Website security checks that find weak spots before attackers do
A page can load normally while its settings leave visitors exposed.
GET / HTTP/2HTTP/2 200content-type: text/html; charset=utf-8cache-control: no-storeset-cookie: session=8f2c41d9; Path=/A cookie is missing security protectionsGET /.env HTTP/2HTTP/2 200Private files may be open to anyonecontent-type: text/plainSiteCMD's free scan reads your home page, its response headers and your domain's public records. It shows what passed, what failed and why it matters.
Check the settings behind your website
Control what browsers allow on your pages
4 guides
Learn which response headers guide browser behavior and where to add them.
How to fix
- Find where your headers are setReview your hosting settings, CDN rules and framework configuration. Find which layer sends the header before adding another copy.
- Inspect the response visitors receiveAfter changing the setting, inspect the live response. Check a range of pages, including error responses, rather than assuming the home page represents the whole site.
Security
Browsers are free to guess file types
Browsers are left free to guess a file's type, which lets a malicious upload be reinterpreted and run as a script.
Fix with your agent
# SiteCMD Fix Brief: Browsers are free to guess file types Site: https://yoursite.com ## What is wrong Browsers are left free to guess a file's type, which lets a malicious upload be reinterpreted and run as a script. ## How to fix Add `X-Content-Type-Options: nosniff` to responses. This is a low-risk header for correctly served assets; it tells browsers not to guess file types, preventing MIME-sniffing attacks. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Read your content policy and investigate blocked scripts before tightening it.
How to fix
- Read the blocked address and the rule it violatesRead the error: it names the blocked address and the directive it violates. Confirm that the resource belongs on the page before changing the policy.
- Correct the specific permission or scriptUpdate the relevant source permission, nonce or hash. Don't replace a narrow rule with a wildcard just to silence the error.
Security
No strong rule on which scripts may run
Your site's rule for which scripts may run is missing or needs work, which may leave less protection if harmful code gets in.
Fix with your agent
# SiteCMD Fix Brief: No strong rule on which scripts may run Site: https://yoursite.com ## What is wrong Your site's rule for which scripts may run is missing or needs work, which may leave less protection if harmful code gets in. ## How to fix Inventory the scripts, styles, frames, fonts, images, workers, forms, and API origins the deployed product intentionally uses, then build a least-privilege policy from that list, using per-response nonces or stable hashes for required inline scripts instead of `unsafe-inline` or wildcards. Deploy the candidate as `Content-Security-Policy-Report-Only` with a monitored reporting endpoint, exercise representative routes and authenticated states while separating application traffic from extension noise, and enforce the reviewed policy only after coverage is representative and required resources work. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Set framing rules so another site can't freely place your page behind its own controls.
How to fix
- This needs a change to your site's code or server settings.
- Stop other websites from showing your pages inside a frame, a window that loads one site within another.
- If partners show some of your pages on their own sites, like a booking widget, identify which partner sites those are, so those pages keep working there.
- Then run the check again.
Security
Your pages can be shown inside other sites
Another website could load your page inside a hidden frame and trick visitors into clicking things unintentionally.
Fix with your agent
# SiteCMD Fix Brief: Your pages can be shown inside other sites Site: https://yoursite.com ## What is wrong Another website could load your page inside a hidden frame and trick visitors into clicking things unintentionally. ## How to fix Decide which origins, if any, may frame each sensitive page, then prefer CSP `frame-ancestors 'none'` or an exact origin list, adding `X-Frame-Options: DENY` or `SAMEORIGIN` as legacy defense when compatible. Keep account and admin flows unframeable, avoid `SAMEORIGIN` for widgets partners must embed cross-origin, and test a hostile iframe plus every intended embedding origin. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Use integrity hashes to tie an outside script or stylesheet to the contents you reviewed.
How to fix
- Choose a file with a fixed versionUse an address whose contents are meant to stay fixed. If the file at a versionless address changes and the hash doesn't, the browser will reject the new file.
- Generate the hash from the file you intend to loadUse a supported SHA-256, SHA-384 or SHA-512 digest in the integrity attribute. Check that the digest comes from the intended file before publishing it.
- Make sure the outside host permits the requestCross-origin integrity checks require CORS. Use the crossorigin attribute and confirm that the resource host returns the needed access-control header.
- Update the file reference and hash togetherWhen upgrading a script, replace its expected hash as part of the same change. Load the page and check for rejected resources before releasing it.
Security
Outside scripts run without an integrity check
Scripts loaded from another site have no integrity check, so a compromised host there could silently change what runs on your page.
Fix with your agent
# SiteCMD Fix Brief: Outside scripts run without an integrity check Site: https://yoursite.com ## What is wrong Scripts loaded from another site have no integrity check, so a compromised host there could silently change what runs on your page. ## How to fix Add SRI only where the URL identifies stable, versioned bytes and, for a cross-origin resource, the server supports the CORS request SRI requires; a rolling vendor URL whose bytes change in place cannot be pinned safely with one hash. For a stable resource, generate a SHA-384 digest of the exact production bytes, add `integrity` with the appropriate cross-origin mode, and test the real page, because a mismatched hash blocks the resource rather than falling back to unverified bytes. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Stop your website from revealing too much
1 guide
Remove unnecessary server and framework details from the response visitors receive.
How to fix
- This needs a change to your site's server or app settings.
- Remove the software name and version number your server adds to its responses, including on error pages.
- Keep that software up to date as well, since hiding the version doesn't fix any weakness in it.
- Then run the check again.
Security
Your server reveals its software and version
Your server's responses reveal the exact software and version it runs, giving attackers a head start on known weaknesses.
Fix with your agent
# SiteCMD Fix Brief: Your server reveals its software and version Site: https://yoursite.com ## What is wrong Your server's responses reveal the exact software and version it runs, giving attackers a head start on known weaknesses. ## How to fix Remove unnecessary product and version detail at the component that adds it, for example Express `app.disable('x-powered-by')`, PHP `expose_php=Off`, or reduced server tokens. Treat this as low-impact hardening, not a substitute for patching, also review error pages and debug endpoints where precise versions are more actionable, and verify representative success and error responses afterward. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Keep other websites from misusing your sign-ins and links
3 guides
Choose cookie attributes that fit your sign-in flow and limit unnecessary exposure.
How to fix
- This needs a change to your site's code or server settings.
- Add the right protections to the cookie we flagged, such as Secure, which sends it only over https, and HttpOnly, which keeps scripts on the page from reading it.
- Test signing in and out afterward, including any sign-in through another service, since these settings can affect how sign-in works.
- Then run the check again.
Security
A cookie is missing security protections
A cookie on your site is missing protections like Secure or HttpOnly, making it easier to steal or intercept.
Fix with your agent
# SiteCMD Fix Brief: A cookie is missing security protections Site: https://yoursite.com ## What is wrong A cookie on your site is missing protections like Secure or HttpOnly, making it easier to steal or intercept. ## How to fix Classify the surfaced cookie first: authentication and session cookies should normally be Secure and HttpOnly with the narrowest viable Domain/Path, a bounded lifetime, and a `__Host-`/`__Secure-` prefix where compatible, while a cookie intentionally read by JavaScript cannot be HttpOnly and should not carry a reusable session secret. Choose SameSite from real same-site, OAuth, and embedded flows, then inspect production `Set-Cookie` headers and test login, logout, and cross-site behavior. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Review cross-site read permissions without opening private responses to every origin.
How to fix
- Check whether the address we flagged is meant to be readable by any website.
- If it's public content that needs no sign-in, letting every site read it is fine.
- If it holds private or signed-in data, change your server settings to allow only your own sites, matched by their exact address.
Security
Any website may read your site's data
Your site's cross-origin sharing settings may let any website read responses meant only for your own pages.
Fix with your agent
# SiteCMD Fix Brief: Any website may read your site's data Site: https://yoursite.com ## What is wrong Your site's cross-origin sharing settings may let any website read responses meant only for your own pages. ## How to fix Decide whether the resource is intentionally readable by any website: `Access-Control-Allow-Origin: *` is valid for public non-credentialed content, while sensitive or credentialed APIs need an exact origin allowlist plus normal authentication and authorization. When selecting an origin dynamically, compare the full serialized origin (scheme, host, port), emit it only on an exact match, add `Vary: Origin`, and never use substring, regex, or blind reflection checks. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Keep a return-address parameter from sending visitors to an unapproved destination.
How to fix
- Use a fixed destination when the flow allows itRemove the destination parameter if everyone should land on the same page. If you need several destinations, map a short identifier to an approved address on the server.
- Check the destination when a full address is necessaryUse a maintained URL parser. Compare the parsed scheme, host and port with the destinations you allow, and limit the path when the flow needs a specific endpoint.
- Reject lookalike and unexpected destinationsDon't rely on a raw starts-with or ends-with test. Use the validated destination for the redirect, and check that unapproved inputs can't select another site.
Security
A link can redirect visitors to outside sites
A link on your site can be tricked into sending visitors to an outside destination while it still looks like your own domain.
Fix with your agent
# SiteCMD Fix Brief: A link can redirect visitors to outside sites Site: https://yoursite.com ## What is wrong A link on your site can be tricked into sending visitors to an outside destination while it still looks like your own domain. ## How to fix Prefer a server-owned destination key or relative application path over a caller-supplied URL. Otherwise parse the target once against a configured canonical base, require the normalized origin (scheme, hostname, and port) to match an exact allowlist, reject scheme-relative, backslash, userinfo, control-character, and non-http(s) input, and fall back to a fixed safe page when validation fails. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Find outdated JavaScript libraries with known vulnerabilities
1 guide
Identify affected front-end library versions and plan an update around the code that uses them.
How to fix
- Locate the affected file and versionIdentify the script in the live page, such as a jQuery or Bootstrap file, and the version it contains. Include files copied directly into the project or loaded from a CDN, as well as packages installed by a package manager.
- Read the advisory before choosing a replacementCheck the affected versions and the conditions described by the advisory. Choose a maintained release that addresses the issue and review its migration guidance.
- Test the parts of the page that use the libraryTry the affected controls and interactions after updating. Replace old integration code where the new release requires it, then confirm the deployed page loads the intended file.
Security
Outdated code with known security flaws
A library loaded on your site has a version with a publicly known vulnerability that attackers can look up and exploit.
Fix with your agent
# SiteCMD Fix Brief: Outdated code with known security flaws Site: https://yoursite.com ## What is wrong A library loaded on your site has a version with a publicly known vulnerability that attackers can look up and exploit. ## How to fix Confirm the library and served version from the exact asset (signatures can be wrong after bundling) and read the primary advisory for affected ranges and whether the vulnerable code path is reachable in this site. Upgrade to a compatible fixed release or remove the library, then re-scan the deployed asset; if no fix exists, remove or replace the vulnerable feature and document a temporary compensating control with an owner and deadline. ## Acceptance criteria After you finish, SiteCMD will re-run the check against https://yoursite.com. The fix is only accepted if that check passes. Make the smallest change that satisfies it.
Scan your code for security vulnerabilities before launch
1 guide
Run the free SiteCMD CLI in your project to find risky code before launch.
How to fix
- Inspect production configuration and built client assets for both the service-role variable name and its resolved value; the name alone is a serious warning but does not prove a live credential shipped. Rotate any credential confirmed in a bundle, public variable, log, or repository, because the service role bypasses RLS and moving an exposed string without rotation does not contain the exposure.
- Use the project's publishable/anon key in the browser only with tested RLS policies, keep service-role use in a trusted server process behind authorization, then confirm the rotated old credential fails and browser requests as two users still enforce the intended boundary.
Security
An admin key may reach the browser (Supabase service role)
A Supabase credential that bypasses all row security may be reachable from code that runs in the browser.
Fix with your agent
# SiteCMD Fix Brief: An admin key may reach the browser (Supabase service role) Site: https://yoursite.com ## What is wrong A Supabase credential that bypasses all row security may be reachable from code that runs in the browser. ## How to fix Inspect production configuration and built client assets for both the service-role variable name and its resolved value; the name alone is a serious warning but does not prove a live credential shipped. Rotate any credential confirmed in a bundle, public variable, log, or repository, because the service role bypasses RLS and moving an exposed string without rotation does not contain the exposure. Use the project's publishable/anon key in the browser only with tested RLS policies, keep service-role use in a trusted server process behind authorization, then confirm the rotated old credential fails and browser requests as two users still enforce the intended boundary. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Limit who can access accounts and data
6 guides
Check that each table's policies keep users to their own rows.
How to fix
- Try the intended database operationUse a local or staging project and the same role your app uses. Confirm the user can read or change the rows the policy is meant to allow.
- Try another user's rowsRepeat the operation as a second user and while signed out. Include writes; permission to read a row doesn't mean permission to change it.
- Write the tests downSave these allow and deny cases as a SQL test file under supabase/tests and run them with supabase test db.
Security
Browser-used tables may have row security off (Supabase RLS)
A Supabase table is reachable from the browser with row security turned off, leaving its raw grants as the only boundary.
Fix with your agent
# SiteCMD Fix Brief: Browser-used tables may have row security off (Supabase RLS) Site: https://yoursite.com ## What is wrong A Supabase table is reachable from the browser with row security turned off, leaving its raw grants as the only boundary. ## How to fix Inventory the table's anon/authenticated grants and every browser operation first; without RLS those grants are the database boundary, and enabling RLS before defining the intended policies can immediately deny ordinary clients. In one reviewed migration, enable RLS and add only the required policies with explicit owner, tenant, or role predicates (using `WITH CHECK` for new row values), then test as unauthenticated, as two different users, and as the trusted server role. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Look for app queries that trust an ID without checking who may use it.
How to fix
- Derive the active tenant from a verified server-side membership or session, never from a request body or unverified token claim, and scope reads, writes, joins, caches, background jobs, and exports to that tenant. Enforce the boundary in a shared repository or query layer with database RLS as defense in depth where supported, and test Tenant A versus Tenant B across list, detail, update, delete, and guessed IDs.
Security
One customer's data may be reachable by another
A query in this multi-tenant app does not scope its results to the current tenant, risking one customer seeing another's data.
Fix with your agent
# SiteCMD Fix Brief: One customer's data may be reachable by another Site: https://yoursite.com ## What is wrong A query in this multi-tenant app does not scope its results to the current tenant, risking one customer seeing another's data. ## How to fix Derive the active tenant from a verified server-side membership or session, never from a request body or unverified token claim, and scope reads, writes, joins, caches, background jobs, and exports to that tenant. Enforce the boundary in a shared repository or query layer with database RLS as defense in depth where supported, and test Tenant A versus Tenant B across list, detail, update, delete, and guessed IDs. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Separate reading a token from deciding whether to trust its claims.
How to fix
- Verify the JWT signature with a maintained library and trusted key source before using any claim for identity or authorization; decoding alone only parses attacker-supplied bytes. Configure an explicit algorithm allowlist, expected issuer and audience, and expiry validation, and keep decode-only use limited to non-authoritative UI hints.
Security
Sign-in tokens may lack a signature check (JWT)
A token's contents are read and trusted without ever verifying its signature, so anyone could forge one.
Fix with your agent
# SiteCMD Fix Brief: Sign-in tokens may lack a signature check (JWT) Site: https://yoursite.com ## What is wrong A token's contents are read and trusted without ever verifying its signature, so anyone could forge one. ## How to fix Verify the JWT signature with a maintained library and trusted key source before using any claim for identity or authorization; decoding alone only parses attacker-supplied bytes. Configure an explicit algorithm allowlist, expected issuer and audience, and expiry validation, and keep decode-only use limited to non-authoritative UI hints. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Choose token storage based on script access and requests that carry cookies.
How to fix
- Pick the architecture first: for a same-site web app, prefer a server-managed opaque session in a Secure, HttpOnly cookie, and because a cookie is sent automatically add CSRF protection, rotation, and revocation rather than copying a JWT into a cookie. Stop persisting long-lived bearer tokens in localStorage, clear and rotate previously exposed tokens, and keep any browser access token short-lived and in memory only when the architecture requires it.
Security
Sign-in details may be readable by page scripts
A long-lived authentication token is stored in the browser's localStorage, where any script on the page can read it.
Fix with your agent
# SiteCMD Fix Brief: Sign-in details may be readable by page scripts Site: https://yoursite.com ## What is wrong A long-lived authentication token is stored in the browser's localStorage, where any script on the page can read it. ## How to fix Pick the architecture first: for a same-site web app, prefer a server-managed opaque session in a Secure, HttpOnly cookie, and because a cookie is sent automatically add CSRF protection, rotation, and revocation rather than copying a JWT into a cookie. Stop persisting long-lived bearer tokens in localStorage, clear and rotate previously exposed tokens, and keep any browser access token short-lived and in memory only when the architecture requires it. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Tie each third-party sign-in callback to the browser that started it.
How to fix
- Generate a cryptographically random, single-use state value with at least 128 bits of entropy per authorization attempt, bind it to the initiating browser session, and require an exact match with the unexpired stored value before exchanging the code, consuming it atomically. Test missing, altered, replayed, and cross-session callbacks; PKCE protects the authorization code and does not by itself replace this binding in every flow.
Security
Sign-in may accept replies it didn't request (OAuth state)
This OAuth callback does not verify a random state value, leaving it open to a forced login into an attacker's account.
Fix with your agent
# SiteCMD Fix Brief: Sign-in may accept replies it didn't request (OAuth state) Site: https://yoursite.com ## What is wrong This OAuth callback does not verify a random state value, leaving it open to a forced login into an attacker's account. ## How to fix Generate a cryptographically random, single-use state value with at least 128 bits of entropy per authorization attempt, bind it to the initiating browser session, and require an exact match with the unexpired stored value before exchanging the code, consuming it atomically. Test missing, altered, replayed, and cross-session callbacks; PKCE protects the authorization code and does not by itself replace this binding in every flow. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Store a slow, salted hash instead of the password itself.
How to fix
- Hash passwords with the framework's maintained password API or a reviewed library: prefer Argon2id where supported; bcrypt, scrypt, or PBKDF2 can be appropriate with parameters benchmarked on production-class hardware, not one universal work factor. Use the library's salt and verification APIs, rate-limit attempts, and plan migration of existing plaintext credentials, forcing resets when safe rehash-on-login is not possible.
Security
Passwords may be saved in readable form
Passwords appear to be stored without being hashed, so a database breach would expose every password in readable form.
Fix with your agent
# SiteCMD Fix Brief: Passwords may be saved in readable form Site: https://yoursite.com ## What is wrong Passwords appear to be stored without being hashed, so a database breach would expose every password in readable form. ## How to fix Hash passwords with the framework's maintained password API or a reviewed library: prefer Argon2id where supported; bcrypt, scrypt, or PBKDF2 can be appropriate with parameters benchmarked on production-class hardware, not one universal work factor. Use the library's salt and verification APIs, rate-limit attempts, and plan migration of existing plaintext credentials, forcing resets when safe rehash-on-login is not possible. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Verify payments and webhooks before acting on them
2 guides
Confirm each payment event came from Stripe, and keep prices and return URLs on the server.
How to fix
- Verify the provider's signature against the exact raw request bytes before parsing, using the provider SDK or constant-time comparison, with the secret in server-side secret storage. Validate the signed timestamp with a narrow replay tolerance where the scheme provides one, and reject missing, stale, wrong-key, and wrong-body signatures before any side effect; a valid signature does not stop replay by itself.
Security
Incoming notifications (webhooks) may not be verified
An incoming webhook is trusted and processed without verifying it was actually signed by the provider that sent it.
Fix with your agent
# SiteCMD Fix Brief: Incoming notifications (webhooks) may not be verified Site: https://yoursite.com ## What is wrong An incoming webhook is trusted and processed without verifying it was actually signed by the provider that sent it. ## How to fix Verify the provider's signature against the exact raw request bytes before parsing, using the provider SDK or constant-time comparison, with the secret in server-side secret storage. Validate the signed timestamp with a narrow replay tolerance where the scheme provides one, and reject missing, stale, wrong-key, and wrong-body signatures before any side effect; a valid signature does not stop replay by itself. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Authenticate incoming provider messages before their handlers change state.
How to fix
- Verify the provider's signature against the exact raw request bytes before parsing, using the provider SDK or constant-time comparison, with the secret in server-side secret storage. Validate the signed timestamp with a narrow replay tolerance where the scheme provides one, and reject missing, stale, wrong-key, and wrong-body signatures before any side effect; a valid signature does not stop replay by itself.
Security
Incoming notifications (webhooks) may not be verified
An incoming webhook is trusted and processed without verifying it was actually signed by the provider that sent it.
Fix with your agent
# SiteCMD Fix Brief: Incoming notifications (webhooks) may not be verified Site: https://yoursite.com ## What is wrong An incoming webhook is trusted and processed without verifying it was actually signed by the provider that sent it. ## How to fix Verify the provider's signature against the exact raw request bytes before parsing, using the provider SDK or constant-time comparison, with the secret in server-side secret storage. Validate the signed timestamp with a narrow replay tolerance where the scheme provides one, and reject missing, stale, wrong-key, and wrong-body signatures before any side effect; a valid signature does not stop replay by itself. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Set limits on AI calls, endpoints and uploads
3 guides
Limit paid model work by caller and request size, with controls around each call.
How to fix
- Rate-limit every path that can start provider work, including streaming and background jobs, with limits derived from the feature's latency and spend budget rather than one universal requests-per-minute number. Key authenticated limits by user plus tenant or billing account, add a separate pre-auth control for anonymous entry points, use a shared atomic store across instances, and return 429 with a valid
Retry-Afterso rejected requests never reach the provider.
Security
AI calls may have no rate limit
Nothing limits how often a visitor can trigger this AI call, leaving it open to abuse and runaway cost.
Fix with your agent
# SiteCMD Fix Brief: AI calls may have no rate limit Site: https://yoursite.com ## What is wrong Nothing limits how often a visitor can trigger this AI call, leaving it open to abuse and runaway cost. ## How to fix Rate-limit every path that can start provider work, including streaming and background jobs, with limits derived from the feature's latency and spend budget rather than one universal requests-per-minute number. Key authenticated limits by user plus tenant or billing account, add a separate pre-auth control for anonymous entry points, use a shared atomic store across instances, and return 429 with a valid `Retry-After` so rejected requests never reach the provider. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Put limits on sign-in, password reset, forms and other public endpoints.
How to fix
- Send requests across app instancesTest bursts and sustained traffic from more than one caller. Confirm the intended allowance applies across the deployed entry points.
- Check what a rejected request startsCheck that a refused request doesn't send the email, write the record or call the protected service. Confirm the HTTP response and retry guidance match the policy.
Security
A public-facing action may lack a request rate limit
An endpoint anyone can reach anonymously has no rate limit, leaving it open to abuse or brute-force attempts.
Fix with your agent
# SiteCMD Fix Brief: A public-facing action may lack a request rate limit Site: https://yoursite.com ## What is wrong An endpoint anyone can reach anonymously has no rate limit, leaving it open to abuse or brute-force attempts. ## How to fix Identify each anonymous or abuse-sensitive operation and choose controls from its cost and abuse mode rather than copying one requests-per-window number across login, reset, search, and uploads. Use layered keys such as account or identifier plus network signals, an atomic shared store across instances, and a 429 response with `Retry-After`; test bursts, distributed IPs, and multi-instance deployments. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Check uploaded files on the server and keep each user's storage keys their own.
How to fix
- Enforce the exact formats and size limits the feature needs server-side while streaming, treating the declared Content-Type and extension only as hints; inspect magic bytes and fully parse the file with a maintained library where feasible. Store uploads outside executable and static roots under opaque server-generated keys, serve them with an explicit safe Content-Type, and authorize every object operation.
Security
Uploaded files may lack size or file-type checks
A file upload is accepted with no real check on its type, size, or contents before it gets stored.
Fix with your agent
# SiteCMD Fix Brief: Uploaded files may lack size or file-type checks Site: https://yoursite.com ## What is wrong A file upload is accepted with no real check on its type, size, or contents before it gets stored. ## How to fix Enforce the exact formats and size limits the feature needs server-side while streaming, treating the declared Content-Type and extension only as hints; inspect magic bytes and fully parse the file with a maintained library where feasible. Store uploads outside executable and static roots under opaque server-generated keys, serve them with an explicit safe Content-Type, and authorize every object operation. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Keep secrets out of your repository and vet its packages and workflows
3 guides
Find possible credentials in source and configuration before sharing the project.
How to fix
- First verify the matched value is a live or plausibly usable credential rather than a public identifier, test fixture, or placeholder. If it is real, revoke or rotate it at the issuing service before treating the code change as remediation, then load the replacement from a server-side secret store or injected environment and search repository history, logs, and build artifacts for the old value.
Security
A possible password or key is written into source code
A value that looks like a real credential is written directly into the source code instead of a secret store.
Fix with your agent
# SiteCMD Fix Brief: A possible password or key is written into source code Site: https://yoursite.com ## What is wrong A value that looks like a real credential is written directly into the source code instead of a secret store. ## How to fix First verify the matched value is a live or plausibly usable credential rather than a public identifier, test fixture, or placeholder. If it is real, revoke or rotate it at the issuing service before treating the code change as remediation, then load the replacement from a server-side secret store or injected environment and search repository history, logs, and build artifacts for the old value. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Watch for look-alike package names, weak lockfiles and missing release delays.
How to fix
- Check the exact import spelling against the library the code intends to use; this finding is based on string similarity and does not prove the near-match package exists or is malicious. Because the import is undeclared, run a clean type-check or build first to rule out a local alias or framework-provided module, and if a registry dependency is genuinely needed, verify its publisher and provenance before installing the intended exact package.
Security
Code imports a package whose name looks like a typo
An imported package name closely resembles a popular library's name, a common way malicious lookalike packages get in.
Fix with your agent
# SiteCMD Fix Brief: Code imports a package whose name looks like a typo Site: https://yoursite.com ## What is wrong An imported package name closely resembles a popular library's name, a common way malicious lookalike packages get in. ## How to fix Check the exact import spelling against the library the code intends to use; this finding is based on string similarity and does not prove the near-match package exists or is malicious. Because the import is undeclared, run a clean type-check or build first to rule out a local alias or framework-provided module, and if a registry dependency is genuinely needed, verify its publisher and provenance before installing the intended exact package. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Pin actions and keep untrusted pull request input away from your secrets.
How to fix
- Replace each third-party action's tag or branch reference with the full 40-character commit SHA of the release you trust (
uses: owner/action@<commit-sha> # v4), keeping the human-readable version in a comment. Enable Dependabot or Renovate for thegithub-actionsecosystem so pins get reviewed updates; prioritize third-party actions since they run with your workflow's permissions and secrets.
Security
Outside GitHub Actions aren't locked to a fixed version
A workflow references a third-party GitHub Action by a movable tag instead of a fixed commit, so its code can change under you.
Fix with your agent
# SiteCMD Fix Brief: Outside GitHub Actions aren't locked to a fixed version Site: https://yoursite.com ## What is wrong A workflow references a third-party GitHub Action by a movable tag instead of a fixed commit, so its code can change under you. ## How to fix Replace each third-party action's tag or branch reference with the full 40-character commit SHA of the release you trust (`uses: owner/action@<commit-sha> # v4`), keeping the human-readable version in a comment. Enable Dependabot or Renovate for the `github-actions` ecosystem so pins get reviewed updates; prioritize third-party actions since they run with your workflow's permissions and secrets. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Protect your database queries from SQL injection
1 guide
Keep request values out of the raw SQL your ORM runs.
How to fix
- Trace the matched value into the executed query and confirm whether a driver binding or trusted wrapper already separates data from SQL structure. Where values are interpolated, replace them with the driver's bound-parameter API, and map identifier choices such as table, column, or sort direction through a strict server-owned allowlist because they usually cannot be parameterized.
Security
Database queries may be built from raw text (SQL)
User input is inserted directly into a SQL query's text instead of through a safe parameter, opening the door to injection.
Fix with your agent
# SiteCMD Fix Brief: Database queries may be built from raw text (SQL) Site: https://yoursite.com ## What is wrong User input is inserted directly into a SQL query's text instead of through a safe parameter, opening the door to injection. ## How to fix Trace the matched value into the executed query and confirm whether a driver binding or trusted wrapper already separates data from SQL structure. Where values are interpolated, replace them with the driver's bound-parameter API, and map identifier choices such as table, column, or sort direction through a strict server-owned allowlist because they usually cannot be parameterized. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Upgrade your runtime before it reaches end of life
1 guide
Find version declarations that select or still allow an unsupported runtime.
How to fix
- Check the runtime's official support schedule and determine what the flagged declaration controls: a version-manager file selects a tool version, while
engines.node,requires-python, or Composerrequire.phpusually declare compatibility and do not prove the deployed runtime. Update every environment intended to match, run the build and tests on the newly supported line, and treat a raised library compatibility floor as a release-impacting change.
Security
Settings allow an unsupported Node.js, Python or PHP version
This project targets a runtime version that is at or approaching the end of its official support window.
Fix with your agent
# SiteCMD Fix Brief: Settings allow an unsupported Node.js, Python or PHP version Site: https://yoursite.com ## What is wrong This project targets a runtime version that is at or approaching the end of its official support window. ## How to fix Check the runtime's official support schedule and determine what the flagged declaration controls: a version-manager file selects a tool version, while `engines.node`, `requires-python`, or Composer `require.php` usually declare compatibility and do not prove the deployed runtime. Update every environment intended to match, run the build and tests on the newly supported line, and treat a raised library compatibility floor as a release-impacting change. ## Acceptance criteria SiteCMD accepts the fix when an authenticated code submission bound to this attempt, from your CI or the SiteCMD CLI, no longer reports the finding. SiteCMD records the result as self-attested. Make the smallest change that fixes the reported risk.
Frequently asked questions
Does HTTPS mean my website is secure?
No. HTTPS encrypts the connection and helps authenticate the server. Other settings control which scripts a page runs and what can access its cookies.
Does the free scan check my SSL certificate?
No. SiteCMD checks the certificate in hosted scans and emails you before it expires.
Why do some security checks need proof that the site is mine?
Some checks send test requests to your site. Others report findings only the owner should see. Prove the site is yours with a DNS record, a file or a meta tag, and SiteCMD runs them.
How can I check my project code for security issues?
Use the free SiteCMD CLI to audit your project code. You don't need an account, and findings stay where the CLI runs.
How do I fix an issue the scan finds?
A fix may mean changing your hosting settings, a DNS record or project code. SiteCMD gives guidance for each finding and can prepare a brief for your coding agent.