Self-hosted OSINT exposure checker — is your email, username, phone or password leaked? Keyless sources: breaches, infostealer logs, COMB credentials, HIBP k-anonymity, 200-platform username grid. Docker one-command install, no API keys, no AI, Apache-2.0.
PHP breach-checkcredential-exposuredata-breachdigital-footprintemail-osintfree-osinthaveibeenpwnedip-reputationleaked-credentialsleaked-dataosintosint-apiosint-toolsphone-lookupprivacy-toolspublic-datapwned-passwordsself-hostedusername-checkusername-osint
Changelog
[1.8.4] — 2026-11-11
Documentation
SECURITY.mdnow documents the static-analysis coverage stack: CodeQL has no PHP support, so PHP coverage is provided by Codacy (Semgrep/PHPMD/PHPCS) plus blocking PHPStan and the CI lint + contract-test matrix — this preempts the common "code scanning is not analyzing the primary language" confusion on the GitHub Security tab.
[1.8.3] — 2026-11-11
Supply chain & observability
shivammathur/setup-phppinned to a full commit SHA inci.ymlandphpstan.yml(third-party actions should not float on a moving tag; was@v2).OsintLogoService::clearMissno longer swallows storage failures in an empty catch block — cleanup errors now log a warning instead of dying silently.
[1.8.2] — 2026-11-10
Static analysis — for real this time
- The PHPStan workflow was a silent no-op:
vendor/bin/phpstanwas never a dependency (the "No such file or directory" failure was masked by|| trueplus acontinue-on-errorjob). It now runs for real and blocks the build. phpstan/phpstan^2.3 added to require-dev;phpstan.neon.distadded (level 5).- All 60 level-5 findings resolved:
- Model
@property/@methoddocblocks added (ExposureCheck,Setting,CommunityLicense) so static analysis understands the Laravel columns. - 16 code cleanups, all behavior-neutral: doctor PHP floor now checks the real
8.4 requirement (was claiming 8.2), redundant
is_array/instanceof/??guards removed, unusedFRESH_TTLconstant dropped,OsintCheckServiceuses an explicitLogfacade import, dead$osintconstructor property removed,SiteProbeSourcecaches the decoy verdict as the string'unreliable'(same contract asSiteHealthCommand) instead of leaking a bool into the verdict cache.
- Model
composer.lockcontent-hash refreshed —composer validate --strictis clean again (the hash had been stale since the v1.7.2 PHP-floor change).
Docs
- README badge + nav anchors fixed to the current heading
(
#️⃣-the-26-source-lanes-310-site-checks— the old#the-25-sourcesanchor has been dead since the v1.8.0 rename). - Counts removed from
docs/FAQ.mdanddocs/SOURCES.md(the "202 scanner modules" and "25 sources" figures were stale; the main README keeps the verified 310-checks / 26-lanes numbers, which match the shipped registry).
[1.8.1] — 2026-11-08
Quality — static-analysis zero on the v1.8.0 diff
- PSR-12 cleanup of the v1.8.0 diff (external analysis had reported 14
new findings): multi-line brace style in
SiteProbeSourceaccessors, line-length wraps inSiteProbeSource/SiteHealthCommand/OsintCheckService/SiteProbeSourceTest, registry config header spacing. - Defensive guard in
SiteProbeSource::parse()— the previously unused interface parameter ($type) now short-circuits non-username types instead of relying solely onsupports(). - Test suite renamed to camelCase (58 methods, 9 files) — PSR-1/PSR-12 method naming; removes the whole issue class for future releases.
- No behavior changes; suite green on PHP 8.4 and 8.5.
[1.8.0] — 2026-11-07
Sources — Site Probe Registry (W109): 85 new decoy-gated sites
- Source lane 26: Site Probe Registry — native username enumeration across 85 independently re-derived and live-verified sites (tier A = decoy-gated AND exemplar-proven; tier B = decoy-gated). Every entry passed the DECOY GATE on 2026-11-07: a random nonexistent username must NOT look "found" — the false-positive suppressor for soft-404/wildcard pages. Hosts already covered by the Maigret lane were deduped (36 overlaps removed).
- License firewall — URL patterns and status codes are uncopyrightable facts, re-derived from scratch and verified from our own egress; no third-party registry file was copied (GPL projects consulted conceptually only). Registry provenance header documents this.
- Decoy cache — decoy verdicts are username-independent and cached 7 days
(refreshed by
site:health), so a check pays the decoy cost once per site, never per username. site:healthcommand (weekly, Mon 04:20) — registry self-test: fresh decoy probe per site + exemplar drift detection; tier-A failures exit 1 (CI-friendly). Unreliable sites are skipped by checks automatically.- Tristate honesty per site:
hit|clean|unavailable— transport errors and ambiguous rules are never counted as clean. - Total: 26 source lanes · 310 keyless site checks (25 API sources + Maigret 200 + Site Probe 85).
- Tests: SiteProbeSourceTest (registry integrity, unique hosts, rule shapes, parse tristates) — suite 58 tests / 1,183 assertions green on PHP 8.4/8.5.
[1.7.1] — 2026-11-05
UX — Maigret "not found" summary (W96)
- Maigret clean-sites summary (
clean_sites) — platforms assessed with a definite "username free" verdict now ship in the source data and render as a collapsed "not found on N other assessed platforms" line at the END of the results card (dashboard and printable report). Found accounts stay big and first — what matters is where you WERE found; the quiet platforms fold away.
[1.7.0] — 2026-10-29
Sources — W58 OSINT repo deep-pass, round 2
- Source 24: Keybase linked identities — the identity-graph pivot: one keyless
lookup returns the account's deliberately-public verified proofs (Twitter,
GitHub, Reddit, HackerNews, DNS domains, PGP fingerprint). A bare account is a
low-severity footprint; 3+ verified links is a real deanonymization surface
(medium). Documented 205 semantics for missing users; verified proofs
(
state=1) only — unverified claims never count. - Source 25: GitHub commit author-email pivot — the reverse of the existing
github_public email check: given a USERNAME, which real addresses did it author
public commits with (e.g.
torvalds@…)? Addresses are MASKED (t***s@…); noreply-only accounts report good hygiene; and the card carries the PROBABLE caveat the adoption review required — git author fields are spoofable, so a match is evidence of exposure, not proof. Zero-match semantics live-verified: qualifier-only author searches 422 on zero matches -> clean, never unavailable. - Maigret corpus 150 -> 200 sites — 290 fresh WhatsMyName candidates, differential-self-checked (claimed vs unclaimed username) from the production egress IP: 136 passed, top 50 selected by platform recognition and category priority, with platform-family dedup (Mastodon instances capped, adult sites excluded, already-covered platforms skipped). ArtStation, Bandcamp, Dribbble, metacritic, cnet, freesound, Genius, IFTTT and more join the grid.
[1.6.0] — 2026-10-29
Sources — W58 OSINT adoption research
- Source 23: DNS blocklist reputation (Spamhaus ZEN, SpamCop, Barracuda) — the one keyless personal-IP signal the CE was missing: "is MY IP listed as a spam/abuse source?" (compromised device, hijacked router, prior abuse). Pure DNS queries (no HTTP, no keys), same offline-source pattern as Bot-CIDR.
- Honest-degradation design (live-verified in production): DNSBL refusals are
ambiguous — Spamhaus sometimes answers open/public-resolver queries with
127.255.255.x error codes and sometimes with EMPTY answers. Each zone is therefore
probed first with the permanently-listed test address 127.0.0.2; a zone whose test
probe does not demonstrably return a listing is reported
not assessed, never a fake clean. IPv6 and restricted resolvers degrade honestly too. - Adoption research (dual independent model reviews over omnisint, PublicEye, n1xYosint, osint-toolbox-mcp, Consensus-OSINT, tiktok-scraper + awesome lists): DNSBL (adopt, CE) and selective Maigret-corpus expansion (follow-up) were the two clear keyless + personal + license-safe wins. Crypto-address checks, TikTok, ORCID and Wayback were evaluated and deliberately deferred or rejected (weak exposure semantics, fragility, or keyless-availability risk) — reasoning kept in the W58 research notes.
[1.5.0] — 2026-10-29
Licensing — v2 "No Wall" (same policy as the scanner CE)
- The 14-day trial wall is GONE. Every check, source, report and export works forever with no key, no countdown and no phone-home. v1 blocked new checks after a 14-day trial unless a free Community key was activated; that wall contradicted the "free forever" promise and was removed across the codebase.
CommunityLicenseServicev2:status()is now purely informational (licensed/unlicensed/key-revoked— notrial,graceorblockedstates);checksAllowed()always returns true (kept for blade/route compatibility). The optional free key powers exactly two things: the aggregate install counter and free premium update notifications.- Zero telemetry by default. The daily heartbeat and weekly re-validation schedules
were removed from
routes/console.php. The license client only talks to the server when you explicitly activate, refresh or deactivate a key. Opt-in heartbeat for the public install map:LICENSE_HEARTBEAT=on(off by default). - Key revocation now only stops update notifications — checks keep working.
EnsureCommunityLicensedmiddleware never blocks; it shows a once-per-session hint about the optional free key.- License page rebuilt for v2 (no trial countdown, no "blocked" chip); README,
docs/PRIVACY.md,docs/FAQ.mdanddocs/SELF-HOSTING.mdupdated to match.
Maintenance
- CI:
actions/checkoutv4 → v7 (dependabot #4). - Version constant synced (
config/app.php1.1.0 → 1.5.0 — it had drifted from the changelog). - Note: the print-ready PDFs in
pdfs/still describe the v1 trial policy for the license sections; they will be regenerated in a follow-up release (the markdown docs are authoritative).
[1.4.0] — 2026-10-23
Sources
- Source 22: LeakLens breach-registry search (phone + username) — keyless
MCP endpoint (
leaklens.org/api/mcp, stateless JSON-RPCtools/call). First phone breach-exposure source (PhoneValidity only checks E.164 validity) and a breach-flavored complement to the Maigret platform-presence grid for usernames. 21 → 22 keyless sources across README, badge, docs and composer description; the dashboard count stays registry-derived. - Deliberate scope: no email (XposedOrNot already queries the same upstream corpus directly), no password (the LeakLens audit endpoint is an HIBP range proxy; PwnedPasswordsSource queries HIBP directly and the browser-side k-anonymity matcher is keyed to it), no domain (EC surfaces domain infrastructure only — crt.sh + RDAP; website scanning stays in the scanner CE).
- Honest degradation (their semantics meet ours):
isExposed:falsewithdegraded:truemaps tounavailable, never a fake clean; hits record which upstreams did not answer. Result items are reduced to aggregates (breach name, date, record count) — credential-like and third-party-identity fields are dropped before anything reaches the client. - License note: leaklens.org registry data is free for non-commercial, civic and human-rights use — fine for this self-hosted personal checker; not licensed for the commercial platform (this source ships CE-only, by design).
- New dedicated unit test (
LeakLensSourceTest, 10 cases, mocked MCP envelopes — hit/clean/degraded/isError/transport/malformed/cap/ethics).
[1.3.1] — 2026-10-23
Fixed
- v1.3.0 follow-up — the two new sources were shipped but not wired. v1.3.0 added
CrtShSourceandRdapSourceas files and updated every count claim to 21, but theOsintCheckService::SOURCESregistry still listed 19, so the dashboard rendered "19 keyless sources" while the README said 21, and the new sources never ran. Both are now registered, thedomaincheck type is live (mirroring the hosted product), and every count is registry-derived again. The domain check surfaces a domain's public infrastructure footprint only (CT-log subdomain history via crt.sh, registry records via RDAP) — website vulnerability scanning stays in the scanner CE. - Domain normalization is stricter than the hosted product:
FILTER_VALIDATE_DOMAINalone accepts spaces and underscores, so the CE requires a dot-joined hostname shape (and strips scheme, path and port). "not a domain" now 422s honestly. - The scope-rule test (
test_check_types_do_not_include_domain) becametest_check_types_include_domain_infra: the 5-check identity is now 6 checks (email, username, phone, password, domain, ip). README, docs/CHECKS.md and the dashboard tabs/placeholder/auto-detect updated in the same commit.
[1.3.0] — 2026-10-23
Sources
- Sources 20–21: crt.sh certificate transparency + RDAP domain registration records — both keyless, both free public services; 19 → 21 keyless sources across README, badge, docs and composer description. CI continues to derive the count from the source registry.
[1.2.1] — 2026-10-22
Security
- PHP base image now builds against PHP 8.3.36 (released 2026-10-21, fixes CVE-2026-109618 and CVE-2026-109619, the "PHP2Shell" unauthenticated RCE pair, CVSS v4.0 9.2). The Dockerfile uses the floating "php:8.3-apache" tag, so any image built after 2026-10-21 automatically includes the fixed runtime; no Dockerfile change was needed. Self-hosters: pull + rebuild + restart Apache.
- CI now derives the keyless-source count from the source registry (the stale hardcoded count broke CI after v1.2.0).
[1.2.0] — 2026-10-21
Added
- 19th source: Bot CIDR Attribution (offline, zero requests) — IP checks now say whether the IP belongs to an operator-published crawler network range (Googlebot, Bingbot, Meta crawler, Applebot). In-range traffic is verifiable crawler traffic; OUT-of-range traffic claiming those user-agents is spoofed. Includes the operator-documented reverse-DNS verification hint per range. Inspired by CrawlerScope research (S32).
Changed
- Source count 18 → 19 across README, dashboard copy, Docker label and FAQ.
All notable changes to Exposure-Check-OSINT are documented here.
[1.1.0] — 2026-10-21
Added
- JSON / CSV export — every stored check can now be downloaded as a JSON document (full per-source detail) or a flat CSV (one row per source verdict) from the report page, the history table and right after a live check. Exports keep identifiers masked and re-strip credential material belt-and-braces, exactly like the stored rows.
php artisan ec:doctor— install self-test: PHP version and required extensions, app key, admin password, storage writability, maigret site database validity, database + migration state, source-registry health, and a live reachability probe for all 18 sources (parallel, short timeout). Table output by default, structured JSON with--json; non-zero exit code when anything FAILs. Run it after install and any time results look wrong.- Username confidence tiers — every username hit is now labeled
confirmed(page-content or redirect verified) orprobable(status-code only) — on the dashboard logo grid, in the report table and in the JSON/CSV exports. Older stored checks render without badges (graceful degradation).
Fixed
- Reported version:
config/app.versionwas still1.0.0inside the v1.0.1 release — now bumped correctly (report header shows v1.1.0).
[1.0.1] — 2026-10-21
Fixed
- Report page: the exposure score now shows its scale (
X / 100, 100 = clean) and a prominent risk label (CRITICAL EXPOSURE/ELEVATED/MINOR/LOW/NO EXPOSURE FOUND). Previously a bare0was ambiguous — readers could mistake maximum exposure for "no exposure". - Dashboard: the daily-quota chip no longer renders as a stray empty pill
before the first check of the day (
[hidden]display guard added). - Logout now fully destroys the session record (
invalidate()+regenerateToken());regenerate()alone kept the old session alive until expiry. - Privacy contract: IP addresses are now masked in stored reports
(
103.***.***.12, IPv6 first+last group) — previously the raw IP was stored, violating the "identifiers never stored raw" promise.
Changed
- README screenshots replaced with REAL live-product captures (dashboard, live results view with score gauge, printable report) — no mockups.
Verified
- Full AI-assisted security review (seed-2-0-pro): 2 real findings fixed (both above); 4 other reported findings verified false and documented (Laravel 11 auto-CSRF; identical rate-limiter keys; registry-whitelisted logo proxy; quota consumed only on success).
- CI green: PHP 8.2 / 8.3 / 8.4 matrix, lint + contract tests + Docker build-context verification.
[1.0.0] — 2026-10-21
Added
- Initial Community Edition release.
- 5 check types: email, username, phone, password (k-anonymity), IP.
- 18 keyless source adapters extracted from the 2MNY production engine with per-source failure isolation. (The ~120-platform linked-account probe is v1.1: it requires the optional browser-service add-on.)
- Single-admin dashboard: parallel check page with platform logo grid, exposure score gauge, category sections, honest three-state verdicts.
- History with per-record delete + purge-all (identifiers stored as salted hashes + masked forms only; password checks store no rows).
- Printable HTML report (browser print -> PDF; no server-side PDF dependency).
- Settings page: daily quota, burst limit, cache TTL, honest UA string, per-source enable/disable.
- Community license (14-day trial -> free key from 2mny.com, weekly
re-validation, 7-day offline grace) — product code
exposure-check. - One-command Docker install (Apache + PHP 8.3 + SQLite, single container).
- Contract test suite (interface, key uniqueness, HTTPS-only specs, normalization, masking, credential-line stripping) + GitHub Actions CI (PHP 8.2/8.3/8.4 matrix, lint every file).
Security
- Password flow regression-fixed from day one: only the 5-hex SHA-1 prefix is accepted (the production Oct-20 rewrite had broken this contract).
- Display-image proxy is path-traversal-guarded and domain-registered.