Skip to main content
Security templateReviewed

Penetration testing rules of engagement: a template for staging and production

A rules of engagement template for web application and API tests, with staging and production side by side.

Each rule sets out the choice to make, a starting default for each environment and what both parties write down before the first request, from request limits and stop phrases to critical-finding escalation and clean-up.

Rules of engagement builder: staging and production side by side

Fourteen rules, each with the choice to make, a starting default for staging, a starting default for production and what to write down. Work through it with your provider before the window opens: keep or change each default, fill every prompt in round brackets with the agreed value, and have both parties sign the finished version.

How to use it

The defaults are starting points to discuss, not a description of any provider's practice, and agreed values such as rates, hours and contacts are left as prompts on purpose. Denial of service, destructive payloads, social engineering and third-party platforms sit outside the default rules, and each host's own testing policy still needs its own check. Not legal advice.

14 rows

Rules of engagement builder: staging and production side by side

RuleChoice to makeStaging defaultProduction defaultWhat to write down
Request limitsHow many requests per second each target takes, how many run at once, and what happens when the application or its firewall pushes back.A higher ceiling than production if nothing else shares staging during the window.A ceiling below the point where your host or firewall starts throttling a single address, checked against normal traffic for the hours of the window. Requests slow down, then stop on that target, if rate-limit or error responses persist.(requests per second and concurrent requests, for each target), who confirmed the throttling point, and the response codes that trigger a slow-down or a pause.
Testing hours and blackout periodsWhen testing may run, and the dates and times it must not.Hours when staging is not needed for a release, a demo or user acceptance testing. Tell the tester when a deployment lands mid-window.Hours when the monitoring contact can answer the phone. Quiet trading hours affect fewer customers but may mean fewer people watching, so choose hours that balance the two. No testing during busy trading periods, billing or payroll runs, releases or maintenance windows.(testing days and hours, with time zone), (blackout dates and times), any deployment freeze, and who announces an unplanned release.
Exploitation depthHow far a confirmed weakness is taken: proved and stopped, or followed to see what else it reaches.Each weakness is confirmed far enough to show its impact, and may be chained into a second weakness where the rules allow, stopping before any step that would damage data that cannot be restored.The least intrusive proof: one read of a test record across an account boundary, or one action on a test record through a missing role check. No chaining into other systems, and no use of a password, key or session found during testing, without written approval for that specific step.The depth for each environment, and (the named person who can approve going further on one finding, in writing).
Data creation and changesWhich records the tester may create, change or delete, and how test records are marked.Test records may be created, changed and deleted freely. A snapshot taken before the window lets staging be restored.Records are created only for the test, carry an agreed marker, and are the only records changed or deleted. Records that existed before the test are not changed or deleted. Payment flows stay on staging and the payment provider's test mode unless a specific production check is agreed with its refund route.(marker for test records, such as a prefix in names), (who takes and restores the staging snapshot), and any production record the tester may change, listed by id.
Accounts and outgoing messagesWhat may happen to accounts, and who may receive email, text messages or notifications the test triggers.Outgoing email and messages go to a capture inbox or test numbers. Lockout, password reset and multi-factor checks run on test accounts.Nothing is sent to a real person: routes that send messages use addresses and numbers the tester controls. Reset, lockout and deletion checks run only on test accounts set aside for them, because the lockout test needs an account someone can afford to lock.(test accounts for each role, marking those set aside for lockout and reset), (tester-controlled addresses and numbers), and any mailing list or automation that must not fire.
Prohibited techniquesWhat is off limits, written so it leaves no room for interpretation.Denial-of-service, load and volumetric testing; destructive payloads; social engineering of staff or customers; physical access; testing of third-party platforms; any host not named in the scope.Everything prohibited on staging, plus whatever the rules allow on staging only, such as chained exploits, bulk data changes or password guessing against accounts that are not set aside for it.The prohibited list by name, and the separate written agreement and approvals needed to bring any item in.
Third-party services and host policyWhich services on the tested routes belong to someone else, and what each target's host allows.Third-party services run in their sandbox or test modes where they offer them. Your application's handling of each service is tested; the service itself is not.Your application's side of each integration is in scope and the other business's platform is not, unless its owner consents in writing. The current testing policy of each target's host is read before the window.Each third-party service and its owner, any written consent, (host for each target), and the date its testing policy was read, with what it requires.
Monitoring and source addressesWhether test traffic is let through the firewall and rate limits, and who watches alerts during the window.Source addresses let through the firewall and rate limits, so the test reaches the application itself.The firewall stays as customers meet it. Your monitoring team, host and any security provider are told the source addresses and window, so test alerts are recognised rather than reported as an attack.(source IP addresses), whether each is let through or left visible, and who monitors alerts for your team, the host and any security provider.
Contacts and channelsWho is reachable during the window, on which channel, and who stands in when they are not.A technical contact during testing hours who can answer questions, restart services and restore the snapshot.A technical contact and a backup, both reachable for the whole window, plus the approver for changes to these rules. At least one channel sits outside the system under test.(name, role, mobile and email) for the tester, your technical contact, the backup and the approver, and the channel for urgent messages.
Stop conditions and stop phraseWhat halts testing, how a stop is called, and who can restart it.Pause the affected target on rising errors, an unplanned deployment or a request from your contact. Resume when the technical contact clears it.Pause the target on persistent errors, slow responses, firewall blocks or a message reaching a real person. Stop every target on the stop phrase or any sign of an incident, leave test records in place, and restart only on written approval from the approver.(stop phrase), the channels it is valid on, who may use it, how the tester confirms a stop (time and last request sent), and who can restart.
Critical-finding escalationWhich findings are reported before the report, to whom, and how fast.Reported on the agreed channel within (agreed time after confirmation). Testing of other areas continues.Reported by phone and in writing within (agreed time after confirmation). Exploitation of that finding stops at the first proof, and the route is left alone until your contact replies.The trigger list (for example cross-customer access, administrator access for a lower role, an exposed live secret, a payment step that can be skipped), the recipient and backup, the channel, the time, what the first notice contains, and whether a fix made during the window is checked before the retest.
Evidence handlingWhat evidence is kept, how it is protected and sent, and when it is destroyed.Requests and responses kept for each finding, with secrets redacted even where the data is seeded.The minimum that proves each finding, with passwords, tokens, keys and personal details masked. One record shows a read across accounts; bulk exports are not downloaded to show scale.(where evidence is stored, and in which country), encryption, who can access it, how the report is sent, (retention period), and how deletion is confirmed.
Post-test clean-upWhat the test leaves behind, and who removes it.Restore the snapshot, or remove the listed test records, accounts and files.Each item is logged as it is created, then removed or listed by id for your team. Uploaded files, created keys and raised roles go first, because they hold access or can run code. Clean-up waits while an incident is open.The clean-up list, who removes each item, (date), how removal is checked, and what the report lists as left in place.
Sign-off and changesWho signs the rules, and how a rule changes once testing has started.Signed by the person leading the test and someone with authority over the target. Changes approved in writing by the named approver.As for staging, with the signatory of the authorisation letter as approver, and each change recorded with its time before the tester acts on it.(signatories, with names and dates), (approver for changes), and the version number in use.

Rule

Choice to make

  1. Request limits
    Choice to make
    How many requests per second each target takes, how many run at once, and what happens when the application or its firewall pushes back.
    Staging default
    A higher ceiling than production if nothing else shares staging during the window.
    Production default
    A ceiling below the point where your host or firewall starts throttling a single address, checked against normal traffic for the hours of the window. Requests slow down, then stop on that target, if rate-limit or error responses persist.
    What to write down
    (requests per second and concurrent requests, for each target), who confirmed the throttling point, and the response codes that trigger a slow-down or a pause.
  2. Testing hours and blackout periods
    Choice to make
    When testing may run, and the dates and times it must not.
    Staging default
    Hours when staging is not needed for a release, a demo or user acceptance testing. Tell the tester when a deployment lands mid-window.
    Production default
    Hours when the monitoring contact can answer the phone. Quiet trading hours affect fewer customers but may mean fewer people watching, so choose hours that balance the two. No testing during busy trading periods, billing or payroll runs, releases or maintenance windows.
    What to write down
    (testing days and hours, with time zone), (blackout dates and times), any deployment freeze, and who announces an unplanned release.
  3. Exploitation depth
    Choice to make
    How far a confirmed weakness is taken: proved and stopped, or followed to see what else it reaches.
    Staging default
    Each weakness is confirmed far enough to show its impact, and may be chained into a second weakness where the rules allow, stopping before any step that would damage data that cannot be restored.
    Production default
    The least intrusive proof: one read of a test record across an account boundary, or one action on a test record through a missing role check. No chaining into other systems, and no use of a password, key or session found during testing, without written approval for that specific step.
    What to write down
    The depth for each environment, and (the named person who can approve going further on one finding, in writing).
  4. Data creation and changes
    Choice to make
    Which records the tester may create, change or delete, and how test records are marked.
    Staging default
    Test records may be created, changed and deleted freely. A snapshot taken before the window lets staging be restored.
    Production default
    Records are created only for the test, carry an agreed marker, and are the only records changed or deleted. Records that existed before the test are not changed or deleted. Payment flows stay on staging and the payment provider's test mode unless a specific production check is agreed with its refund route.
    What to write down
    (marker for test records, such as a prefix in names), (who takes and restores the staging snapshot), and any production record the tester may change, listed by id.
  5. Accounts and outgoing messages
    Choice to make
    What may happen to accounts, and who may receive email, text messages or notifications the test triggers.
    Staging default
    Outgoing email and messages go to a capture inbox or test numbers. Lockout, password reset and multi-factor checks run on test accounts.
    Production default
    Nothing is sent to a real person: routes that send messages use addresses and numbers the tester controls. Reset, lockout and deletion checks run only on test accounts set aside for them, because the lockout test needs an account someone can afford to lock.
    What to write down
    (test accounts for each role, marking those set aside for lockout and reset), (tester-controlled addresses and numbers), and any mailing list or automation that must not fire.
  6. Prohibited techniques
    Choice to make
    What is off limits, written so it leaves no room for interpretation.
    Staging default
    Denial-of-service, load and volumetric testing; destructive payloads; social engineering of staff or customers; physical access; testing of third-party platforms; any host not named in the scope.
    Production default
    Everything prohibited on staging, plus whatever the rules allow on staging only, such as chained exploits, bulk data changes or password guessing against accounts that are not set aside for it.
    What to write down
    The prohibited list by name, and the separate written agreement and approvals needed to bring any item in.
  7. Third-party services and host policy
    Choice to make
    Which services on the tested routes belong to someone else, and what each target's host allows.
    Staging default
    Third-party services run in their sandbox or test modes where they offer them. Your application's handling of each service is tested; the service itself is not.
    Production default
    Your application's side of each integration is in scope and the other business's platform is not, unless its owner consents in writing. The current testing policy of each target's host is read before the window.
    What to write down
    Each third-party service and its owner, any written consent, (host for each target), and the date its testing policy was read, with what it requires.
  8. Monitoring and source addresses
    Choice to make
    Whether test traffic is let through the firewall and rate limits, and who watches alerts during the window.
    Staging default
    Source addresses let through the firewall and rate limits, so the test reaches the application itself.
    Production default
    The firewall stays as customers meet it. Your monitoring team, host and any security provider are told the source addresses and window, so test alerts are recognised rather than reported as an attack.
    What to write down
    (source IP addresses), whether each is let through or left visible, and who monitors alerts for your team, the host and any security provider.
  9. Contacts and channels
    Choice to make
    Who is reachable during the window, on which channel, and who stands in when they are not.
    Staging default
    A technical contact during testing hours who can answer questions, restart services and restore the snapshot.
    Production default
    A technical contact and a backup, both reachable for the whole window, plus the approver for changes to these rules. At least one channel sits outside the system under test.
    What to write down
    (name, role, mobile and email) for the tester, your technical contact, the backup and the approver, and the channel for urgent messages.
  10. Stop conditions and stop phrase
    Choice to make
    What halts testing, how a stop is called, and who can restart it.
    Staging default
    Pause the affected target on rising errors, an unplanned deployment or a request from your contact. Resume when the technical contact clears it.
    Production default
    Pause the target on persistent errors, slow responses, firewall blocks or a message reaching a real person. Stop every target on the stop phrase or any sign of an incident, leave test records in place, and restart only on written approval from the approver.
    What to write down
    (stop phrase), the channels it is valid on, who may use it, how the tester confirms a stop (time and last request sent), and who can restart.
  11. Critical-finding escalation
    Choice to make
    Which findings are reported before the report, to whom, and how fast.
    Staging default
    Reported on the agreed channel within (agreed time after confirmation). Testing of other areas continues.
    Production default
    Reported by phone and in writing within (agreed time after confirmation). Exploitation of that finding stops at the first proof, and the route is left alone until your contact replies.
    What to write down
    The trigger list (for example cross-customer access, administrator access for a lower role, an exposed live secret, a payment step that can be skipped), the recipient and backup, the channel, the time, what the first notice contains, and whether a fix made during the window is checked before the retest.
  12. Evidence handling
    Choice to make
    What evidence is kept, how it is protected and sent, and when it is destroyed.
    Staging default
    Requests and responses kept for each finding, with secrets redacted even where the data is seeded.
    Production default
    The minimum that proves each finding, with passwords, tokens, keys and personal details masked. One record shows a read across accounts; bulk exports are not downloaded to show scale.
    What to write down
    (where evidence is stored, and in which country), encryption, who can access it, how the report is sent, (retention period), and how deletion is confirmed.
  13. Post-test clean-up
    Choice to make
    What the test leaves behind, and who removes it.
    Staging default
    Restore the snapshot, or remove the listed test records, accounts and files.
    Production default
    Each item is logged as it is created, then removed or listed by id for your team. Uploaded files, created keys and raised roles go first, because they hold access or can run code. Clean-up waits while an incident is open.
    What to write down
    The clean-up list, who removes each item, (date), how removal is checked, and what the report lists as left in place.
  14. Sign-off and changes
    Choice to make
    Who signs the rules, and how a rule changes once testing has started.
    Staging default
    Signed by the person leading the test and someone with authority over the target. Changes approved in writing by the named approver.
    Production default
    As for staging, with the signatory of the authorisation letter as approver, and each change recorded with its time before the tester acts on it.
    What to write down
    (signatories, with names and dates), (approver for changes), and the version number in use.

What rules of engagement settle in a penetration test

Rules of engagement are the operating agreement for the testing window.

They decide how fast requests go, how far an exploit is taken, what may be created or changed, when testing stops, who hears about a serious finding and how quickly, and what happens to evidence and test data once the window closes. Without them a test still runs, but each of those questions gets answered on the spot by whoever is online at the time.

NIST's Technical Guide to Information Security Testing and Assessment, SP 800-115, defines rules of engagement as detailed guidelines and constraints on how security testing is carried out. They are set before the test starts, and they give the test team authority to carry out the defined activities without needing further permission.

The guide says they carry the same information as an assessment plan and also cover testing activities an organisation's policies usually prohibit, such as issuing attacks to compromise systems, which the rules then authorise as part of the assessment.

SP 800-115 includes a rules of engagement template in its Appendix B and calls its structure illustrative, leaving organisations free to arrange their own however they choose. It runs from purpose, scope, assumptions and risks, through personnel, schedule, test site and equipment, to communication and incident handling, the target systems with an exclude list of systems not authorised for testing, testing execution and data handling, reporting, and a signature page.

The guide is dated September 2008 and covers networks and premises as well as applications, so some of its contents, such as badges, escorts and access to computer rooms, rarely matter to a web application.

The builder on this page keeps the parts that fit and adds rows a web application test needs, such as request rates, account lockouts and messages sent to real people.

Scope, authorisation and rules of engagement do different jobs

Questions in the middle of a test are easier to settle when each document has one job and nobody expects another document to cover it.

  • The scope names what is tested: hostnames, API paths, environments, roles and test accounts, and what is excluded.
  • The authorisation letter is the permission. It records who signed, their authority over each target, the window and the source addresses.
  • The rules of engagement govern conduct inside that permission: pace, depth, changes to data, stops, escalation, evidence and clean-up.

The rules cannot widen the permission

If the letter covers staging only, no rule in this builder can move testing to production; that takes a new or amended letter. The reverse also holds. A letter that names production says nothing about how hard production may be pushed, which is why the builder gives production its own column of defaults.

Changing a rule once testing has started

SP 800-115 says the plan or rules of engagement should be followed unless specific permission to deviate has been obtained, normally in writing, from the original signatory or the individual in command, and it recommends that testers review them periodically during the assessment.

In practice that means one named approver, a written record of each change with its time, and a numbered copy of the current rules open while testing runs.

Why staging and production get separate columns

Staging is where depth is cheap.

Test records can be created, changed and thrown away, and a crash costs an afternoon rather than customers. Production is where a result means the most, because it is the system attackers reach, and it is also where a mistake lands on someone real.

SP 800-115 says that in some cases organisations should consider whether to test production or a similarly configured non-production system, or to restrict some techniques to off-hours, and it names factors to weigh:

  • The possible impact on production. A technique likely to cause a denial of service should probably be used against a non-production system.
  • Sensitive personal information. Where testing could expose it to people not authorised to see it, the guide suggests considering a non-production system with test data in place of the real information.
  • How closely the two environments match. The guide notes there are usually inconsistencies between test and production environments, which can result in missed vulnerabilities if non-production systems are used.

Where web staging copies usually differ

For web applications those inconsistencies are predictable. Staging often runs without the production firewall or its rate limits, sends email through a sandbox, uses the payment provider's test mode, carries different feature flags and holds a fraction of the data.

One workable split is depth on staging and confirmation on production: the full set of checks runs where data can be thrown away, and production gets a short list of read-only or test-record checks that show whether the staging results hold where customers are.

Whichever split you choose, record the environment and build behind each result, so a finding confirmed only on staging says so.

Account state is data too

Lockouts, password resets and multi-factor prompts change an account as surely as an edit changes a record. The lockout test in the OWASP Web Security Testing Guide v4.2 (WSTG-ATHN-03) says the tester needs access to an account they are willing or can afford to lock.

On production that means accounts set aside for the purpose and named in the rules, so a lockout or a reset email does not land on a customer or a member of staff.

Stop conditions and the stop phrase

Stop conditions are the rules most likely to be needed in a hurry, so they should be specific enough that nobody has to interpret them.

SP 800-115's template asks for criteria for halting the testing, what the test team does if a test procedure negatively affects the network or an attacker strikes while testing is underway, the incident response chain of command in a quick-reference format, and a process for reinstating the test team and resuming. For a web application, these are the conditions worth writing down:

  • Error responses or response times rise and stay up after the request rate drops.
  • The firewall or host starts blocking the source addresses, or rate-limit responses keep coming back.
  • An email, text message or notification reaches a real person, or a record that existed before the test changes.
  • Signs that someone else got there first: an administrator account nobody recognises, unexpected files on the server, or data where it should not be.
  • Your team detects an incident, whether or not the test caused it.
  • A deployment, migration or outage starts on a target during the window.
  • The next step would cause damage or reach data outside the test. SP 800-115 notes that penetration testing can be designed to stop when the tester reaches a point where an additional action will cause damage.
  • Either side uses the stop phrase.

Pause, stop and restart

Two levels work better than one. A pause halts one target and resumes when the technical contact clears it. A stop halts every target and resumes only when the named approver says so in writing. SP 800-115 makes a related point: it says the extent to which testing is suspended during an incident varies, and that in many cases only activities involving the systems directly involved are suspended.

The stop phrase is a short phrase agreed in advance that will not come up by accident, such as two unrelated words picked together at the kickoff call. Anyone on the contact list can use it on any agreed channel without giving a reason. The tester stops first, confirms in writing with the time and the last request sent, and asks questions afterwards.

When the stop is about an incident

SP 800-115 says the presence of an attacker found during an assessment should be reported immediately to the appropriate person, and recommends that testers stop assessing the systems involved while the organisation carries out its response. Leave test records and uploaded files where they are until the people running the response agree to their removal, because clean-up can destroy evidence.

Hand them the tester's request log as well: the guide says a step-by-step log of the testers' activities provides an audit trail and lets an organisation tell the testers' actions apart from a real attacker's.

Escalating a critical finding before the report

Some findings cannot wait for the report.

SP 800-115 says proper coordination during an assessment helps ensure that appropriate people are informed of any critical high-impact vulnerabilities as soon as they are discovered. Elsewhere it gives the example of a penetration test that exploits a vulnerability to gain administrator rights on a critical system: the testers should immediately notify the person identified in the assessment plan or rules of engagement.

Define the triggers before testing starts, so the decision is not made under pressure. For a web application they usually include:

  • One customer can read or change another customer's records.
  • A lower role reaches administrator functions.
  • A live secret key, password or token is exposed where anyone can collect it.
  • A payment or approval step can be skipped, creating orders or accounts that should not exist.
  • Signs of an earlier compromise, which are handled under the stop conditions as an incident rather than as a finding.

The mechanics to agree

Name the person who receives the call and a backup. Pick a channel that sits outside the system under test, in case that system is the problem. Set how long after confirmation the first notice is due, and whether testing of the affected route continues. The first notice needs only enough to act on: the route, the accounts involved, redacted evidence and whether any real records were touched.

Fixes made during the window

Your developers may want to fix a serious finding straight away. SP 800-115 notes that administrators may want to mitigate weaknesses immediately and expect a quick re-assessment, and says testers should communicate the importance of following the organisation's change management policies and procedures.

Agree whether a fix made during the window gets a quick confirmation or waits for the retest, and record the build each result applies to, so the report does not mix results from two versions of the application.

Evidence handling and post-test clean-up

Test evidence is a map of how to attack your application, so it deserves the same care as the data it describes.

SP 800-115's template asks the rules to set guidelines for gathering, storing, transmitting and destroying test data, and says the results of any security test identify vulnerabilities an adversary can exploit and should be considered sensitive. Where an organisation has no stated retention requirement for assessment data, the guide says to specify one in the assessment plan or rules of engagement.

The WSTG v4.2 reporting chapter covers the report itself: possibly sensitive data in a finding, such as passwords, personal information or credit card details, should be masked. Production evidence needs one more rule, about volume. One record read across an account boundary proves the flaw. Downloading a full export to show scale adds risk without adding proof.

Clean-up starts before testing does. The SP 800-115 template asks whether files may be installed, created, modified or executed to support the testing, and what must happen to them once testing is completed. In a web application the list grows with every request that creates something: test accounts, orders, bookings, uploaded files, comments, API keys, webhooks and scheduled jobs.

Ask for the list to be kept as items are created, then agree who removes each one, by when, and how the removal is checked.

Anything that holds access or can run code, such as an uploaded file, a created key or a raised role, comes off first.

What the default rules leave out

Some activities do not belong in a default set of rules, because they need a different kind of provider, another owner's consent, or both.

They stay out unless a separate written agreement brings them in, with the approvals each one needs. SP 800-115 says an assessment plan should explicitly state any prohibited activities in a way that leaves no room for interpretation, so list them by name.

  • Denial-of-service, load and volumetric testing. These need a load-testing specialist and the hosting provider's written approval, planned as their own piece of work.
  • Destructive payloads: anything built to delete, corrupt or overwrite data, fill storage or exhaust a resource. A finding that suggests one would work is reported, not demonstrated.
  • Social engineering: phishing, phone calls or messages aimed at staff or customers. That is red team work with its own consent and rules.
  • Third-party platforms: a hosted payment page, a booking or email service, a content delivery network or the host's own platform. Your application's side of each integration can be tested; the other business's platform needs its owner's consent. SP 800-115 says that where an organisation does not control part of its network, such as systems housed on a third party's network, the owner of the other network usually must also consent in writing to the assessment plan.

Host policies still need their own check

Hosting providers set their own conditions for security testing, and they differ. Read the current policy for each target's host before the window opens, and record what it requires and the date you read it. Agreeing a rule in this builder does not override a host's terms, and the builder does not replace that check.

Filling in the builder with your provider

The builder works with any provider.

If yours has its own form, compare the two row by row and add whatever theirs covers that this one does not. It is a working document, not legal advice; if the rules sit inside a contract, have that contract reviewed as you would any other.

  • Start with the scope and the authorisation letter beside you, so every target in the rules is one the letter covers.
  • Go row by row. Keep each default, change it or strike it out, and fill every prompt in round brackets with the agreed value.
  • Ask whoever runs your hosting and integrations for the throttling point and the list of third-party services. They are the two rows a business owner is least likely to know offhand.
  • Walk through the stop phrase and the escalation row on the kickoff call, with everyone on the contact list present.
  • Sign it, number the version, and keep it open while testing runs. SP 800-115's template ends with a signature page and says that, at a minimum, the test team leader and the organisation's senior management should sign it, stating that they understand the test's scope and boundaries.
  • File the signed rules with the report. SP 800-115 lists assessment plans and rules of engagement among the information testers should store to provide an audit trail of testing activities.

How Clever Ops applies these rules

On a Clever Ops engagement, the working rules below go into the rules of engagement, and changing any of them needs your written agreement first.

The values the builder leaves as prompts, such as the request ceiling, the testing window and who to call, are agreed with you before the first request.

  • 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.
  • Nothing is sent by email or text message to a real person. Routes that send messages are tested with addresses the tester controls.
  • Requests stay under the agreed ceiling. Testing on a target pauses when a firewall starts blocking it or when your technical contact asks.
  • Testing begins on a staging copy wherever one matches production.
  • Each piece of test data is logged when it is created, then removed or listed in the report for your team to remove.
  • Evidence is redacted, stored encrypted in Australia and deleted on the schedule in the data handling terms.
  • Your hosting provider's testing policy is checked for each target before testing begins. Denial-of-service, load and volumetric testing are not part of any engagement, and neither are phishing or other social engineering.

Frequently asked questions

For a web application test: request limits, testing hours and blackout periods, how far exploits are taken, what data may be created or changed, how accounts and outgoing messages are handled, prohibited techniques, third-party services and host policies, monitoring and contacts, stop conditions and a stop phrase, critical-finding escalation, evidence handling, clean-up, and who signs and approves changes. The builder on this page has a row for each.

SP 800-115's template has the test team leader and the organisation's senior management sign, at a minimum. In a smaller business the natural signatory is the person who signed the authorisation letter, alongside whoever leads the test. Your technical contact should read the rules too, because they will carry out most of them.

Errors or slow responses that persist after requests slow down, the firewall or host blocking test traffic, a message reaching a real person or an existing record changing, signs of an earlier compromise, an incident your team detects, an unplanned deployment or outage, and the agreed stop phrase. Write each one down with what the tester does next and who can restart testing.

No single number suits every application. Start from the traffic production normally handles during the hours the window covers, find out what your host and firewall do when requests from one address climb, and set the ceiling well under the point where either starts throttling. Write it per target, with the number of concurrent requests, and say what happens when rate-limit responses appear. Staging can take a higher ceiling if nothing else shares it.

It depends on what you want the result to describe. Letting the source addresses through tests the application itself, which suits staging and reaches flaws a firewall only hides. Leaving the firewall in place tests what an attacker meets on production. SP 800-115 says the testers' IP addresses should be identified in the assessment plan so administrators can tell test activity from real attacks, and that, if appropriate for the goals of the assessment, monitoring can be configured to ignore activity from those addresses during testing. Record the choice either way, because it changes what the report can claim.

Yes, because a retest sends requests to the same targets. Carry the rules over and update what has changed: the build, the window, the contacts and any test accounts. A retest replays specific findings rather than exploring, so the request ceiling and the data changes it needs can often be tighter than in the original test.

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.