# Coverage What each check reads and where its limits are. This is not a measure of real-world detection or false-positive rates. Run `--describe` for the full check catalogue. Each report lists the checks that actually ran and the gaps found on that input. Browsers are looked for at their default locations (on Linux, also common Snap and Flatpak packages, snap data in its current revision folder, and Firefox 147's `~/.config/mozilla/firefox`): - **Chromium-based:** Chrome, Edge and Brave (stable, beta, dev, canary and nightly channels), Chromium, Vivaldi, Opera, Opera GX, Arc and Yandex. - **Firefox-based:** Firefox, LibreWolf, Waterfox, Floorp, Zen, Pale Moon and Tor Browser, each reported under its own name. Portable installs elsewhere are not discovered. - **Safari:** macOS only. | Check family | Reads | Limits | |---|---|---| | Chromium browsers and desktop apps | History, bookmarks, cookies, favicons, site storage, connection records, HSTS, download URL chains; separate standard Linux/macOS browser caches (and Opera's on Windows); raw app caches in full mode. Desktop apps are found by folder contents with no fixed depth limit | Open/locked files, deleted data, schema changes. Bookmarks and raw cache strings are references. Download hosts are attempted or contacted according to status and bytes received; neither proves execution. Explicit targets do not expand into an unselected separate cache. App discovery is bounded by folder count and time; reaching the bound is a gap. | | Firefox family | History, cookies, favicons, cache, text and binary HSTS stores | Partition metadata is not treated as the contacted host. | | Desktop Safari | Main and profile `History.db` visits on macOS, or selected copies. From the Safari sandbox and `~/Library/Safari`: cookies and HSTS (contacted); tabs, iCloud tabs, URL cache, WebKit `observations.db`, bookmarks, Top Sites and recently closed tabs (references); downloads (contacted or attempted by bytes received) | Unknown visit origin or load status stays a reference with a gap; saved references and synced visits do not establish local contact. Damaged stores are gaps. Selected copies cover `History.db` only. | | File fingerprints | Platform-specific user, application and temporary roots; Windows candidate extensions; on Linux/macOS also extensionless/executable files, native libraries and scripts; all extensions in full/targeted mode | Exact known hashes only; 512 MiB per file; inaccessible paths. Full mode does not cover all drives or accounts. | | Windows host and existing traces | DNS, COM, download origins, Win+R, Defender, PowerShell event logs, ShimCache and PCA | Logging may be disabled; records expire. ShimCache shows presence, not execution. | | Windows antivirus status | Product names, states and signature status from Windows Security Center, with no vendor allowlist | Windows 8 or later desktop API; Windows Server and unregistered products are not covered. Reported state is not an independent health test or proof of protection at past event times. API failures are gaps. | | Added Windows records | PSReadLine, Explorer TypedPaths, Run/RunOnce, services, Startup files including supported `.lnk` metadata, task actions; opt-in Sysmon/Security events | Rights and logging affect coverage. BAM and SRUM are unsupported. Saved input and arbitrary Startup text are references; recognised batch commands, Internet Shortcut URLs and literal shell-link launch metadata are persistence configuration. Targets are never accessed or resolved. Shortcut IDLists, unfamiliar blocks/flags, ambiguous ANSI encodings, expansions, unsupported batch syntax and other unknown formats are gaps. | | Windows Prefetch and Amcache | Supported SCCA Prefetch layouts and allocated Amcache hive records, live on Windows or from selected copies on any OS | Prefetch executable names are `execution-record`, not full-path or content identification; metric paths are references. Amcache is `inventory-record`, never proof of execution or current presence. Compressed Prefetch needs Windows; dirty hives need transaction-log replay, which is unsupported. Locked files, unsupported layouts and exhausted bounds are gaps. | | Linux startup configuration | Standard user/system autostart `.desktop` files, systemd units and one-level `.d/*.conf` overrides, readable cron configuration (including BusyBox `/etc/crontabs`) and hourly/daily/weekly/monthly script directories. Links are not followed; masked units and aliases of inspected files are not gaps | Unit/cron text and network strings are references. Only supported unconditional desktop executable paths can be configured evidence. No activation or effective-override resolution, recursive script execution, binary journal analysis or other users' homes. | | macOS startup configuration | Standard user/system LaunchAgent and LaunchDaemon XML/binary property lists, including Apple's App cryptex folders; literal `Program`/`ProgramArguments`, compared as an executable path and, by file name, as a process name | Configuration does not establish activation or execution. No managed login items, unified logs or memory. Unsupported or unreadable files are gaps. | | Unix shell records and text logs | Standard Bash/Zsh/fish startup files and saved history, including macOS Terminal's per-session `~/.zsh_sessions` / `~/.bash_sessions` history; selected current Linux syslog/auth files and macOS `system.log`. Executable paths are matched as shell words, including inside `$( )`, backticks and after `;`, `\|`, `&` | References only. At most the recent 4 MiB per file is kept and older records are reported as omitted. No custom history locations, unsaved sessions, rotated/compressed logs or binary/unified-log decoding. Permissions and read/count limits leave gaps. | | Unix hosts file | `/etc/hosts` on Linux, `/private/etc/hosts` on macOS | An indicator pointed at a real address is configured evidence; one pointed at a block address (0.0.0.0, loopback) is protective and only counted. | | Guided Android acquisition | Pinned official AndroidQF download, owner consent, local ZIP extraction, optional public age recipients | The collector changes device state. Windows collector unsigned; SHA-256 is publisher integrity, not an independent audit. Encrypted archives need separate decryption. | | Android bug report (made on the phone) | The bug report `.zip` (or an older phone's `.txt`) read in place: system services' state, system properties, settings (main user), logs, the app list from the package service (installer and install time), battery history of apps installed and removed, crash records, the running apps named by the activity manager; then the same checks as an AndroidQF folder | No app files, so app fingerprints and signing certificates are not compared; no file list, root-tool check, SELinux record or text messages - each listed as not checked. Recognised by its contents (the dumpstate header), not its name. Main text capped at 128 MB; a longer report is listed as not fully checked. | | AndroidQF folders | Apps, process inventories (`processes.json`, `processes.txt`, or `ps.txt` from older AndroidQF), settings, files, logs, crashes, install history and SMS/MMS backup | Collection permissions and Android versions limit what is present. JSON processes need a positive integer `pid` and a `command_line` string array; only the first command names the app. Only inbox messages are `received`; sent, draft and failed messages are references. MMS attachments are not opened. Certificate coverage gaps count SHA-256 and SHA-1 separately. Archives must be extracted first. Encrypted SMS needs the password; legacy password encodings may be unsupported. | | Android intrusion logs | Decrypted newline-JSON `.txt` exports and supported ZIP members: DNS events, TCP connection attempts, process-start security events | Logging may be off or absent. DNS is `dns-record`; connection attempts and process records are `device-log`, not proof of successful contact. Unknown events/schemas and encrypted input are gaps. No live collection or decryption. | | iPhone local evidence | Local backups, independently encrypted artifacts/WALs, extracted file systems and selected sysdiagnose TAR/directory artifacts: messages, profiles, browser and process records, running-process lists, installation logs, metadata and diagnostics (see [IPHONE.md](IPHONE.md)). Table layouts of iOS 15, 16, 17 and 18 are in the test suite | Unsupported schemas or encoded-message variants, damaged WALs, encrypted root-manifest WALs, unavailable protection classes and record caps are gaps. No live/cloud acquisition, deleted-record recovery or unified-log decoding. Source hashes identify bytes read, not device authenticity. | | Wi-Fi settings | Windows: saved profiles exported without keys (authentication, encryption, automatic connection, hidden network); Linux: NetworkManager security type, automatic connection and hidden setting; macOS: security of the network in use. Nearby look-alike networks from the system's last scan | Prevention advice only, never a finding. The macOS saved list needs administrator rights; only NetworkManager on Linux. Router settings are not checked (that would mean contacting the router). A computer without Wi-Fi is not a gap. Look-alikes are seen only if in range during the last system scan. | | Certificates and network settings | Trusted root certificates (Windows stores; macOS custom trust settings; Linux local anchors) against the Microsoft and Mozilla CCADB lists; system proxy and proxy script; DNS servers; the router's hardware address in the neighbour table | Leads only. Browser-specific stores (e.g. Firefox's own) and per-app proxies are not read. An unrecognised root is named by kind from its subject, which the certificate's maker chose. | | Link or message check (separate) | Every link and bare address in pasted text, defanged spellings included, against the indicator lists (with parent domains) and documented lure patterns: Signal linkdevice links and look-alike Signal/WhatsApp/messenger sites, QR and pairing-code wording, Predator-style login SMS, Pegasus-style short paths and random sub-domains on high ports, shorteners, raw IPs, punycode and mixed scripts, '@' decoys, app/profile/program downloads | Text comparison only: nothing is opened or resolved, so a link's real destination behind a shortener is not known, and a brand-new address that copies nothing is not recognised. Look-alike detection covers the commonly impersonated services listed in the code. | | Extraction check (separate) | AndroidQF collection: adb_keys (authorised computers), adb_enabled and developer options, ro.boot.verifiedbootstate / ro.debuggable / ro.secure, side-loaded apps holding accessibility services, Cellebrite's imitated USB devices in logcat/dmesg/bugreport (id together with the published version), app ids against the indicator lists. iPhone backup: Info.plist / Manifest.plist details, backup encryption, published Cellebrite iOS strings | An extraction can leave no trace at all, stated in every report. A backup does not hold the iPhone's pairing records, so trusted computers cannot be seen (the report says how to clear them on the phone). adb_keys needs root to collect. Vendor-specific indicators are only those published in the cited reports. | | Email check (separate) | A saved message (.eml / raw source): From vs envelope sender vs Reply-To, Authentication-Results (SPF/DKIM/DMARC/compauth), Received path, display-name imitation and look-alike letters, link href vs visible text, attachment types (runs / macros / archives / double extension / RTL override), pressure wording; links and text through the link check | Only as good as the copy you save - a forwarded message loses the provider's own verdict. A real account that was taken over passes every check. Attachments are described, never opened (use the file check). No S/MIME or PGP signature checking. | | File check (separate) | One file, read only (first 256 MB for content checks): SHA-256/SHA-1/MD5 and spyware addresses inside against the indicator lists; real type from its bytes; deceptive names (double endings, right-to-left characters, padding); Windows signature, imphash, packer sections, what a Windows program says it is (company, description, original name: a famous maker its signature does not back, or a Windows tool saved under another name); who signed a Mac program (developer and Team ID, ad hoc, or not signed); .lnk target and arguments (judged on what it runs, not its description); ZIP directory (programs/shortcuts/disk images anywhere, scripts at the top or in small archives, password protection, nested archives), TAR and 7-Zip listed (7-Zip's own compressed list is unpacked; locked file names are said to be locked), RAR/gzip named; ISO 9660 / Joliet root folder, UDF, VHD, VHDX and DMG recognised; Office: macros, Excel 4.0 macro sheets, embedded objects, external relationships (remote template, Follina, external objects; ordinary hyperlinks ignored), DDE, Equation Editor class id (also in RTF); PDF: JavaScript (with or without opening actions), /Launch, embedded files, XFA, also inside FlateDecode streams (32 MB cap); web pages and SVG: HTML smuggling, encoded payloads, password boxes, scripts; OneNote embedded files; Android app package and permissions | No unpacking of archives, no RAR listing, no listing inside VHD/VHDX (worth a look: Windows mounts them on a double-click) or DMG/UDF (reported as incomplete because their contents were not seen, as are RAR and gzip), an ISO image is listed at its top folder only; old binary Office internals (Excel 4.0 sheets in .xls, OLE Package objects, attachments inside .msg) are not parsed; MSI installers are named like programs, not judged, no YARA, no execution or emulation. Encrypted Office files and PDFs are not decrypted. A clean result means nothing known or obviously risky was seen, not that the file is safe. | | QR code check (separate) | PNG (8/16-bit, grey, colour, palette, alpha), GIF (incl. interlaced), BMP (1/4/8/24/32-bit), PPM/PGM/PBM; Otsu threshold, finder-pattern alignment, all four rotations and mirrored, versions 1-40, levels L/M/Q/H, all four text modes, Reed-Solomon repair with a refusal beyond capacity; the text then goes through the link check | JPEG and HEIC must be saved as PNG first. No camera. Interlaced PNG, very skewed or curved photographs, and codes smaller than 21 modules across are not read. Micro QR is not supported. | | Hash look-up (separate) | Files and folders given (up to 20,000 files); SHA-256/SHA-1/MD5; kind from magic bytes; PE imphash (pefile-compatible, ordinal names for ws2_32/wsock32/oleaut32) and Authenticode signer (read from the embedded certificate; validity checked offline on Windows); offline indicator match; opt-in CIRCL bulk SHA-1, MalwareBazaar, ThreatFox, VirusTotal v3, Hybrid Analysis v2; local cache (30 days found, 3 days unknown) | No similarity hashes (TLSH/ssdeep). Catalogue-signed Windows files show no signer name. Programs with more than ~4,096 imports get a complete imphash where pefile's is truncated. Online answers reflect each service's data at that moment. | | Close the doors found open (separate) | Windows: Remote Desktop (fDenyTSConnections), automatic sign-in (AutoAdminLogon), firewall (all three profiles), SMB1 feature. Each: current value read, JSON undo record written first, one elevated command, value read back, result recorded. Managed-machine check (dsregcmd, MDM, macOS profiles) | Does not touch disk encryption, Secure Boot, accounts or stored passwords - those are explained instead. macOS and Linux print the sudo command rather than running it. SMB1 needs a restart to finish. An undo record cannot restore a password that was never read. | | Windows' own records (separate) | Live: Get-WinEvent over PowerShell/Operational, Windows PowerShell, TaskScheduler, WMI-Activity, Defender, TerminalServices (2), RdpCoreTS, BITS, CodeIntegrity, AppLocker, Sysmon, System, Application (+ Security with administrator rights). Files: own .evtx reader (any OS). 30 rules; base64 `-EncodedCommand` decoded before matching; indicator lists matched against every record; findings grouped with counts, first and last seen | Only what the logs still hold, and a small log holds little. Most channels are not switched on by default on Home editions, so absence proves nothing. Process creation needs Sysmon or audit policy. No live memory, no network capture (see the recording check). Rules are our own, not a full attack catalogue. | | Send evidence to a helper (separate) | Files and folders the person adds (links not followed, network paths refused); AndroidQF collection detected and kept at the top; age v1 X25519 encryption streamed (no plaintext on disk); Ed25519-signed manifest (SHA-256, size, modified time); acquisition.json + hashes.csv; safety number and package code; `--verify-package` | Proves who sent a package only if the safety number is compared by phone or in person. Files are copied as they are at packaging time, not as a forensic image. No STIX export. | | What starts automatically (separate) | Windows: Run/RunOnce (HKCU, HKLM, WOW64, policy) with Task Manager's StartupApproved switch, Startup folders (.lnk and .cmd/.bat resolved), scheduled tasks (schtasks export, no admin), auto-start services (ServiceDll for svchost) and drivers, Winlogon Shell/Userinit, IFEO Debugger, AppInit_DLLs; offline Authenticode (embedded + catalogue). macOS: LaunchAgents/Daemons outside /System, BTM background items, TCC grants. Linux: autostart, enabled systemd units, cron. SHA-256/SHA-1 indicator matches; unsigned in user-writable folders, encoded PowerShell, downloaders, remote-access programs, hijacks, missing targets, unquoted paths | WMI event subscriptions, COM hijacks (the scan's COM check covers per-user ones), browser extensions (Doors to close), Winsock/LSA providers and Office add-ins are not listed. macOS BTM and TCC need sudo / Full Disk Access. Signatures name no publisher on Windows. | | Network recording check (separate) | pcap / pcapng (several interfaces; Ethernet, raw IP, Linux SLL/SLL2, BSD loopback); DNS questions and A/AAAA answers (other record types skipped safely), TLS ClientHello SNI with TCP re-assembly, HTTP Host, QUIC by address only (never decrypted), PCAPdroid per-app trailer; indicator matches with parent domains and IP-to-name from DNS answers; clockwork contact, dynamic-DNS names, TLS on unusual ports, plain HTTP, encrypted DNS | Only the time recorded. Encrypted DNS (DoH/DoT), Encrypted Client Hello and QUIC hide names. Server certificates are not examined. | | Doors to close | Windows: Remote Desktop, automatic sign-in (stored-password flag only), password on wake, firewall profiles, BitLocker/device encryption, Secure Boot, SMB1, local administrators, driver files (System32\drivers and every kernel-driver service's file, local paths only, 64 MB per file) fingerprinted against LOLDrivers' verified vulnerable/malicious list (Apache-2.0), the vulnerable driver blocklist setting, memory integrity running vs configured (Device Guard report, registry fallback); macOS: FileVault, firewall, SIP, Gatekeeper, listening SSH/VNC ports, auto-login; Linux: LUKS root, listening SSH, display-manager auto-login; installed program names against our remote-access/monitoring list plus entries hand-picked from LOLRMM (Apache-2.0); file names in Downloads (two levels), Desktop and Documents against those programs' own file names (names only, links not followed, 5,000 entries per folder); Chromium and Firefox extensions; signed-in browser accounts (masked); iPhone passcode and Lockdown Mode from the backup; Android updates, encryption, Play Protect, debugging | Prevention advice, never findings. Programs installed under another name, portable copies renamed or kept elsewhere, and browser-based remote access are not recognised. If none of Downloads, Desktop and Documents can be read, the download check says so instead of reporting nothing found. A setting that cannot be read is listed as could not check. Accounts are a guided checklist only: they cannot be seen without signing in, which this tool never does. | | DNS logs (`--dns-log`) | dnsmasq/Pi-hole text logs, Pi-hole FTL database, AdGuard Home querylog.json, BIND, Unbound, Windows DNS Server debug logs, Zeek dns.log | Only the period the log covers; devices using their own DNS (DoH, VPN, mobile data) are absent. A lookup is dns-record, a filtered lookup attempted; neither proves contact. Files over 4 GB or time limits leave the rest as a gap. | | Home network (`--home-network`, separate) | Local /24 only, private addresses only: neighbour table after one packet per address, mDNS and SSDP answers, TCP connect to ~25 ports per device, the router's UPnP description and port mappings; the router's answers to five test lookups (fixed names, a made-up name, the resolver it uses) | Asks first. Devices that are off, asleep or on another network are not seen. Open ports are reported, not tested further; nothing is logged into. Router lookups: a router that does not answer DNS (or a VPN blocking it) is a gap; a filter the person chose can explain a wrong answer; the resolver's owner is named only for well-known public resolvers. | | Household report (`--combine`) | Saved full JSON reports, sharing summaries and home-network JSON from this program and the phone-check page | Summarises; the device reports hold the details. A sharing summary has no findings detail. Kinds of check not supplied are listed as missing. | | iPhone backups (window button; `--make-iphone-backup` on Linux) | Windows/Mac: opens Apple's app and watches the usual backup folders for a finished new backup. Linux: a verified pymobiledevice3 helper makes an encrypted full backup | Backups saved elsewhere are not watched (choose the folder instead). The Linux helper needs usbmuxd; it is tested to install and start with Python 3.11 and 3.12 in Linux containers, not yet against a real iPhone. | | Bluetooth trackers (`--trackers`, separate) | Find My (AirTag), SmartTag, Tile and Find Hub advertisements, with "away from owner" where the format says it | Receives only; needs Bluetooth on the computer (the programs include the library; the Python version needs requirements-bluetooth.txt). Trackers out of range, off or silent are not heard, and one whose owner's phone is close by often stays quiet (a tracker following you is away from its owner and calls out every few seconds); one place is not enough to tell travelling with you from nearby. | | Out-of-date software | Installed versions (Windows, macOS) and the iOS version an iPhone backup records, against CISA's Known Exploited Vulnerabilities list and the fixed versions in each CVE record, per release line | Needs the list saved by `--update` (the web page ships one). A version on a release line the record does not list is not compared. Prevention advice only. | | Indicator files | Curated sets, downloaded feeds and STIX bundles; a STIX `file:name` holding a full path is used as a path; built-in copies of the 14 MVT-hosted files (exactly the bytes this release pins, used when no downloaded copy loads); downloaded known-good lists (MISP warninglists) that set aside an upstream indicator only when it exactly equals a listed value, or falls inside one of a few fixed address ranges | Missing or refused sources are reported, and each source's age is shown; stale or undated sets are flagged. An update that would cut a source's indicator count by more than half, or to zero, is refused and the previous copy kept; `--update --accept-shrink SOURCE_ID` accepts a genuine change. | | Built-in set of spyware used against journalists, activists and minorities (`iocs/curated_targeted.txt`) | SHA-256 file fingerprints, or SHA-1 where the publisher gives only that (file check, Android app files and file lists), and Android package names (installed apps, process and log records) from ESET, Volexity, Cisco Talos, SentinelLabs, Meta, Palo Alto Networks Unit 42, Lookout and Amnesty International Security Lab; each line names its source, and each publisher's licence notice is kept in the file header | Exact builds only: a rebuilt or updated sample has a new fingerprint. No web or IP addresses from these reports, because old ones may now belong to someone else. Package names that a real app could share are left out. MD5 values are not used. | | Context and related records | Publisher assertions, validity dates, withdrawal labels and matching records across artifacts | Publisher labels are not independent corroboration. A validity period is not necessarily a campaign period. Related records do not establish causation. | | Report inspection and saving | Completed findings, references, gaps and fixed advice; readable report, full JSON and minimal sharing summary | No extra detection happens in the menu. Quitting before a scan returns `4`, not a clean verdict. Reports never replace existing files in selected evidence or in app data and app cache folders. | | Private evidence handoff | Full reports, indicator fingerprints and explicitly selected local files in a ZIP with an integrity manifest | Unencrypted and never uploaded. Up to 128 files, 64 MiB each, 256 MiB total payload including reports. Files are copied at export time, not kept from the scan; hashes do not prove original authenticity or infection. | | Optional file patterns | The separately documented Candiru rules ([PATTERNS.md](PATTERNS.md)) | Experimental; no memory scan or automatic dependency installation. |