<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>CyberSib field notes</title>
    <link>https://www.cybersib.co/blog</link>
    <description>Notes on how we test, what public reports show, and how disclosure works.</description>
    <item>
        <title>How We Run a Penetration Test, From Scope to Retest</title>
        <link>https://www.cybersib.co/blog/how-we-run-a-penetration-test</link>
        <guid>https://www.cybersib.co/blog/how-we-run-a-penetration-test</guid>
        <pubDate>Fri, 20 Nov 2026 00:00:00 GMT</pubDate>
        <category>penetration-testing</category><category>methodology</category><category>access-control</category><category>reporting</category>
        <description>A walkthrough of a CyberSib penetration test: how we set scope, how we move from reconnaissance to privilege escalation, and what a finding has to contain before it reaches your report.</description>
        <content:encoded><![CDATA[
<p>A penetration test shows how an attacker gets past your security controls and how far they get once inside. This post walks through how we run one and what you receive at the end.</p>
<p>A penetration test shows an attacker getting past the security controls on your network, and how far that goes, including privileged access on critical systems. The useful part is the demonstration. Scanner output lists what might be wrong. A penetration test shows what an attacker can do with it, step by step, with the evidence.</p>
<h2><span class="mono">§ 1</span><a href="#the-engagement-at-a-glance">The engagement at a glance</a></h2>
<p>Every engagement follows the same five phases. The testing itself sits in the middle. The phases around it decide whether the results are useful to you.</p>
<figure><figcaption><span class="mono">Diagram</span></figcaption><div><span class="edgeLabel"></span></div><div><span class="edgeLabel"></span></div><div><span class="edgeLabel"></span></div><div><span class="edgeLabel"></span></div><div><span class="edgeLabel"><p>findings shared as they are confirmed</p></span></div><div><span class="nodeLabel"><p>01 Project kickoff</p></span></div><div><span class="nodeLabel"><p>02 Scoping and planning</p></span></div><div><span class="nodeLabel"><p>03 Security assessment</p></span></div><div><span class="nodeLabel"><p>04 Progress tracking</p></span></div><div><span class="nodeLabel"><p>05 Final delivery</p></span></div></figure>
<ol>
<li><strong>Project kickoff.</strong> We gather your security requirements, define scope and boundaries, allocate the team, and agree on a timeline.</li>
<li><strong>Scoping and planning.</strong> We analyze the technical environment, select the testing methodology, set up the risk rating framework, and prepare tooling.</li>
<li><strong>Security assessment.</strong> We combine vulnerability scanning with manual testing and evaluate each security control in scope.</li>
<li><strong>Progress tracking.</strong> You get weekly progress reports, documented vulnerabilities as we confirm them, updated risk ratings, and remediation recommendations.</li>
<li><strong>Final delivery.</strong> We hand over a detailed vulnerability report, an executive summary, a remediation strategy, and recommendations for future work.</li>
</ol>
<h2><span class="mono">§ 2</span><a href="#scope-is-a-security-control">Scope is a security control</a></h2>
<p>A weak penetration test usually fails in the scope, not in the testing. If the scope lists hostnames but leaves out the roles a user can hold, the tester never learns that a “viewer” can become an “admin.” If the scope excludes the export service because it is “internal,” the tester never tries to reach it from the outside.</p>
<p>At kickoff we write two documents with you:</p>
<ul>
<li><strong>A scope document</strong> that lists systems, environments, user roles, and test accounts for each role. For web and API targets, we ask for at least two accounts per role so we can test whether one user can reach another user’s data.</li>
<li><strong>Rules of engagement</strong> that cover testing windows, rate limits, systems that must not be disrupted, how to reach your on-call contact, and what we do if we find evidence of a prior compromise.</li>
</ul>
<h2><span class="mono">§ 3</span><a href="#testing-methodology-five-steps">Testing methodology: five steps</a></h2>
<p>Inside the security assessment phase, we work through five steps. Each step is documented, so the report shows exactly how we reached each finding.</p>
<ol>
<li><strong>Initial scoping and planning.</strong> We confirm what is reachable and build a test plan for each component.</li>
<li><strong>Reconnaissance and information gathering.</strong> We map the attack surface: hosts, services, API routes, client-side code, roles, and the trust relationships between components.</li>
<li><strong>Vulnerability identification.</strong> We test each component for known weakness classes and for flaws specific to how your application works.</li>
<li><strong>Exploitation and privilege escalation.</strong> We confirm a weakness by exploiting it within the agreed rules, then try to chain it toward higher privileges or more sensitive data.</li>
<li><strong>Post-exploitation and reporting.</strong> We record what an attacker could reach from each foothold and write the finding with the evidence that proves it.</li>
</ol>
<h2><span class="mono">§ 4</span><a href="#a-worked-example-privilege-escalation-through-a-role-parameter">A worked example: privilege escalation through a role parameter</a></h2>
<p>The following example is illustrative. It is not client data. It shows the type of finding that automated scanners rarely report and that a manual tester checks on every engagement.</p>
<p>A web portal lets users update their display name. The browser sends a <code>PATCH</code> request to the user API. The tester adds one field that the UI never sends:</p>
<figure><figcaption><span class="mono">http</span>Copy</figcaption><pre><code><span class="cl"><span class="ln">1</span><span class="token request-line"><span class="token method property">PATCH</span> <span class="token request-target url">/api/v2/users/4182</span> <span class="token http-version property">HTTP/1.1</span></span></span><span class="cl"><span class="ln">2</span><span class="token header"><span class="token header-name keyword">Host</span><span class="token punctuation">:</span> <span class="token header-value">portal.internal.example</span></span></span><span class="cl"><span class="ln">3</span><span class="token header"><span class="token header-name keyword">Cookie</span><span class="token punctuation">:</span> <span class="token header-value">sid=9f2c…; role=viewer</span></span></span><span class="cl"><span class="ln">4</span><span class="token header"><span class="token header-name keyword">Content-Type</span><span class="token punctuation">:</span> <span class="token header-value">application/json</span></span></span><span class="cl"><span class="ln">5</span></span><span class="cl"><span class="ln">6</span><span class="token application-json"><span class="token punctuation">{</span><span class="token string-property property">"display_name"</span><span class="token operator">:</span> <span class="token string">"j.doe"</span><span class="token punctuation">,</span> <span class="token string-property property">"role"</span><span class="token operator">:</span> <span class="token string">"admin"</span><span class="token punctuation">}</span></span></span></code></pre></figure>
<p>The server responds:</p>
<figure><figcaption><span class="mono">http</span>Copy</figcaption><pre><code><span class="cl"><span class="ln">1</span><span class="token response-status"><span class="token http-version property">HTTP/1.1</span> <span class="token status-code number">200</span> <span class="token reason-phrase string">OK</span></span></span><span class="cl"><span class="ln">2</span></span><span class="cl"><span class="ln">3</span>{"id": 4182, "role": "admin", "updated": true}</span></code></pre></figure>
<p>The API binds every field in the request body to the user record, including <code>role</code>. The tester now holds an admin session and requests the bulk export endpoint:</p>
<figure><figcaption><span class="mono">http</span>Copy</figcaption><pre><code><span class="cl"><span class="ln">1</span><span class="token request-line"><span class="token method property">GET</span> <span class="token request-target url">/admin/export?all=true</span> <span class="token http-version property">HTTP/1.1</span></span></span><span class="cl"><span class="ln">2</span><span class="token header"><span class="token header-name keyword">Host</span><span class="token punctuation">:</span> <span class="token header-value">portal.internal.example</span></span></span><span class="cl"><span class="ln">3</span><span class="token header"><span class="token header-name keyword">Cookie</span><span class="token punctuation">:</span> <span class="token header-value">sid=9f2c…</span></span></span></code></pre></figure>
<p>The response is an 18 MB CSV file of customer records.</p>
<figure><figcaption><span class="mono">Diagram</span></figcaption>Admin exportUser APITester (viewer account)Admin exportUser APITester (viewer account)PATCH /users/4182 {"role":"admin"}200 OK, role=adminGET /admin/export?all=true200 OK, full customer CSV</figure>
<p>A scanner sees two <code>200 OK</code> responses and moves on. A tester sees a viewer account that now owns the customer database. The report finding states the root cause (the update handler accepts fields the client should never control), the impact (any authenticated user can become an administrator), and the fix (an explicit allowlist of writable fields, plus a server-side authorization check on role changes).</p>
<h2><span class="mono">§ 5</span><a href="#why-we-spend-most-of-our-time-on-access-control">Why we spend most of our time on access control</a></h2>
<p>Access control flaws are common in public bug bounty data. The <a href="https://hexbrainai.com/browse">HexBrain</a> index of 15,244 publicly disclosed HackerOne reports lists 1,494 reports in its Access Control class, 464 of them tagged as privilege escalation. The same index lists 373 more reports as insecure direct object references (IDOR).</p>
<p>These bugs are hard for tools because the request is valid. Only someone who knows that user A must not see user B’s invoice can tell that the response is wrong. That is why we ask for multiple accounts per role during scoping.</p>
<h2><span class="mono">§ 6</span><a href="#retesting-matters-as-much-as-testing">Retesting matters as much as testing</a></h2>
<p>A fix can close the exact request in a report and leave the underlying flaw open. A public example: in <a href="https://hackerone.com/reports/791775">HackerOne report #791775</a>, a researcher bypassed Shopify’s email verification, which allowed takeover of store accounts through Shopify’s SSO. According to the program’s public summary, Shopify deployed a permanent fix three hours after disabling the affected feature. The researcher then bypassed email verification again, and Shopify found “another bug with a separate root cause,” which it asked him to file as a <a href="https://hackerone.com/reports/796808">separate report</a>.</p>
<p>HexBrain’s <a href="https://hexbrainai.com/bypasses">bypass library</a> groups 256 disclosed reports under “earlier fixes,” meaning reports that got past the fix for an earlier report. That pattern is why every finding in our report includes the root cause, not only the reproduction steps, and why we recommend a retest once fixes ship.</p>
<h2><span class="mono">§ 7</span><a href="#what-a-finding-must-contain">What a finding must contain</a></h2>
<p>Before a finding reaches your report, it has to answer five questions:</p>
<div><table>
<thead>
<tr>
<th>Question</th>
<th>What the report includes</th>
</tr>
</thead>
<tbody>
<tr>
<td>What is wrong?</td>
<td>A one-line title and a plain-language summary</td>
</tr>
<tr>
<td>How did we reach it?</td>
<td>Numbered reproduction steps with the exact requests and responses</td>
</tr>
<tr>
<td>What can an attacker do?</td>
<td>Impact written for both engineers and the business owner</td>
</tr>
<tr>
<td>How severe is it?</td>
<td>A severity rating with the scoring rationale</td>
</tr>
<tr>
<td>How do you close it?</td>
<td>A specific fix for the root cause, plus defense in depth where it applies</td>
</tr>
</tbody>
</table></div>
<p>If we can’t prove a finding with evidence, it goes in the report as an observation, not as a vulnerability. That keeps your engineers working on real issues.</p>
<h2><span class="mono">§ 8</span><a href="#questions-to-ask-any-penetration-testing-vendor">Questions to ask any penetration testing vendor</a></h2>
<p>Whether or not you work with us, ask these before you sign:</p>
<ul>
<li>How many accounts per role will you test with?</li>
<li>Will you attempt privilege escalation, or stop at identification?</li>
<li>How do you report a critical finding during the engagement, not only at the end?</li>
<li>Does every finding include the evidence and the root cause?</li>
<li>Is a retest included?</li>
</ul>
<h2><span class="mono">§ 9</span><a href="#next-step">Next step</a></h2>
<p>If you want a penetration test scoped to your environment, <a href="/contact">request a scoping call</a>. We confirm scope, boundaries, and timeline at kickoff, before any testing starts.</p>
<h2><span class="mono">§ 10</span><a href="#references">References</a></h2>
<ul>
<li>HexBrain, <a href="https://hexbrainai.com/browse">Browse disclosed bug bounty reports</a> (class counts retrieved November 20, 2026)</li>
<li>HexBrain, <a href="https://hexbrainai.com/bypasses">The bypass library</a></li>
<li>HackerOne, <a href="https://hackerone.com/reports/791775">Report #791775: Email confirmation bypass in myshop.myshopify.com</a></li>
<li>HackerOne, <a href="https://hackerone.com/reports/796808">Report #796808: Email confirmation bypass, part II</a></li>
</ul>
]]></content:encoded>
        <author>info@cybersib.co (CyberSib Team)</author>
        <comments>7 min read</comments>
      </item>
  </channel>
</rss>