recommend · 0xsteph — Claude Code command: route pentest tasks to the right agent
recommend turns a freeform pentest task into a routing decision: names one or two specialist agents and lists the next commands. Useful when you know the goal but not which agent to pick.
MIT ★ 2,482 updated 2026-10-06
.claude/commands/recommend.mdWhat recommend does
The command reads a task description and classifies it by domain: recon, web, AD, cloud, mobile, wireless, social engineering, malware, RE, forensics, detection, planning, reporting, CTF. It then picks a primary agent from the catalog and at most one hand-off agent.
It restates scope, environment type and rules of engagement, or asks a single question if scope is missing. The output includes 3–5 runnable commands with placeholders such as nmap -sV --top-ports 1000 <target>, plus one sentence on a pitfall or pivot signal.
The Claude Code install path is .claude/commands/recommend.md. Requests for mass targeting, unauthorized DDoS or attacks on third parties are refused outright.
What it helps with
- classify a pentest task by domain
- pick the best agent plus one hand-off
- produce 3–5 concrete next commands
- ask for the missing scope
- refuse out-of-scope requests
Example prompts for the agent
/recommendmap the external perimeter and tell me what to run next/recommendI have AD access and need a privilege escalation plan/recommendroute this SQL injection finding to the right agent
How to install recommend
Manually: put the files in place
.claude/commands/recommend.mdCall it with /recommend. For all projects use ~/.claude/commands/
Manually: put the files in place
~/.codex/prompts/recommend.mdCall it with /prompts:recommend. Codex marks this format deprecated in favour of skills
Manually: put the files in place
.github/prompts/recommend.prompt.mdManually: put the files in place
.gemini/commands/recommend.tomlGemini stores commands as TOML: the prompt text goes into the prompt field
Questions about recommend
How do I install recommend in Claude Code?
Place it at .claude/commands/recommend.md. Claude Code will expose it as the /recommend command.
What does recommend do?
It classifies a freeform pentest task, picks one or two agents from the pentest-ai-agents catalog, and gives commands for the next steps.
What happens with out-of-scope requests?
Mass targeting, unauthorized DDoS, and attacks on third parties are refused: the command stops instead of inventing authorization.
Source text
recommend.md · MIT
--- description: Recommend the right pentest-ai agent and concrete next steps for a freeform task description. argument-hint: <freeform task description> --- You are routing a pentesting task to the right specialist agent. The user's task is below. User task: $ARGUMENTS Do this in order: 1. **Classify the task.** Identify the primary domain (recon, web, AD, cloud, mobile, wireless, social engineering, malware, RE, forensics, detection, planning, reporting, CTF) and any cross-cutting concerns. 2. **Pick the agent (or two).** From the catalog below, name the single best agent. If a second agent would meaningfully help, name it as a hand-off. Don't list more than two. 3. **State the assumed scope.** Restate the in-scope assets, environment type (lab, internal, external, bug bounty, CTF), and any rules of engagement constraints visible in the request. If scope is missing, ask one direct question to fill the gap. 4. **Give 3–5 concrete next commands.** Real CLI invocations with placeholders the user can fill (`<target>`, `<wordlist>`). Not methodology paragraphs. Not "consider running nmap" — just `nmap -sV --top-ports 1000 <target>`. 5. **Note one thing to watch for.** A common pitfall, OPSEC consideration, or pivot signal that's specific to this task. One sentence. Agent catalog (for routing): - engagement-planner — pentest scoping, ROE, MITRE-mapped phased plans - recon-advisor — Nmap/Nessus/BloodHound output analysis, target prioritization (Tier 2) - osint-collector — domain recon, email harvest, breach data, social profiling - exploit-guide — methodology for AD/web/cloud/post-ex (advisory) - privesc-advisor — Linux/Windows privilege escalation - cloud-security — AWS/Azure/GCP, IAM escalation, container escape, serverless - api-security — REST/GraphQL/WebSocket, JWT, OAuth, OWASP API Top 10 - mobile-pentester — Android/iOS, Frida, Objection, MASTG/MASVS - wireless-pentester — WPA/WPA2/WPA3, evil twin, 802.1X, Bluetooth - social-engineer — phishing strategy, pretexting, vishing methodology - phishing-operator — Evilginx, GoPhish, dnstwist, live campaign tooling - vuln-scanner — Nuclei, Nikto, Nmap NSE, RouterSploit (Tier 2) - web-hunter — ffuf, gobuster, sqlmap, dalfox, Commix (Tier 2) - credential-tester — Hydra, Hashcat, John, CrackMapExec, wordlist generation - attack-planner — multi-step attack chain construction with stealth/impact scoring - bug-bounty — HackerOne/Bugcrowd methodology, dedup, report writing - ad-attacker — BloodHound, Impacket, NetExec, Certipy, Kerberos abuse (Tier 2) - exploit-chainer — chain low-severity findings into full compromise (Tier 2) - poc-validator — generate and safely run PoC scripts to eliminate false positives (Tier 2) - payload-crafter — msfvenom, Donut, custom loaders, paired with detection content - reverse-engineer — Ghidra, JadX, Radare2, Binwalk, static analysis workflows - swarm-orchestrator — coordinates multiple agents in parallel - bizlogic-hunter — price manipulation, race conditions, authorization flaws (Tier 2) - cicd-redteam — GitHub Actions, GitLab CI, Jenkins pipeline security - detection-engineer — Sigma, Splunk SPL, Elastic KQL, Sentinel KQL rules - threat-modeler — STRIDE/DREAD, attack trees, data flow diagrams - forensics-analyst — evidence acquisition, memory/disk forensics, timeline - malware-analyst — triage, dynamic analysis, IOC extraction, YARA - stig-analyst — DISA STIG, GPO remediation, keep-open justifications - report-generator — pentest reports, executive summaries, CVSS, remediation - ctf-solver — HackTheBox, TryHackMe, PicoCTF, web/pwn/rev/crypto/forensics Output format: ``` **Primary agent**: <name> **Hand-off to (if any)**: <name or "none"> **Assumed scope**: <one sentence> **Next commands**: 1. <command> 2. <command> 3. <command> **Watch for**: <one sentence> ``` If the user's task is out of scope (mass targeting, unauthorized DDoS, attacks on third parties), say so directly and refuse to route. Do not invent fake authorization context.
© 0xSteph · MIT · 0xSteph/pentest-ai-agents
Similar slash commands
All slash commands for Security →recommend by other authors comes first.