v0.0.190
Release notes
This release is signed with the squirrelscan release key, key id 3d10cf3775665622. If squirrel self update from 0.0.189 or earlier fails with ATTESTATION_FAILED (only when the GitHub CLI is logged in), re-run the install one-liner once (curl -fsSL https://install.squirrelscan.com | bash). It rolls up v0.0.190-beta.1 and v0.0.190-beta.2, with key-signed releases, the stale-HTML re-audit fix and em-dash-free output as the headline changes.
Added
- Smart-audit reports state, per rule check, how many pages held an open finding going in, how many this run re-checked and how many stayed carried (
meta.coverage.findings), and the requested page set when one is given (meta.coverage.requested), so a consumer can decide whether an issue may resolve after a partial re-audit. JSON schema 1.30. (#2079) squirrel audit --target-urlaccepts--compare-url(repeatable): a comparison set fetched once next to the targets, socontent/duplicate-descriptionandcrawl/schema-noindex-conflictrun over those pages instead of skipping withtargeted-run. Every other site rule is reported as not checked. (#2096)
Changed
- CLI output and rule messages no longer contain em dashes; they use a colon, comma or parentheses instead, and a guard test keeps them out. (#2211)
- The console report labels its counts: section headers count findings ("3 warning findings") and the Group and Category Breakdown count checks ("Category Breakdown (checks):"), so the numbers no longer read as a contradiction. (#2212)
schema/entity-publisher-mismatchnow says "two entities share the name X" and names both@ids when the publisher and the primary organization have the same name, instead of calling the publisher "not the primary organization" (#2227).security/form-captcharecognises honeypot fields: a public form whose only protection is a hidden decoy input withtabindex="-1"andautocomplete="off"now reports as info naming the honeypot instead of a warning. (#2228)- A targeted audit (
squirrel audit --target-url) now runs the site rules that can be judged from the listed pages alone:links/broken-links,links/broken-anchorsandlinks/broken-external-linksreport broken links on the listed pages,security/leaked-secretsscans those pages and the scripts they load, andsecurity/csp,security/hstsandsecurity/x-frame-optionsjudge each listed page's own response headers, each finding naming the pages it was read from. The links on the listed pages are status-checked again, each target once and nothing else: the internal link check sends its HEAD control to a listed page instead of the start URL, and external link checks follow the run's level and--external-links/--no-external-links(the note that--external-linksis ignored is gone, andmeta.auditLevel.settings.externalLinksis the run's real setting). Rules that need the whole site (orphan pages, sitemap coverage, duplicates across pages) still skip withtargeted-run.squirrel report --difflists a site finding that names no page as not compared, with the new reasontargeted-run, instead of counting it fixed after a targeted run. The scan line now reads "rules that need the whole site were not run". (#1778) - Correction to an earlier release note: the rule cache adds under 3% of CPU time on a cold run for typical sites (measured at +2.2% on squirrelscan.com and +0.6% on docs.squirrelscan.com), rather than the earlier cost being gone entirely. Very large pages are still being measured in #2206. (#1971)
security/supabase-rlswaives an intentionally public table by its bare name:[[ignore]]withitems = ["posts"]waives thepoststable in every project. Each finding item is now labelled with the table name alone, with the row count beside it, and the fail message no longer lists table names, so a waived table no longer appears in it (#2103)security/supabase-rlsnow probes up to 30 tables per Supabase project and 40 per audit (was 8 and 10), picks them in a fixed name order so the same site probes the same tables on every run, and says how many tables were checked and how many were left unchecked in a not-checked result and in a fail that did not reach every table (#2104)a11y/paste-inputsnow reports paste blocking on password and one-time-code fields only (type="password", orautocompleteofcurrent-password,new-passwordorone-time-code) and maps to WCAG 2.2 success criterion 3.3.8 Accessible Authentication (Minimum), so--wcag-level AAkeeps it. Paste, copy or cut blocking on other fields, and copy or cut blocking on password fields, is no longer reported. (#2106)crawl/server-errorsnow reports each 5xx page as its own finding, so findings store, compare and resolve per page: a page that recovers resolves while other pages still fail, and a page the crawl did not fetch keeps its open finding (#2130).- Team is now a flat plan: $89 a month or $890 a year for 15,000 pooled credits, with unlimited members and no seat charge. Per-seat Team subscriptions keep their price, seats and credits (#2148).
Fixed
a11y/duplicate-id-ariano longer errors on a missingaria-controlsoraria-ownstarget of a collapsed disclosure (aria-expanded="false") oraria-haspopuptrigger. Missing ids now read "ARIA reference to a missing id", separate from duplicated ids; a missingaria-labelledbyoraria-describedbytarget stays an error and other missing targets are a warning. (#2224)crawl/sitemap-domainno longer flags URLs from a sitemap declared in robots.txt on another host (or its child sitemaps), and its messages say whether the sitemap was robots-declared. (#2225)squirrel analyzeno longer counts warnings from advisory rules in its warning total, so its totals match the audit report, and it prints a line saying why the totals can still differ slightly (#2213).a11y/aria-command-nameno longer flags a link or button named by a childsvg role="img"witharia-labeloraria-labelledby(#2223).- Re-audits no longer reuse a stored page for up to a day when a site sends CDN cache headers (
s-maxage,stale-while-revalidate): the crawler ignoress-maxageand revalidates stale pages with a conditional request (#2229). - A leaking current-user endpoint such as
/api/meis now always reported exactly once.security/unauthenticated-apireports every endpoint it finds, and squirrel keeps one finding per endpoint after the site rules finish: thesecurity/auth-endpoint-disclosurefinding when that rule reported the endpoint, otherwise thesecurity/unauthenticated-apione. Before, an endpoint handed tosecurity/auth-endpoint-disclosurelost its finding when that rule threw an error or ran past its time budget. (#2082) For JSON consumers: thedeferredToAuthdetail on thesecurity/unauthenticated-apiclean row is replaced byreportedByandreportedElsewhere. - A password, password hash or secret field that both
security/json-sensitive-fieldsandsecurity/auth-endpoint-disclosurefind on a current-user or session endpoint such as/api/accountis now reported once, bysecurity/auth-endpoint-disclosure, and the endpoint is requested once per audit. Each rule reports on its own, and after the site rules finish squirrel drops only the overlapping field fromsecurity/json-sensitive-fields, which keeps its other fields, such as card numbers and access tokens. (#2083) - Console output: correct singular counts ("1 error", "1 warning"), the
analyzeConfig line printed above the spinner instead of onto it,squirrel entitieslisting every entity the "no @id" count includes with unnamed entities printed once, and the--probe aggressivehint shown once in the exposed files note (#2086). - links/anchor-mismatch no longer flags links on translated stores: inflected anchor words ("Sadler" for "cykelsadel"), trailing counts ("Shoes (15)") and locale prefixes in the target path ("/da-dk/") are matched correctly (#2092).
- Fixed the date column in the
squirrel reportlist running into the Pages column. (#2095) - Re-audits that crawl only part of a site no longer count a check that was skipped on a carried-over page (for example
perf/ttfbwith no timing data) as a pass: the page row records what its last crawl could not evaluate, in the cloud and in the CLI's local store, so the carried page scores as it did in the full audit (#2099). squirrel report --inputnow keepsmeta.partialAuditfrom the saved report and publishes with it, so a partial audit is no longer republished as complete. (#2102)a11y/color-contrastnow holds large text (18pt, or 14pt bold) to the WCAG 3:1 ratio instead of 4.5:1, so large headings and buttons are no longer flagged when they pass (#2105).- a11y/reduced-motion counts a reduced-motion rule as handling only for the selectors it covers, or for the whole page when it is a global reset (
*,html,body,:root), so a library stylesheet such as Font Awesome 6 no longer hides a page's own motion; a motion library's own reduce rule ([data-aos],.animate__animated,.animated) still handles that library. Matching is by element, not selector text, so Tailwind'smotion-reduce:animate-noneon ananimate-spinelement and animate.css infinite animations covered by the library's reduce rule no longer warn, and a reduce rule for.Spinno longer clears.spin(#2111) - The
a11y/zoom-disabledrule page and solution now say WCAG 1.4.4 requires 200% text resize and that themaximum-scalecutoff of 5 is the Lighthouse threshold. Rules with no WCAG criterion can now list related criteria as links on their rule page. (#2115) - Quick crawls now fetch the start URL when the sitemap lists it past maxPages, so home-page-unreachable no longer fires falsely. (#2127)
- Rules near-duplicate, conversion-path and commerce-readiness report a coded gap (or keep a stored finding open) for input they did not fully check (#2134)
squirrel entitiesno longer fails or bloats on sites with very large entity maps: stored references per page and edges per source are capped at 1000, and a corrupt references row is skipped with a warning instead of aborting the command (#2135).- An audit whose config
[rules]selection plus the WCAG level leaves zero rules to run is now refused before the crawl starts, instead of crawling the site and reporting an empty audit (#2110). - Local incremental re-audit totals and score now count carried findings in the full audit's check units. (#2159)
- A sourced
install.shno longer sends failure reports, and a report with no message names what failed. Test harnesses and wrappers that source the script were reporting made-up failures as real install failures. A failure with no message now says which command failed, on which line, how the script was run and which bash ran it, and a curl exit code is only blamed on curl when a request was made (repo#5020). install.shnow providesopensslon Alpine/musl (as root, withapk, next tolibstdc++), so signed releases install on a bare Alpine image. On other Linux systems withoutopensslor a logged-ingh, the signed-release error names the fix (apt-get install -y openssl,dnf install opensslorapk add openssl, the line for your distro first). It never runs a package manager for you outside Alpine.
Security
install.sh,install.ps1andsquirrel self updateaccept a release signed by a pinned squirrelscan release key (ECDSA P-256, checked withopenssl, .NET and WebCrypto) as well as one with a GitHub build attestation. A signature from an unpinned or revoked key, or one that does not verify, stops the install even if an attestation would pass. Fromfirst_signed_versionup, sha256 alone is never accepted: a missing.sigwith no verifiable attestation is a hard failure, with or withoutgh. The current key, id 3d10cf3775665622, is pinned, and this release is the first one signed with it (docs) (squirrelscan/repo#5032)
Docs
- The agent skills served over MCP (the
io.modelcontextprotocol/skillsextension ofsquirrel mcpand the hosted server) now carry audit-website 3.3 and squirrelscan 2.4: hostedrun_auditacceptsrefresh: trueto re-audit after a deploy. - The rule pages for 39 more rules (adblock, ai, analytics, i18n, integrity, legal, local, mobile, reliability, social and video) list the check names each rule emits and which clean check clears which problem check (#2218).
- The
squirrel reporthelp says issues are available in squirrelscan cloud, and the MCP docs state that the old issue fieldsrule_id,occurrencesandaffected_pagesare kept with no removal planned, withcovered_findingsas their replacement (#2031) - The agent skills served over MCP (the
io.modelcontextprotocol/skillsextension ofsquirrel mcpand the hosted server) now point firewall and connection reset troubleshooting at the new firewall guide, and carry squirrelscan 2.3 (#2141). - A new guide, Firewalls, bot protection and custom headers, shows each firewall, WAF, access gateway and bot protection exception for cloud, CLI and MCP audits side by side, with header scoping and the fix for sites that reset browser-like clients (#2141).
- Docs prices no longer render as LaTeX math on the cloud team, credits, CLI credits and FAQ pages, and
make validatenow fails on unescaped currency amounts (#2204)
npm: this version is published to npm without provenance. It was built off GitHub Actions, and npm provenance needs a GitHub OIDC identity. The binaries carry the squirrelscan release-key signature instead.
Signed with the squirrelscan release key 3d10cf3775665622: every asset has a .sig (DER ECDSA P-256 over SHA-256), verify with openssl dgst -sha256 -verify.
Update to this version
squirrelscan will auto-update, or run this command to update now.