Web application penetration testing checklist: a seven-day readiness runbook for your team
A countdown your own team works through before the first day of testing: who takes and restores the backup, who sets up test accounts and synthetic data, who gets the host's answer and proves domain control, who freezes changes and briefs the people watching alerts, and who signs the go or no-go.
The seven days are your preparation plan, counted back from a testing date you have already agreed. They are not a provider's lead time.
On this page
Seven-day readiness runbook with a go or no-go sign-off
Print this, or copy it into a shared document, and write a real name against each role before the first step. Days count back from the first day of testing in your statement of work: the countdown is your team's preparation plan, not a provider's lead time.
How to use itHide the instructions
Each step names an owner by role, what done looks like, and what happens to the test if it is not done. Tick steps as they finish, and carry anything unfinished into the go or no-go as a written exception or a hold on the target it affects. The last step is the sign-off sheet. Readiness confirms preparation only: it does not replace the signed authorisation, a host's or other third party's approval, or the agreed scope.
What a penetration test readiness checklist covers
Search for a penetration testing checklist and you find two different things: lists of tests for the tester to run, and lists of preparation for the business being tested.
This page is the second kind. The OWASP Web Security Testing Guide is a well-known example of the first.
The scope, the authorisation letter and the rules of engagement all make promises about the first day of testing. The scope promises accounts in every role. The letter promises that the business controls each hostname and that the host has been dealt with.
The rules promise a contact who answers the phone and a backup that can be restored. The UK National Cyber Security Centre (NCSC) says the scoping document should state the testing team's requirements, so the customer can do any necessary preparation before the date of the test, such as creating test accounts. A readiness checklist is that preparation, with a named person on your side for each promise.
Most of it is ordinary administration: creating accounts, loading test data, reading a host's terms, publishing a DNS record, telling the people who watch your alerts. It slips because it is spread across several people, which can include an agency or IT provider outside the business, and nobody owns the whole list.
The runbook on this page gives each step an owner by role, a definition of done and the consequence for the test if it is not done, and it ends with a sign-off sheet you can print.
Readiness confirms preparation only. A completed checklist does not authorise anyone to test anything, it does not stand in for a host's or another business's approval, and it does not add to or settle the scope. Those come from the signed authorisation, the third party's own answer and the agreed scope. The checklist records that the work those documents rely on has been done.
Who owns each step
The runbook names owners by role rather than by name, because the roles exist in every business while the people change.
In a small business one person may hold three of them. Write a real name against each role before the countdown starts, including people who work for your agency, host or IT provider.
- Engagement owner: the person on your side who owns the test, keeps this list, chases the other owners and runs the go or no-go meeting. For example, an operations manager, product owner or general manager.
- Signatory: the person whose authority to approve testing is evidenced in the authorisation letter. They sign the letter and the final decision.
- Lead developer: your in-house developer or your agency. Owns the test environment, the accounts, the seeded data, the frozen build and the restore rehearsal.
- Hosting account holder: whoever holds the hosting or cloud account, which may be the agency or IT provider. Owns the host's testing terms, any notice they ask for, and access to backups held at the host.
- DNS manager: whoever can change your DNS records. Owns proof of domain control and the check that each hostname in scope still leads somewhere you run.
- Monitoring contact: whoever watches alerts for the tested systems, inside the business or at a host or monitoring service. Owns the briefing that lets test traffic be recognised.
Why a named list matters
NIST SP 800-115, the US Technical Guide to Information Security Testing and Assessment, asks an assessment plan to cover the logistics of an engagement, including a call plan with current contact information and the organisation's main point of contact for the assessment. Where parent organisations, law enforcement or an incident response team need to be told about a test, it says the person responsible for telling them should be identified.
For a web application test in a business running one application, the six roles above are that list.
Why the countdown runs in this order
The order follows how long each step waits on someone else, not how important it is.
Every step on the list has to be done; the question is which ones cannot be hurried.
- Seven and six days out: anything that depends on another business. The host's answer, a vendor's permission, an agency publishing a DNS record and the signatory's diary are the steps you cannot speed up, so they start first.
- Five and four days out: the environment. Backups, the restore rehearsal, test accounts and synthetic data need the environment in its final shape, and any of them can turn up a problem that needs time to fix.
- Three days out: the freeze. Once the build stops changing, accounts and data are checked against the version that will be tested rather than one about to be replaced.
- Two days out: the people. The monitoring briefing, the call plan, the stop phrase and access to findings are rehearsed with everyone who will be reachable during the window.
- One day out: the decision. Every gate is ticked or turned into a written exception, and the sign-off sheet is signed.
When the date is closer, or further away
If testing starts in fewer than seven days, compress the plan rather than dropping steps, because the gates on the sign-off sheet apply however short the run-up. If it is weeks away, start the steps that wait on other businesses now and leave the rest to the countdown.
Either way, the dates are yours to set: the runbook says nothing about how long any provider takes to start, test or report.
Backups and a restore someone has actually run
Testing a working system carries a risk nobody can take to zero.
NIST SP 800-115 says systems may be damaged or otherwise rendered inoperable during penetration testing, and that although experienced testers can mitigate this risk, it can never be fully eliminated. It says penetration testing should be performed only after careful consideration, notification and planning.
The NCSC guidance puts it from the tester's side: testers should make every effort to avoid undue impact on the system, but the nature of penetration testing means no one can promise that testing will cause no unexpected reactions.
A backup is a fallback for that risk only if it restores. The NIST Cybersecurity Framework 2.0 lists, among its outcomes, that backups of data are created, protected, maintained and tested (PR.DS-11), and that the integrity of backups and other restoration assets is verified before they are used for restoration (RC.RP-03).
For readiness, a tested restore means four facts written down: when the backup was taken, where it was restored to (a separate environment, not over the live system), what was checked afterwards, and how long the restore took.
Checks worth making are that the application starts, a sample of records reads correctly and uploaded files open. Include file storage as well as the database, because uploads, generated documents and images often live somewhere a database backup does not reach.
Then name the person who can run a restore during testing hours, and a second person in case the first is unavailable.
If production is in scope and no restore has worked yet, hold production testing until one has. Work on a separate staging copy can still go ahead, and a staging snapshot taken after the freeze gives that copy a clean point to return to once the window closes.
Test accounts and synthetic data that work on the first morning
Accounts can be found broken when testing starts: a password that was never set, multi-factor sign-in tied to a developer's phone, an invitation link that expired, or an account created in the wrong organisation.
Each one costs testing time, and if it cannot be fixed inside the window it becomes a gap the report has to explain. The reporting chapter of the OWASP Web Security Testing Guide (WSTG) v4.2 lists lack of access or credentials among the limitations a report can record.
Plan for two accounts in each role in scope and, where one database serves several customer organisations, accounts in two of them. WSTG v4.2 says the best way to test direct object references is with at least two users, often more, who own different objects.
Mark which accounts may be locked or reset, or set aside extra ones for those checks, since the WSTG lockout test needs an account the tester is willing or can afford to lock.
If administrator accounts use multi-factor sign-in, decide by day four how the second factor reaches the tester and write it into the handover note.
Synthetic data is what gives those accounts something to test. Each account needs records of its own, and records in the other organisation it should not be able to reach. NIST SP 800-115 suggests that where testing could expose sensitive personal information to people not authorised to see it, organisations consider testing on a non-production system with test data in place of the real information.
In practice that means names, addresses, phone numbers and files written for the test, an agreed prefix on every test record, payments in the provider's test mode, and outgoing email and text messages routed to a capture inbox or to destinations the tester controls.
SP 800-115 also recommends involving legal advisers in intrusive tests such as penetration testing, while leaving that choice to the organisation, and says the legal department should address any privacy concerns the organisation may have. If the tested environment holds a copy of real customer data, put that question to your adviser during the countdown, not on the first day of testing.
Host approval and proof that you control each domain
Two steps depend on parties outside the business, and together they decide whether a target can be tested at all.
The host, and anyone else whose platform is on a tested route
Hosting and cloud providers word their terms for customer security testing differently, and some ask for notice or written approval before testing starts. The hosting account holder reads the current terms for each target's plan, saves a copy with the date it was read, and sends any notice or request the terms ask for.
Done means an outcome is on file for each target: no approval needed under the terms as read, notice sent and acknowledged, or written approval received.
Nobody else can supply that answer, including the testing provider, and an authorisation signed by your business does not replace it.
The same goes for any other business whose platform sits on a tested route, such as a hosted sign-in service or a payment provider: its written permission, published terms that allow what the scope describes, or an exclusion. Readiness is when the answer is chased. The decision about what is in scope was made earlier, in the scope.
Proof of control, and a look at where each hostname leads
The authorisation letter states that your business controls each target. Readiness is the DNS manager publishing the record, or placing the file, that the provider asks for, and the provider confirming it can see it. While the DNS manager is in your records, have them look up every hostname in scope and match it to a system you run today.
A hostname that resolves to a server you no longer pay for, or to a supplier you stopped using, comes out of scope rather than going into the test.
NIST SP 800-115 makes a similar recommendation to assessors about addresses: given a list of authorised IP addresses to target, they should verify before testing begins that all the public addresses on it are under the organisation's purview.
A target without its host outcome or its proof of control stays out of the window until both are on file. The rest of the scope can still go ahead.
A change freeze, and why not to tidy up for the test
A test describes one build.
If code, configuration or firewall rules change mid-window, a finding can apply to a version that has already been replaced, and a check that found nothing can describe a version that no longer runs. NIST SP 800-115 says proper coordination helps ensure an assessment does not take place during upgrades, new technology integration or other times when the system's security is being altered.
Agree the freeze three days out. Write down what it covers: deployments to the tested environment, configuration and feature flags, firewall, bot protection and rate-limit settings, data migrations, and scheduled jobs that rewrite data.
Name one person who can approve an urgent exception, and agree that the tester hears about it before it lands, with the new build reference. Record the frozen build or commit on the sign-off sheet so the report can name it.
If testing runs on a separate staging copy, production releases can usually carry on, provided nothing they change reaches staging; say so in the freeze note.
Resist the urge to harden the system for the test week. SP 800-115 describes this as a lack of realism: in preparing for an assessment, users and administrators sometimes change settings to make their systems more secure or resistant to attack, and such changes are generally kept only for the duration of the assessment before systems return to their previous configurations.
A fix you mean to keep belongs before the freeze. A setting switched on for the week and off again afterwards leaves you with a report about a system you do not run.
Monitoring contacts, stop conditions and access to evidence
The last working day before the decision is about people: who will see the test, who can stop it, and who receives what it finds.
Everyone who might mistake the test for an attack
List everyone who watches alerts for the tested systems: your own team, the host's support desk if it monitors you, a monitoring or security service, and any agency with alerts on the site. Give each one the window, the time zone and the tester's source addresses.
NIST SP 800-115 says those addresses should be identified in the assessment plan so administrators can tell assessment activity apart from real attacks. Then ring every number on the call plan once and replace any that reaches voicemail or someone who has moved on.
Stops, restarts and a one-page incident plan
The stop conditions themselves belong in the rules of engagement. Readiness checks that the people who use them are ready: everyone on the call plan knows the stop phrase and the channels it is valid on, the person who can approve a restart is named and reachable, and at least one channel does not depend on the system being tested.
SP 800-115 recommends that an established incident response plan be in place during testing, because of the risk of operational impact. That plan can fit on a page: who to call if the site goes down or something looks like a real attack, and who decides whether testing pauses.
The hacked website guide covers the first hour if a stop turns out to be a real incident.
Who receives the findings, and where they are kept
Test evidence describes how your application can be attacked, so decide before the window who receives critical-finding calls, the report and the evidence behind it, and how each reaches them. SP 800-115 says organisations should store sensitive assessment data, including assessment reports, securely and limit access to the people whose roles need it, and that the assessment plan or rules of engagement should cover how sensitive information is transmitted. Have each recipient test the delivery channel, and choose where the report will be kept afterwards.
Check your own logs as well. Make sure web server, application and firewall logs for the tested environment are switched on and kept for longer than the window. SP 800-115 asks assessors to keep a step-by-step log of their activities so the organisation can tell their actions apart from a real adversary's, and that comparison only works if your side has kept its logs too.
Making the go or no-go decision
The decision is made the working day before testing, by the engagement owner and the signatory together, and the provider acknowledges it.
It has three outcomes: go, go with written exceptions, or no-go with a new date.
Sort the gates into two kinds before the meeting. Permission and safety gates decide whether a target is tested at all: a signed authorisation that covers it, the agreed scope, the host's outcome on file, proof of control, a working restore where production is in scope, and contacts who answer.
A target that fails one of these is not tested until it passes, and if it is the only target, the answer is no-go. Completeness gates decide what the report can say: one role's accounts missing, an integration switched off on staging, a vendor's permission still pending. These can go ahead as written exceptions, each with the limitation the report will record.
The NCSC describes a similar choice for systems found during a test that lie outside the scope but could affect the systems in it: change the scope, which is likely to alter time frames and cost, or record the exclusion as a limitation on testing, with the decision generally down to the risk owner.
Applying that logic the day before is simpler than meeting it mid-test, and it keeps a missing account from quietly becoming a gap nobody chose.
A go records that the preparation is done. It is not a fresh permission. It does not approve anything the authorisation letter leaves out, and it does not stand in for a host's or another business's approval.
Adjusting the list for a website security audit
A website security audit that stays outside the login needs a shorter list.
The account and synthetic data steps mostly fall away, because nothing is tested behind a sign-in. The rest still applies: an authorisation and proof of control for each hostname, the host's testing terms, monitoring contacts who know the window and the source addresses, a stop phrase, and a named recipient for the findings.
A WordPress audit that signs in as each role brings the account steps back, one set per role, and a store adds payment test mode to the list.
Using the runbook with Clever Ops or another provider
The runbook is yours and works with any provider.
Hand the signed sheet to whoever is testing, along with the scope, the authorisation letter and the rules of engagement, and file a copy with the report.
On a Clever Ops engagement, testing begins once the authorisation and access are in place: written authorisation from the owner of each target, the exact targets, your hosting provider's testing policy checked for each one, two test accounts for each role in scope plus accounts in two customer organisations where the application serves more than one, an environment to test, with a staging copy that matches production preferred, and a technical contact for the window who can answer questions, restore data and pause testing.
Destructive actions are off by default for real people's accounts and records that existed before the test, and reset, lockout and deletion checks run only on test accounts and test records where the rules of engagement allow them.
If Clever Ops built or hosts the application, the work is a security review rather than an independent test, and the report says who built and who hosts it. Where a tender, panel or contract requires CREST, Clever Ops is not CREST accredited, so we refer that work to a provider that is.
Frequently asked questions
An owner by role for each step, plus: a backup and a restore that has been run, two test accounts per role (and per customer organisation where there are several), synthetic data for each account, the host's outcome on file, proof of control for each hostname, a change freeze with the build recorded, monitoring contacts who know the window and source addresses, a stop phrase and a named restart approver, recipients and a channel for findings, and a signed go or no-go.
No. Permission comes from the authorisation letter signed by someone with authority over each target. A host's or vendor's approval comes from that business, and what is tested comes from the agreed scope. The checklist only records that the preparation those documents rely on is done, and its sign-off sheet says so.
The runbook spreads the work over the seven days before testing so that each step has room to fail once and be fixed before the next one relies on it. That is a planning suggestion for your team, not any provider's notice period. Start earlier where you depend on someone else, such as a host that asks for notice or approval, a vendor's permission, or an agency that manages your DNS.
It depends on the gate. A missing authorisation, host outcome, proof of control, working restore for production, or reachable contact keeps the affected target out of testing until it is fixed. Missing accounts, data or integrations can go ahead as written exceptions, and the report records each one as a limitation. If every target fails a permission or safety gate, move the date.
Fix what you would fix anyway, before the freeze, and tell the tester what you already know. The NCSC guidance says that in an ideal world you should know what the testers are going to find before they find it, and its model engagement, which assumes you run your own vulnerability assessment, has you share a current vulnerability assessment with the testers at the scoping stage. What to avoid is changing settings only for the test week and changing them back afterwards.
No. A list of tests, such as the checklist in the OWASP Web Security Testing Guide, tells the tester what to try. A readiness checklist tells your team what to have in place before the tester starts. A thorough test needs both, and only the second is your team's job.
Scope the right test for your application
Tell us what you have built, who uses it and when it needs to launch. You will get a plain-English recommendation: a website audit, a web application test or a referral to a specialist.