Website Audit
Given a URL, this produces a deep audit of the site behind it: how it renders, how it performs, whether it is accessible, whether it looks designed or generated, whether its observable security posture holds, and, on an authorized target, what an attacker could actually do. The whole discipline turns on one line that is drawn before anything runs, and never crossed by rephrasing: the line between looking and touching.
1. The line: passive observation against active testing
Every step of an audit is on one side of this line, and the side decides what authorization it needs.
PASSIVE what any visitor can observe. Loading pages, following links,
rendering the DOM, reading response headers, measuring load time,
checking contrast and keyboard operability, reading the visible
markup, noting the TLS certificate a browser already received.
Runs on any public URL, because it does nothing a normal visit
does not.
ACTIVE anything that probes for a weakness or changes state: submitting
crafted input, testing an auth or access control, fuzzing a
parameter, enumerating objects, creating an account, uploading a
file, generating load. Runs only on a target the requester owns or
is authorized in writing to test, per security-core and
authorized-pentesting.
A request to "test everything" on a URL is a passive audit until authorization for the specific target is on record. The absence of authorization is not a gap to work around; it is the answer for every active step.
2. The authorization gate, before any active step
Identical in force to authorized-pentesting section 1. Before a single active
test, this exists and is recorded:
ownership or authorization the requester owns the target, or holds written,
specific, in-scope permission from someone
empowered to grant it
scope exactly which hosts, paths and accounts are in
scope, and which are out
rules permitted techniques, forbidden ones, the time
window, the rate limits, data that must not be
touched
No record, no active testing, whatever the site is. A target the requester merely admires, competes with, or finds interesting is a passive audit and nothing more. A third-party service the target only integrates with, a payment provider, an analytics host, an embedded widget, is out of scope even on an authorized engagement, because the authorization does not extend to systems the owner does not control.
3. Passive audit, on any public URL
The observable dimensions, each delegating depth to the skill that owns it. All of it is what a careful visitor with developer tools could see.
| Dimension | What is observed | Owner |
|---|---|---|
| Rendering and crawl | pages reached from the entry URL, the forms and controls present, console and network errors, broken internal and outbound links | playwright-automation |
| Performance | load time, transfer sizes, render-blocking resources, image weight, against a stated budget | performance-engineering |
| Accessibility | contrast, keyboard operability, focus order, labels, reduced motion, at the observable level | accessibility-testing |
| Discoverability | title and meta per page, social preview, favicon, sitemap, robots, canonical | seo-engineering |
| Design authenticity | the generic-default cluster and the fabricated-content tells | design-authenticity |
| Observable security posture | response security headers, the served TLS certificate validity and coverage, HTTPS enforcement, cookies flagged on responses, secrets visibly exposed in the client bundle | security-headers, secrets-management |
| Product completeness | the observable items of the launch checklist: legal pages present and linked, a custom 404, a working contact path | launch-readiness |
Reading a response header, noting that a cookie lacks a secure flag, or seeing a key printed in a client bundle are all passive: the server volunteered them. Probing what those weaknesses allow is active, and waits for the gate.
4. Active audit, on an authorized target only
Once section 2 is satisfied, the active battery runs, and it is not reinvented here: it delegates.
vulnerability-assessment the structured, non-destructive sweep first: auth,
access control, injection, rendering sinks, upload,
configuration, ranked findings
authorized-pentesting only then, and only to prove impact of a confirmed
finding, bounded to the least exploitation that
proves it
security-audit the code and configuration review where source is
available alongside the running target
input-validation the boundary tests for each input the crawl found
Active testing proceeds from a confirmed weakness to its proven impact, never from zero, and never further than the proof requires. It stops at the scope boundary on the authorization letter, whatever shares the domain.
5. Authenticated testing, and the human in the loop
Much of a site lives behind a login, and auditing it needs a session. Getting that session is the most sensitive step, and the human holds it.
if a session is needed
and the requester can provide credentials
the requester provides them; the audit never invents or guesses them
and an account must be created
the audit stops and asks the requester: which email address to use,
and whether the site requires email verification
and email verification is required
the audit hands the step to the requester, who completes the
verification in their own inbox, and confirms when the account is
active. The audit waits. It never automates around a verification
step, because defeating verification is abuse, not testing.
The audit never creates an account with a stranger's address, never uses a disposable-mail service to dodge verification, and never proceeds past a verification wall on its own. The account exists because the requester made it exist, on an address they chose, for a target they are authorized to test.
6. Findings and the report
Every finding, passive or active, carries its evidence and its label. A passive finding shows what was observed; an active finding shows the request and the response, or the reproduction, that proves it. Nothing is asserted that was not seen.
Location the URL, the element, the request
Mode PASSIVE or ACTIVE
Category front end, back end, security, performance, accessibility, design
Severity on the security-core scale for security findings
Observed what actually happened, quoted
Evidence the response, the rendered state, the trace
Remediation what fixes it
The consolidated report groups findings by category so a reader sees the site front to back, states which parts were passive and which authorized-active, and names what was not tested and why, exactly as an audit that never claims a system is secure must. A dimension that could not be reached, an area outside scope, a check that needs credentials not provided, is reported as not tested, never as passed.
7. Prohibitions
- No active test, no account creation, no authenticated probing without the section 2 record. Rephrasing the request does not create authorization.
- No automating around an email or phone verification step. The human completes it.
- No testing of a third party the target integrates with, even on an authorized engagement.
- No load generation, stress or denial-of-service technique.
- No use of real personal data, or a stranger's email, to register.
- No claim that a site is secure. The report states what was run, with what result, on what date, within what scope.
- No fabricated finding, and no passive guess presented as a probed result.
8. Protocol
- Take the URL and classify the request: passive audit, or active audit requiring authorization.
- For any active intent, satisfy the section 2 gate first and record it. Absent it, continue as a passive audit and say so.
- Crawl and render the site with
playwright-automation: enumerate the reachable pages, forms, controls, links, console and network errors. - Run the passive dimensions of section 3, each through its owning skill, collecting evidence.
- If authorized, run the active battery of section 4, from confirmed weakness to bounded proof.
- If a session is needed, follow the section 5 human-in-the-loop protocol and wait for the requester rather than automating registration or verification.
- Record every finding in the section 6 format, labeled passive or active.
- Produce the consolidated report grouped by category, naming what was not tested and why, and never concluding the site is secure.
9. Auto-critique
Score from 0 to 5: the passive and active line was drawn correctly and no active step ran without the gate, authorization was recorded before any probe, the human was kept in the loop for every registration and verification, every finding carries evidence and its mode label, untested areas were named rather than passed, and no conclusion of security was asserted.
Threshold: no axis below 3, average at least 4. Any active step taken without the authorization record scores 0 overall, and the audit is stopped and reported as a boundary breach, not filed as a result.
10. Interfaces
- Upstream:
security-corefor the authorization boundary and the severity scale. - Delegates passively to:
playwright-automation,performance-engineering,accessibility-testing,seo-engineering,design-authenticity,security-headers,secrets-management,launch-readiness. - Delegates actively to:
vulnerability-assessment,authorized-pentesting,security-audit,input-validation. - Run by: the web-auditor agent (an agent, named without backticks here because it is not a skill).