mr-onion

Phase 35 — Security Testing: Tool & Capability Research

Status: Research / pre-decision (nothing locked) Date: 2026-10-13 Context: Feeds Phase 35 (“Comprehensive testing”), security area — the first Comprehensive area by the locked area order (security → SEO → highest-risk areas). Author model: Drafted in chat under Emil’s direction; implementation and commits to route through Claude Code per standing workflow rule.


1. Purpose

Emil compiled a list of ~26 pentesting / OSINT / red-team tools (from various videos) and asked whether TestForge could call them or replicate their functionality during testing. This document evaluates that list against TestForge’s actual scope, recommends what to build, flags what to exclude and why, and proposes a build order and design constraints.

2. Scoping principle (the thing that decides everything)

TestForge is a tool you point at a website to map paths and run reproducible suites. Its security area is therefore authorised web-application security testing (DAST) — not infrastructure pentesting, not OSINT, not red-teaming.

Two consequences drive every verdict below:

Anything that crosses from testing into exploitation (gaining access, cracking credentials, escalating privilege) is excluded regardless of technical feasibility — it changes the product’s legal and ethical footprint entirely.

3. Evaluation of the source list

Name corrections applied: cnesys → Censys, “exif metadat” → ExifTool, bloohound → BloodHound, mao → nmap. “grams” dropped (defunct dark-web market search, ~2017).

Verdict key: ✅ core · 🔧 helper/replicate · ⚠️ optional/gated · ⛔ out of scope

Tool What it is Integration path Verdict
nuclei Template-based vuln scanner, severity-tagged, large community template set CLI, JSONL, rate-limitable, tag/severity scoping ✅ Core — highest-value single add
ffuf Fast web content/parameter fuzzer CLI, clean JSON ✅ Core
gobuster Directory/DNS/vhost brute-forcing CLI ✅ Core (overlaps ffuf — one as primary)
ExifTool File/image metadata extraction CLI, JSON 🔧 Helper (info-leakage checks), low risk
CyberChef Data transform swiss-army knife (decode/encode/hash/JWT) Node library, embed in-process 🔧 Helper, low risk
FOCA Document-metadata harvester Windows GUI, not automatable 🔧 Replicate useful part with ExifTool/parsers; don’t call
nmap Port/service/version scanner + NSE scripts CLI -oX, parse via python-libnmap ⚠️ Optional, Comprehensive-only — active, infra-focused; narrow web-app value
Burp Suite Intercepting proxy + scanner Community=GUI only; Pro=Java ext API; Enterprise=REST/GraphQL (paid) ⚠️ Gated — only Enterprise cleanly automatable
Caido Lighter modern Burp alternative Scriptable workflows + plugin SDK, interaction-centric ⚠️ Gated/partial
Shodan Internet-exposed-service search (passive, 3rd-party DB) REST/Python API, paid tiers ⚠️ Optional passive recon, scope-adjacent
Censys Internet asset/cert search (passive, 3rd-party DB) REST API, free + paid ⚠️ Optional passive recon, scope-adjacent
SpiderFoot OSINT automation framework Web UI + REST API ⛔ Recon platform — scope creep
recon-ng Modular recon framework Interactive; resource-file scriptable ⛔ Recon platform — scope creep
bbot Modern async recon engine CLI + Python API, NDJSON ⛔ Recon platform — scope creep
Ahmia Clearnet search for Tor hidden services HTTP query ⛔ Irrelevant to web-app QA
Sherlock Username enumeration across sites CLI, JSON ⛔ People-targeting — privacy risk
PhoneInfoga Phone-number OSINT CLI + REST API ⛔ People-targeting
Metasploit Exploitation framework RPC API (automatable) ⛔ Exploitation ≠ QA (legal line)
Hydra Online login brute-forcer CLI ⛔ Intrusive; pointless vs lockout-exempt automation accounts
Hashcat Offline hash cracking CLI, needs hashes + GPU ⛔ No web-testing fit
LinPEAS / WinPEAS Local priv-esc enumeration Require a shell on target ⛔ Post-exploitation
BloodHound Active Directory attack-path mapping Needs domain creds + collector ⛔ Internal AD pentest
CloudFox AWS/Azure/GCP attack-surface enumeration Needs cloud credentials ⛔ Different product area
netcat Generic connect/listen/transfer primitive CLI ⛔ Trivially replaced by code; mostly an exploitation primitive

Net from the source list: the genuinely useful web-app pieces are nuclei, ffuf/gobuster, ExifTool, CyberChef, with nmap / Burp-Caido / Shodan-Censys as optional, mostly-paid or infra-adjacent extras. The majority of the list is full-scope pentest/OSINT tooling that belongs to a different product with a different authorisation model, not a feature of a web-app tester.

The source list was weighted toward popular red-team tooling and missing the automation-first DAST tools that actually fit a URL-driven tester.

Addition What it adds Integration Where it fits
OWASP ZAP Full DAST — passive + active web scanning Docker + REST API + Automation Framework YAML; or proxy-mode behind Playwright Core. Passive = Standard-safe; active = Comprehensive
Security-header + cookie checks CSP, HSTS, X-Frame-Options, Secure/HttpOnly/SameSite None — responses already captured via Playwright; pure code Standard-safe, zero new dependency
Retire.js Flags known-vulnerable JS libraries on loaded pages Node lib / CLI, runs on already-loaded pages Standard-safe, passive
testssl.sh / sslyze TLS/cert configuration grading CLI, JSON Standard-safe-ish (one connection)
sqlmap Automated SQL-injection testing CLI, scriptable Comprehensive-only, heavily gated; matches “break fields with bad data”
Nikto Quick web-server misconfig / known-file scan CLI, noisy, fast Comprehensive, optional (ZAP overlaps most)
SecLists Standard wordlist/payload corpus Data, not a tool — feeds ffuf/gobuster/nuclei Dependency for fuzzing tools

4.1 Highlight: the ZAP + Playwright proxy pattern

The strongest single idea in this research. TestForge already crawls, maps journeys (Phase 19) and tests the authenticated state (Phase 20). Run ZAP as a proxy behind Playwright: the browser drives the app exactly as it already does, ZAP passively scans every request/response the browser touches, then optionally active-scans.

Why it matters:

This is a differentiator, not just another scanner bolted on.

5. Structural backbone: OWASP ASVS

Rather than “we ran five scanners,” map every check/tool to the OWASP ASVS requirements it covers (~250 structured web-security requirements). This produces, for security, the same thing Phase 42 is building for functional testing: a security coverage map with honest gaps, indexed to drive test selection and allocation. It gives the security area a spine that matches TestForge’s existing “what’s tested / not tested / why” philosophy.

6. AI skills (the AI-native angle)

Scanners produce noise; triage is the hard part, and this is where TestForge’s AI layer earns its place.

7. Cross-cutting design constraints

These matter more than the individual tool picks:

  1. Passive, zero-dependency wins first: security-header + cookie checks, Retire.js. Pure code over responses already captured. Standard-safe.
  2. nuclei as the security engine — passive/safe templates by default, intrusive severities behind opt-in. Severity-tagged findings straight into the structured report.
  3. ZAP passive via the Playwright-proxy pattern — the key architectural piece.
  4. ffuf (or gobuster) for content discovery.
  5. testssl/sslyze for TLS grading (Standard candidate).
  6. ExifTool + CyberChef as embedded helpers.
  7. Comprehensive, gated: ZAP active scan, sqlmap, nmap (light: common ports + TLS/cert NSE), Nikto (optional).
  8. AI skills (triage, payloads) layered over the above.
  9. ASVS coverage map wrapping the whole area.

Note: nuclei ships network + SSL templates that overlap a light nmap pass and testssl, so sequence to avoid paying for the same coverage twice — nmap earns its place mainly for port/service discovery.

9. Explicitly out of scope (and why)

10. Open questions / decisions still needed

11. Suggested next tickets (draft — to confirm/lock, not locked here)

Proposed for Phase 35 security area, pending Emil’s approval and the usual decision-record + observable-acceptance-criteria treatment:


Research only. No code changes or commits made from this doc; implementation routes through Claude Code per standing workflow.