: The alert or user report
- Asks for: Business context for impact
- Gives: Incident type
- Gives: Priority by business impact
- Gives: Which team it goes to
- Say: "Triage this EDR alert and tell me how urgent it is"

### performing-malware-hash-enrichment-with-virustotal
- Asks for: A file hash (MD5, SHA-1 or SHA-256)
- Asks for: Your VirusTotal access key
- Gives: Detection rates across engines
- Gives: Sandbox behaviour
- Gives: Related indicators
- Say: "Is this SHA-256 hash malicious? Check it on VirusTotal"

## Know before installing
- Not an Anthropic project. Despite the name, the README says it is an independent community project, not affiliated with Anthropic PBC.
- It contains attack procedures on purpose: red-team command-and-control, phishing simulation, exploitation and credential access sit next to the defensive playbooks. About 34 skill names start with "exploiting". The author says use them only on systems you own or have written permission to test.
- It is huge. 818 playbooks compete for your assistant's attention; installing the few folders you need keeps it focused.
- Playbooks tell your assistant to install and run real tools. Many workflows begin with pip install, git clone or wget steps, and the helper scripts can run commands on your machine.
- One emulation script runs attack-simulation commands through the shell. The Atomic Red Team helper executes test commands as shell scripts on the machine it runs on; run it only on a test system.
- Apache 2.0 license: free for personal and commercial use.

## Install (official commands)
- Any agent (skills CLI): `npx skills add mukul975/Anthropic-Cybersecurity-Skills` — Installs the library; you can also copy single folders from skills/ into your agent's skills folder.
- Git clone: `git clone https://github.com/mukul975/Anthropic-Cybersecurity-Skills.git` && `cd Anthropic-Cybersecurity-Skills` — Then point your agent at the skills/ folder.
- Claude Code (plugin marketplace): The repo ships .claude-plugin/marketplace.json with one plugin, cybersecurity-skills (version 1.3.0). The README gives no plugin install command.

## What runs on the user's machine
- Code on the machine: yes (when your assistant chooses to run a skill's helper script or the tool commands in a workflow)
- Network: yes
- Install: Markdown playbooks, references, templates and 1,098 scripts are copied. No hooks, nothing runs at install or at session start.
- A skill is used: The playbook may tell your assistant to install tools (pip, git clone, wget) and run them on files or systems you name.
- Helper scripts: About 345 scripts start other programs, more than 250 make web requests (VirusTotal, Microsoft Graph, NVD, abuse.ch, Shodan and others), about 70 open network sockets; about 140 read access keys from environment variables.
- Attack emulation: The Atomic Red Team helper runs test and cleanup commands through the shell.

## Examples (from the pack's own files)
- The format every playbook follows: output format defined in README.md

## Author's numbers (as reported, not run by AgentGid)
- Framework coverage across the 817 skills: ATT&CK 805, NIST CSF 2.0 804, D3FEND 139, NIST AI RMF 97, MITRE F3 94, ATLAS 93. Source: mukul975/Anthropic-Cybersecurity-Skills/README.md
- Each skill costs about 30 tokens to scan and 500 to 2,000 tokens to load fully. Source: mukul975/Anthropic-Cybersecurity-Skills/README.md

## Steps only the person can do
- Get written permission (rules of engagement) before any penetration test or red-team skill is used on a system (README.md) Then: The assistant follows the offensive playbook only against the systems in scope
- Create accounts and access keys at outside services a helper script needs (for example VirusTotal) and provide them to the script (skills/performing-malware-hash-enrichment-with-virustotal/SKILL.md) Then: The assistant runs the helper script with that key to query the service
- Approve installing the security tools a playbook lists (pip, go install, npm, git clone) (The Prerequisites section of each skill's SKILL.md) Then: The assistant installs the tool and runs the workflow steps

## Mistakes to avoid
- Not: Let the assistant use any playbook whose name looks close to the request. Instead: Read the description's "Do not use for" line and switch to the skill it names. Why: Many skills overlap, and the descriptions carry negative triggers that route to the right neighbour
- Not: Point an attack or exploitation playbook at any system the request mentions. Instead: Use offensive skills only on systems you own or have written permission to test. Why: The library mixes attack procedures with defensive ones on purpose
- Not: Run the Atomic Red Team helper on your everyday computer. Instead: Run attack-emulation scripts only on a disposable test machine. Why: The script passes test and cleanup commands straight to the shell
- Not: Feed a long hash list to the VirusTotal skill on a free key and expect quick results. Instead: Plan for the free-tier rate limit or use a premium key for bulk enrichment. Why: The free key allows only a few lookups per minute

## What leaves the machine
- File hashes you ask about → VirusTotal API, When the VirusTotal skill or its helper script runs. Opt out: Do not use VirusTotal skills; check hashes in an offline or self-hosted tool
- Suspicious files (malware samples) → Joe Sandbox Cloud, or a sandbox you host (Cuckoo), When the malware submission pipeline skill submits a sample. Opt out: Use only the self-hosted Cuckoo path
- Indicators (IPs, domains, hashes) and queries to threat-intel and cloud services such as Shodan, abuse.ch, NVD and Microsoft Graph → Those services' APIs, When a helper script that calls them runs. Opt out: Read the script before running it and skip skills that call outside services

## Lines in the pack aimed at agents (ask the person first)
- "curl -sSfL https://raw.githubusercontent.com/anchore/grype/main/install.sh | sh -s -- -b /usr/local/bin" (skills/scanning-container-images-with-grype/SKILL.md): Pipes a remote install script into the shell and writes to /usr/local/bin; ask the person before installing
- "go install -v github.com/projectdiscovery/subfinder/v2/cmd/subfinder@latest" (skills/performing-subdomain-enumeration-with-subfinder/SKILL.md): Installs an unpinned latest version of a global tool (several skills do the same with @latest); ask first and prefer a pinned version
- "npm install -g snyk" (skills/performing-sca-dependency-scanning-with-snyk/SKILL.md): Global CLI install; needs the person's approval

## May I use it like this?
- commercial-app: yes (Apache-2.0)
- client-work: yes (Apache-2.0)
- redistribute-files: conditions (Apache-2.0: keep the LICENSE (each skill folder has its own copy) and mark changed files)
- inside-app-bundle: conditions (Apache-2.0: ship the license text with the bundled skills)

## Which one to pick
- If You need step-by-step procedures across SOC, incident response, forensics, cloud, threat intel and red teaming: skill:mukul975/Anthropic-Cybersecurity-Skills
- If You want security research, vulnerability detection and code audit workflows: skill:trailofbits/skills
- If You only do web recon and bug hunting (CORS, XSS, SQLi, SSRF, subdomain takeover): skill:uphiago/recon-skills
- If You are solving CTF challenges (pwn, crypto, reverse engineering, forensics): skill:ljagiello/ctf-skills

## Works well with
- tool:semgrep: The implementing-semgrep-for-custom-sast-rules skill writes Semgrep rules and wires them into CI
- api:promptfoo: The continuous-llm-red-teaming-with-promptfoo skill runs Promptfoo red-team tests against LLM apps in CI

## Works in and needs
- Works in Claude Code, GitHub Copilot, Codex, Cursor, Gemini CLI: skills folder (npx skills add or git clone)
- Works in Claude Code: plugin (marketplace.json in the repo)
- Needs Python 3 and pip: Helper scripts are Python and many playbooks install Python tools
- Needs API keys for outside services (VirusTotal and others): Only for skills that query those services
- Needs The security tools each playbook names (Volatility 3: Workflows run them

## How to tell it works
Copy one folder, for example skills/triaging-security-incident, and ask your assistant to triage a sample alert. The answer names an incident type, a priority and a team to route it to, following the skill's steps.

## Safety
Not an official Anthropic project; install only from github.com/mukul975/Anthropic-Cybersecurity-Skills. The library includes attack procedures and scripts that run tools, send web requests and start shell commands. Read a skill and its script before letting your assistant run it, and use attack skills only on systems you are allowed to test.
