penetration-tester

name	penetration-tester
description	Use when the user explicitly requests a penetration test (pentest, ethical hacking, active exploitation, red-team assessment) against an authorized in-scope target — and only then. Runs a methodology-driven offensive assessment that validates exploitability with safe proofs of concept and delivers a risk-rated remediation report. Does not run on a normal code-review, security-review, or resolve-issue pass.
license	MIT
metadata	{"author":"Petr Král (pekral.cz)"}

Penetration Tester

Constraints

• Apply @rules/security/backend.md and @rules/security/frontend.md (and @rules/security/mobile.md when the target ships a mobile app)
• If the target uses Laravel, also apply @rules/laravel/laravel.mdc
• Apply @rules/reports/general.mdc for the delivered report
• Read-only / non-destructive — never modify, stage, commit, or push code; never run a destructive or service-disrupting payload. Output is the assessment report only.
• Authorization is mandatory — never test a target without explicit, confirmed authorization and a defined scope. Refuse and stop when scope or authorization is missing.
• Keep every proof of concept minimal and safe; demonstrate impact without causing damage, exfiltrating real data, or persisting access
• Never reveal or store harvested secrets, tokens, or PII in cleartext in the report — redact them
• Hard limits: this file stays <= 500 lines and <= 5000 tokens

Use when

• The user explicitly asks for a penetration test, pentest, ethical-hacking engagement, active exploitation, or red-team assessment
• An authorized, in-scope target (web app, API, network surface, or infrastructure) is named and the rules of engagement are known
• The goal is to validate real exploitability and demonstrate impact — not to perform a routine diff review

Do not use when

• The request is a normal change-set review → use @skills/security-review/SKILL.md
• The request is to find unknown reportable bugs for disclosure / bounty → use @skills/security-bounty-hunter/SKILL.md
• The request is to remediate a known CVE / advisory → use @skills/security-threat-analysis/SKILL.md
• No explicit pentest request, authorization, or scope exists → refuse and ask for scope and authorization first

How this differs from neighbors

• @skills/security-review/SKILL.md reviews a diff against best practices (passive, runs every CR); this skill actively validates exploitability only on explicit, authorized request.
• @skills/security-bounty-hunter/SKILL.md hunts unknown reportable bugs across the reachable surface; this skill runs a scoped, methodology-driven engagement with rules of engagement and a phased report.
• @skills/security-threat-analysis/SKILL.md remediates a known advisory; this skill discovers and proves impact.

Pre-engagement gate (blocking)

Before any active testing, confirm all of the following. If any is missing, stop and ask the user — never proceed on assumption.

1. Explicit request — the user actually asked for a penetration test (not a review or audit).
2. Authorization — written / confirmed permission to test the named target.
3. Scope — the exact in-scope assets (domains, hosts, repos, API endpoints) and the explicit exclusions.
4. Rules of engagement — testing window, allowed intensity, data-handling rules, and an emergency contact / stop condition.
5. Success criteria — what a complete engagement looks like for this request.

Record the confirmed scope and authorization at the top of the report. Refuse any target outside the stated scope.

Methodology

Run the engagement in ordered phases. Start low-impact, escalate carefully, document everything, and never step outside scope.

1. Reconnaissance (passive first)

• Map the attack surface from in-scope evidence only: routes/endpoints, technology fingerprint, exposed services, version markers.
• For a code-accessible target, enumerate entry points: routes (routes/*.php, php artisan route:list), controllers, FormRequests, Livewire/Filament actions, queued jobs, webhooks, API resources.
• Do not enumerate or probe anything outside the authorized scope.

2. Vulnerability identification

Systematically check the high-value classes (OWASP Top 10 oriented):

• Injection — SQL (whereRaw/DB::raw/DB::statement with interpolation), command (exec/shell_exec/proc_open/Process::run), template, header.
• Broken access control — IDOR / BOLA on route-model binding, missing policy/gate checks, privilege escalation, mass-assignment to privileged columns.
• Authentication & session — auth bypass, weak session handling, token leakage, missing rate limiting on credential endpoints.
• SSRF & external interaction — user-controlled URLs reaching Http::* / Guzzle / file_get_contents, missing allowlists, internal-IP / metadata reachability.
• Unsafe deserialization — unserialize($userInput), unsigned cookie/cache payloads, untrusted ShouldQueue sources.
• XSS — {!! $userInput !!}, Alpine x-html, raw JSON in <script>.
• Security misconfiguration & data exposure — verbose errors leaking stack traces / paths / DB identifiers, unsafe upload handling, path traversal.

3. Exploit validation (safe PoC)

• For each candidate, build the smallest safe proof of concept that proves impact (a single request or short script).
• Confirm the input is genuinely attacker-controlled all the way to the sink.
• Never run a payload that destroys data, disrupts service, persists access, or exfiltrates real user data — demonstrate reachability and impact, then stop.
• Stay strictly within the authorized scope and testing window.

4. Impact & risk assessment

• For each validated finding, assess likelihood and business impact, then assign a severity.
• Distinguish validated exploits from theoretical or environmental issues and from false positives.

5. Remediation & reporting

• Provide a concrete, actionable fix for every finding, complying with @rules/security/backend.md, @rules/php/core-standards.mdc, and (for Laravel) @rules/laravel/architecture.mdc.
• Separate quick wins from strategic / architectural fixes.

Severity

• Critical — directly exploitable for RCE, auth bypass, mass data access, or full compromise.
• High — exploitable with meaningful impact under realistic conditions.
• Medium — exploitable but constrained, or requires uncommon preconditions.
• Low / Informational — hardening gaps and best-practice deviations with no validated exploit.

Report

Deliver a single Markdown report with these sections:

Engagement summary

• Confirmed scope, authorization reference, testing window, and exclusions
• Counts: systems/surfaces tested, findings by severity, exploits validated

Findings

One block per finding, ordered by severity:

• Severity and category (OWASP / CWE)
• Location — file + line, endpoint, or host (in scope only)
• Description — what the vulnerability is and why it matters
• Proof of Concept — minimal safe reproduction (redact secrets, tokens, PII)
• Impact — what an attacker achieves
• Remediation — minimal corrected approach / code that closes the issue

Risk-rated remediation roadmap

• Quick wins, then strategic / architectural fixes, in priority order

Quality gate — before delivering

• Authorization and scope were confirmed before any active testing
• Every reported finding has a validated, safe PoC — no theoretical noise reported as exploited
• No destructive, persistent, or service-disrupting action was taken
• Nothing outside the authorized scope was tested
• Secrets, tokens, and PII are redacted in the report
• No code was modified and no unauthorized exploit was run

Done when

• The pre-engagement gate passed (or the skill refused for missing authorization / scope)
• The methodology phases ran in order within the authorized scope
• Each finding follows the Findings structure and passes every quality-gate item
• The report includes the engagement summary and a risk-rated remediation roadmap

Related skills

• @skills/security-review/SKILL.md — best-practices review of a change set (passive, every CR)
• @skills/security-bounty-hunter/SKILL.md — hunt unknown reportable bugs for disclosure
• @skills/security-threat-analysis/SKILL.md — remediate a referenced advisory / CVE