Repository navigation
Added
- Live cloud collection — CloudTrail (S3 trails or 90-day history), Microsoft 365 audit log, Entra, Azure Activity and GCP audit logs collected as manifest-backed evidence (part of #2313, part of #2314, part of #2315)
- Durable cloud collection jobs — resumable Cloud Collection panel; every page SHA-256'd into an append-only manifest; complete, partial or failed results (closes #2311)
- Cloud collection quotas and window splitting — pre-check estimates, capped windows split, a per-tenant Microsoft query budget, disclosed cutoffs (closes #2312)
- Cloud account vault — AWS accounts, Microsoft tenants and GCP projects in their own Settings tab; secrets sealed with a separate key file (closes #2307)
- Per-case cloud accounts — in team mode only writers of the attached cases see or use an account; the public demo refuses cloud routes (closes #2308)
- Cloud credentials end with the case — closing offers Delete or Keep, archive and delete always remove them; held-credential list with audit and custody (closes #2347)
- Cloud account read check and minimum policies — Check access tests one read per source; minimum AWS, Microsoft and GCP policies shown at add time (closes #2309)
- Cloud TLS and proxy settings — cloud SDK traffic follows
DFIR_CLOUD_CA,DFIR_CLOUD_INSECUREandDFIR_CLOUD_PROXY_URL(closes #2310) - GuardDuty findings — collected live from every detector region or imported, graded by GuardDuty severity with MITRE from the finding type (closes #2361)
- AWS region picker — collection walks the account's enabled regions; untick regions, skipped ones disclosed (closes #2363)
- Drop folder batches small CloudTrail files — up to 100
.json.gzper grading pass, Undo step and analysis run; SANS 504 lab 883 s → 314 s (closes #2540) - Million-row cloud imports — large and gzip cloud exports stream from disk with bounded, disclosed output (closes #2303)
- Cloud record dedup — re-imported or overlapping cloud exports add no duplicates; conflicts flagged; aggregates page back to source records (closes #2301)
- Cloud export recognition — Log Analytics, MES mailbox audit, CloudTrail console CSV, Graph audit records and Azure diagnostic exports reach the right importer (closes #2302, closes #2304)
- Mailbox rules, OAuth grants and MFA state — MES CSVs import as state observed at collection; forwarding, hiding rules and risky scopes grade above Info (closes #2305)
- Cloud tool compatibility matrix — manual page mapping MES, DFIR-O365RC, Log Analytics, Azure, AWS and GCP exports to importers (closes #2300)
- Cloud detections graded from the record's own parameters — audit bypass, POP/IMAP, impersonation, DLP removal, MFA cleared and more; dry runs and failures excluded (closes #2604)
- Terraform note on CloudTrail rows — rows whose user agent claims Terraform say so; tag-only, grades unchanged (closes #2607)
- AWS account weaknesses — root without MFA, routine root use, no Object Lock, S3 data-event gaps; in Cloud Coverage, a CSV and a report section (closes #2560)
- Trusted sources: ours / responder — label an IP, access key or principal; its routine rows get a hideable trusted-activity label, grades unchanged (closes #2561)
- Cloud identities as IOCs, with analyst labels — IAM users, keys and attached policies become IOCs; label IOCs attacker / ours / responder / unknown (closes #2543)
- Collector footprint tag — the collector's own read calls are tagged
dfir-collector, with a Hide collector calls switch (closes #2487) - Failed AWS reads folded per principal — after 10 failed read-only calls the rest fold into one Medium event; writes never fold (closes #2481)
- One synthesis seat per distinct cloud operation — on an over-budget cloud timeline each unseen operation gets a row before repeats (closes #2585)
- Look-alike M365 account and app names — a new user or app named like Microsoft, the tenant or an admin is High (closes #2557)
- Azure caller names — a service-principal caller shows the name learned from Entra uploads when exactly one is known (closes #2490)
- MFA fatigue detection and cloud rule gaps — denied MFA prompts then a success is High; new AWS/GCP and Microsoft-Analyzer-Suite rules (closes #2306)
- Suzaku CloudTrail rules — Yamato Security's ~740 CloudTrail Sigma rules bundled with their own switch and severity floor; Sigma gains
exists(part of #2499) - Sigma grades Microsoft 365 and Azure Activity — Suzaku M365/Azure and SigmaHQ Azure Activity rules graded at import (closes #2500)
- Sigma grades Google Workspace — Suzaku and SigmaHQ Workspace rules graded per event at import (closes #2522)
- Suzaku rule updates — check, review, install and roll back newer Suzaku rules, each file verified against GitHub (closes #2508)
- SigmaHQ main-branch rules — the check counts unreleased main-branch changes; an opt-in install brings them in, verified (closes #2518)
- GCP detection rules — SigmaHQ GCP rules bundled, plus 9 graded and 6 tag-only rules from Panther, Sentinel, Chronicle and Rhino (closes #2346)
- AWS detection rules from TrailDiscover — 12 graded rules and 211 more operations tagged by ATT&CK tactic (closes #2345)
- Detection rule pack — curated tagger rules publish as
rules-*releases; check, install, measure and roll back without an app release (closes #2349) - New built-in rules for edited rules — an edited tagger copy takes new or changed built-in rules while keeping the analyst's edits (closes #2344)
- Detection rules box — one line each for tagger and Sigma, with Review update… only when a newer release exists (closes #2389)
- Detection-overlap check — CI fails when two grading paths disagree on the same cloud sample, unless allowlisted with a reason (closes #2348)
- Upstream rule watch — a weekly job opens an issue when the Microsoft-Analyzer-Suite lists change (closes #2343)
- CrowdStrike Falcon importer — Event Search / LogScale / FDR / Next-Gen SIEM exports with host, user, IP, domain and hash IOCs and logons (part of #2592)
- CrowdStrike Falcon alerts and Next-Gen SIEM data — alerts keep Falcon's severity; third-party logs stored in Falcon go to their own importer (closes #2592)
- CrowdStrike Falcon (CQL) hunt queries — an eighth hunt language in the hunt modal and Query Translator, values escaped (closes #2558)
- Hunts in the case's own query language — the main evidence source picks CQL, KQL, ES|QL, SPL or VQL for hunts and next steps (closes #2559)
- Main evidence source in New case, Settings and AI next steps — set it at creation or in Settings → Server sets; synthesis names that tool (closes #2594)
- Light-task AI model — optional cheaper model for CQL/KQL/ES|QL/SPL hunts and query translation; Diagnostics names each task's model (closes #2593)
- Preview what the AI will see — from the Anon dialog, the first request any AI action would send, clear and masked side by side; nothing sent (closes #2491)
- AI preview as a diff — changed lines red and green, paired tokens, a summary by kind and a show-only-changed switch (closes #2531)
- AI preview for imports, playbook hunts and decision-model checks — more AI actions previewable on the shared masking step (closes #2515)
- All my cases service tokens — one capture token for every case its owner can edit, capped at capture + write, 90 days at most (closes #2474)
- Skip screenshots — a Skip button clears an AI screenshot error without a retry; the screenshots stay as evidence (closes #2498)
- Super-timeline Group repeats — optional view folding identical runs into one ×N row; nothing is merged (closes #2391)
Changed
- Sigma rules settings — split into shared, SigmaHQ and Suzaku sections; the rules table gains a Rule set column and filter
- Faster S3 trail collection — each trail folder listed once, downloads overlap; about 7× faster (closes #2380)
- Sigma "not graded" wording —
Sigma (medium, not graded)instead oftag only, with the reason on hover (closes #2386) - Faster large CloudTrail imports — 400k records 693 s → 509 s, same output (closes #2362)
- CloudTrail imports stay incremental after a summary or object-logging note — SANS 504 lab: 3,567 full merges → 1 (closes #2544)
- Default server set name — "Default" instead of "Testing" on a new install
Fixed
- "Any address" only for a real address — a malformed
/0security-group source is shown as written, without the any-address claim (closes #2625) - Report Case Details stay on one line — a newline in a case field no longer forges a heading or table in Markdown, HTML or DOCX (closes #2618)
- No
[type-seed]marker in reports — seeded questions read cleanly in report, hand-off, Notion and IRIS (closes #2621) - DNS domains checked before they become IOCs — CrowdStrike and GuardDuty DNS names that are not domains stay on the row only (closes #2620, closes #2626)
- No IP fragments in auto-detected internal domains — IP-literal assets no longer yield a bogus parent domain (closes #2624)
- Narrative edit cannot overwrite a fresh generation — Edit is locked while generating; Generate asks to save or cancel an open edit (closes #2623)
- Second Look card refreshes after synthesis — no reload needed to see the new record (closes #2622)
- Folder browser explains unreadable folders — 403 / 400 with a clear message instead of a 500 (closes #2617)
- Faster Mark False Positive on big cases — no longer rewrites the whole forensic timeline (closes #2616)
- Live dashboard updates no longer send the whole case — a small change notice, then a paged fetch; memory no longer grows per dashboard (closes #2615)
- Cloud rows show the deciding value — mailbox audit settings, security-group port and source, share principals; three new graded rules (closes #2605)
- Separate cloud accounts are boundaries, not dwell — quiet time between accounts or tenants is a Low boundary with no T1070, everywhere (closes #2606, closes #2610)
- Synthesis fallback notice — Findings card, run record and report say when the fallback model wrote a synthesis (closes #2603)
- T1552.005 on instance-credential use — instance-role calls from the remote-access source carry it, grade unchanged (part of #2607)
- Offensive-tool sign-ins and MFA disable from UAL exports — AzureHound-style agents grade Medium;
Disable Strong Authenticationis High (closes #2609) - Shares to different recipients stay separate — EBS snapshot and S3 policy/ACL shares to different accounts no longer fold into one row (closes #2611)
- Suggest hunts clears a stale Presidio hold — the old hold line no longer stays on screen while it thinks
- Playbook query cards are styled — CQL / KQL / ES|QL / SPL cards match the Suggested Fleet Hunts cards
- Suggested Fleet Hunts fits a phone screen — the main-source dropdown no longer pushes the view sideways (closes #2595)
- Large drops keep a flat per-file import time — no per-file log re-reads, a debounced duplicate-host check; SANS 504 drop 311 s → 179 s (closes #2584)
- Imported names can no longer forge product notes — note-shaped text shows as ⟦ ⟧ (closes #2574)
- M365 BEC pairings join across batches and uploads — delete-after-forward and new-admin are raised however the export was split (closes #2570)
- Hunt queries escape evidence values per language — a crafted path or URL can no longer break or widen a query (closes #2572)
- M365 audit records no longer doubled — Splunk or overlapping UAL exports of one record are one record (closes #2548)
- Anonymizer cannot be tricked into restoring a real value — mask-shaped text is never reused as a token (closes #2569)
- AI preview marks only real mask tokens — planted
ANON_…text is not styled as masked (closes #2566) - Live collection coverage row shows collection totals — counts summed across pages, later caps listed (closes #2546)
- Look-alike checks catch rn / vv / cl swaps —
jsrnithandarnazon.comare flagged (closes #2568) - EC2 instance credentials used away from the instance — High from two or more public addresses, a Medium note for one (closes #2545)
- US-format dates in M365 CSV exports — read in the file's day/month order, removing a false coverage gap (closes #2552)
- Cloud-only cases get cloud questions — no Windows next steps on a tenant-only case, plus pinned cloud key questions (closes #2555)
- M365 sign-in failures and permission grants read in full — errors named, the whole grantee shown (closes #2553)
- AI findings carry the techniques their events support — a parent becomes the cited sub-technique; unsupported ones are dropped (closes #2556)
- M365 BEC actions graded as attacks — outside forwarding, FullAccess/SendAs, hiding rules and new admins grade High or Critical (closes #2550)
- Audit logging turned on is no longer High — only turning it off is; Microsoft's own upkeep is Low (closes #2549)
- M365 Security & Compliance alerts named and graded — by vendor severity, one row per alert (closes #2551)
- CloudTrail attack steps above the AWS-service noise — SANS 504 S3 lab rows above Info ~15,400 → 37; disproved ATT&CK tags withheld (closes #2542)
- S3 versioning check raised across files — High after its own admin grant or new key, in either order and in separate files (closes #2563)
- Dated object-logging note agrees on every import — the incremental merge matches a full rebuild (closes #2564)
- Cloud imports stay incremental after S3, metadata, storage-key and EC2 rows — no forced full rebuild (closes #2536, closes #2539)
- Live collections no longer push uploads out of Cloud Coverage — one growing row per account (closes #2538)
- GitHub rate limits explained — rule checks name the hourly limit and the retry time in UTC with its date (closes #2520, closes #2525)
- Sigma Check spends less of GitHub's hourly limit — 2 calls instead of 12 when nothing changed (closes #2526)
- Faster live cloud collection import — each page commits only its own rows; 20,000 AWS records 340 s → 160 s (closes #2504)
- Coverage notes survive a High gate — the four coverage notes stay on the forensic timeline at Medium (closes #2497)
- One credential history per collection — AWS lineages built once per run, no duplicate or false "no issuance" rows (closes #2482)
- Microsoft, GCP and Azure summaries per collection — built once, so a privilege path split across pages can reach High (closes #2493)
- Faster large cloud imports — compiled tagger rules reused across pages; 100k AWS records 54.7 → 36.5 min (closes #2478)
- M365 and Azure rows cite their source record — record id and file hash instead of a legacy citation (closes #2485)
- Entra rows read correctly — no "NotAgentic" targets; sign-ins show interactive or not, script agents tagged not graded (closes #2484, closes #2483)
- Clearer cloud row wording — Sigma rows keep their description, hidden Entra risk explained, short GCP titles (closes #2486)
- No "[aggregated: 1 row(s)]" — a single budgeted cloud row keeps its description (closes #2475)
- One Azure call, one row — Started / Accepted / Succeeded fold into one counted row; failures stay separate (closes #2476)
- Readable AI failures — a schema-failing AI reply shows a plain sentence in every AI action (closes #2459, closes #2495)
- Close case dialog — Escape no longer hides it mid-send; a never-closed case says "still open" (closes #2463)
- Busy dialogs stay open — delete, import and integration dialogs ignore Escape, Cancel and backdrop while running (closes #2494)
- Settings clear and on/off fields — Clear buttons for secrets, emptied fields remove their key, on/off settings are drop-downs (closes #2465, closes #2496)
- Case record accuracy — restore explains a closed case, false-positive marks name the analyst, undo/redo in "Since your last review" (closes #2466)
- Zero-event AI import warning — an AI import that keeps 0 of N rows says so (closes #2460)
- Hunt Workbench grammar — the reference and field catalogue load when a case opens (closes #2461)
- Untrusted team-mode certificate — the extension names it instead of saying "companion offline" (closes #2473)
- Quiet console — routine requests answer 200 instead of logging errors; hunt cancel is sent before the abort (closes #2462)
- Empty states — messages fit the case; the timeline names a view's severity floor (closes #2468)
- Host-ranking badge contrast — passes in light and dark (closes #2458)
- Phone and tablet widths — every built-in view fits at 375 and 768 px (closes #2457)
- Screen-reader names — view and lifecycle buttons announce their state; every select is named (closes #2469, part of #2492)
- Manual matches the dashboard — views, page sizes, swimlane export and the hunt-platforms setting (closes #2467)
- Cloud audit row details — Azure role grants name role, assignee and scope; Entra rows list changed properties (#2503)
- Revoked service tokens show as revoked — with the revoke time, not a live Revoke button
- Honest cloud collection progress — download and import steps, saved-file counts and an ETA that says how long is left (closes #2470, closes #2384)
- Cancel imports and Second Look — a running or queued import can be cancelled and rolled back; Second Look from the Jobs panel (closes #2425)
- AI runs as jobs — Sigma re-grade, Jev review and containment check show in Jobs and wait for synthesis or import (part of #2425)
- GCP access problems named — a missing Private Logs Viewer role or a disabled Resource Manager API is said plainly (closes #2450)
- Cloud Check access fixes — offered on the open case's accounts; GCP and Microsoft checks no longer show AWS-only fields
- Settings layout — field rows line up, checkboxes sit beside their labels, the Microsoft form shows only PEM or PFX fields
- Case saves on newer Node releases — no "Value is too large" when an indexed timeline row is rewritten (closes #2445)
- Password and access changes end live dashboards — password change, sign-out, disable, demotion or role removal close them (part of #2441, closes #2449)
- Dashboard case switching — late answers for a case you left are ignored; buttons refuse a case no longer on screen (closes #2442)
- Timesketch replace keeps the old timeline until the new one is in — deleted only after every upload succeeds (part of #2414)
- Closed cases stay frozen — every change outside a reviewed allowlist is refused; closing cancels running AI work (closes #2437)
- Live case lifecycle — another analyst's close, archive or restore refreshes every open dashboard (closes #2443)
- Capture replay memory — a compact digest per capture, about half the memory (closes #2446)
- Linear-time test flake — scan-speed tests measure CPU time, so slow runners no longer fail them (closes #2444)
- Steady Cloud Collection form — polls keep the analyst's edits; regions in a wrapping grid (closes #2379)
- GuardDuty detector pagination — every page read, bounded and disclosed (closes #2372)
- Cloud credential release custody race — custody matches the outcome under concurrent changes (closes #2371)
- Cloud collection survives a briefly locked file — a Windows rename lock is retried (closes #2370)
- Sigma re-grade preview memory guard — refuses with a clear message instead of risking a crash (closes #2366)
- Cloud grading checks the service, not just the operation name — a Transfer Family
CreateUseris not IAM user creation (closes #2358) - Cloud import follow-ups — file hashes on whole-file rows,
.gzre-grade, no replay duplicates, early hang-up on refused uploads (closes #2353) - Tagger rules on Sigma-matched cloud rows — they match again; Run tagger repairs existing cases (closes #2350)
- Detection rule fixes — the real
FileSyncDownloadedFulloperation; un-sharing a snapshot is Low (closes #2338) - Sigma and failed AWS calls — not-found logging/Config failures keep Low; access-denied can still raise (closes #2351)
- Streamed cloud imports — cancellable, and rows carry the source-file hash (part of #2337)
- Cloud collection status — a stop before the first page is partial; resume no longer repeats members (part of #2339)
- Empty CloudTrail S3 window — completes with 0 files instead of failing (closes #2352)
- Collector gaps — empty GCP sink windows are not imported; refused Azure subscriptions are skipped and named (part of #2340)
- Large and gzip uploads — files over 32 MB and every
.gzupload raw (DFIR_MAX_UPLOAD_MB); the drop folder streams them (closes #2336) - Vault temp files — a failed vault write leaves no temp copy of secrets (closes #2342)
- Too-large import hint on Node 22 — the size hint is shown again (closes #2316)
- Large CloudTrail import by path — objects over 256 KiB are detected as AWS (closes #2317)
Verifying downloads. Every release carries SHA256SUMS.txt. On Linux run
sha256sum -c SHA256SUMS.txt --ignore-missing in the download folder. On Windows run
Get-FileHash <file> -Algorithm SHA256 and compare the hash with the matching line in
SHA256SUMS.txt.
Signing status. The Windows EXE and the Linux AppImage are not code-signed, so Windows
SmartScreen warns on first run. The Firefox extension zip is unsigned until it is installed
from addons.mozilla.org. The Chrome extension is signed when installed from the Chrome Web Store.