2026-10-18T0256Z-intel
One pipeline fire, in full · intel run of 2026-10-18 · sub-agent allocation and telemetry, per-iteration verification verdicts and findings, source-list edits, coverage gaps, bridge invocations, and the run's own verification & coverage notes: what was published, what was dropped at the borderline or judged not relevant (and why), single-source carve-outs, and contradictions. Rendered from runs/2026-10-18/2026-10-18T0256Z-intel.md.
Run telemetry
- Items returned
- 3
- Duration
- 16m 47s
- Tool calls
- 0 WebFetch17 WebSearch95 bridge
- Cited sources
- 1 of 21 in slice
- Items returned
- 2
- Duration
- 25m 59s
- Tool calls
- 10 WebFetch47 WebSearch95 bridge
- Cited sources
- 1 of 29 in slice
- Items returned
- 4
- Duration
- 22m 25s
- Tool calls
- 10 WebFetch9 WebSearch107 bridge
- Cited sources
- 3 of 57 in slice
- Items returned
- 2
- Duration
- 22m 21s
- Tool calls
- 14 WebFetch22 WebSearch62 bridge
- Cited sources
- 1 of 12 in slice
Verification
Deep dive
·
Entries this run published (2) and updated (3)
- CVE-2026-88771 / CVE-2026-88772, Citrix NetScaler ADC and Gateway: unauthenticated pre-auth RCE zero-days exploited before a patch existed (CVSS 4.0 9.5)
- CVE-2026-105133 / CVE-2026-105134, AhsayCBS backup management console: an unauthenticated authentication-bypass and code-execution chain exploited since 7 October to drop webshells and a cryptominer; Ahsay says its earlier patches were incomplete and ships hotfixes 10.3.4.45 and 9.15.2.70 through its partner portal
- Danfoss AK-SM and AK-SC, Copeland xweb and Dixell xweb refrigeration controllers compromised, with forced defrost reported at hundreds of French supermarkets; CERT-FR and NCSC Switzerland say to remove internet exposure
- ConnectWise Cloud Backup (formerly SkyKick): the vendor's Microsoft 365 application credential was used to sign in to customer tenants, with directory-role self-assignments in about 50 tenants and added credentials in one
- Rank Math SEO PRO 3.0.123: some packages served for about eleven hours carried malware that can leave hidden backdoors, rogue administrators and cron persistence that the update to 3.0.124 does not remove
Sources changed (this run)
Edits this run made to sources/sources.json · promotions, demotions, new candidates, and fetch-method / category / reliability / url corrections (the run record's sources_changed[]). Paginated; 10 per page.
26 notes · 6 last_successful_fetch · 3 added.
| Source | Change | From → To | Reason |
|---|---|---|---|
| greynoise | notes | · → recipe note appended | 2026-10-18 (S1): `extract` of the listing returns the wrong content (one article body or navigation only); `url --direct` raw HTML is the working listing recipe. |
| hadrian-labs | notes | · → recipe note appended | 2026-10-18 (S1): `extract` of the listing returns the wrong content; `url --direct` raw HTML is the working listing recipe. |
| morphisec | notes | · → recipe note appended | 2026-10-18 (S1): `extract` of the listing returns titles without dates; `url --direct` raw HTML is the working listing recipe. |
| hpe-networking-psirt | notes | · → recipe note appended | 2026-10-18 (S1): only a single advisory URL was readable and no listing route was found; ask for the advisory id when one is known. |
| watchguard-psirt | notes | · → recipe note appended | 2026-10-18 (S1): `extract` of the listing carries no dates; read each advisory page for its date. |
| fortinet-psirt | notes | · → recipe note appended | 2026-10-18 (S1): the Known Exploited and CVE fields appear only in the raw advisory page (`url --direct`), not in the `extract` view. |
| anssi-fr | notes | · → recipe note appended | 2026-10-18 (S2): the CERT-FR version table ('Gestion détaillée du document') shows revision dates and descriptions only in the raw page (`url --direct <alert URL>`); the feed and `extract` hide revisions, so re-read each covered A |
| safeonweb-be | notes | · → recipe note appended | 2026-10-18 (S2): `url` works where WebFetch returns 403. |
| europol-newsroom | notes | · → recipe note appended | 2026-10-18 (S2): take dates only from `url --direct <full item URL>`; never truncate the feed link (a truncated URL falls to the metered reader). |
| prodaft | notes | · → recipe note appended | 2026-10-18 (S2): no dated listing on any transport; reached by WebSearch only. |
| apa-ots | notes | · → recipe note appended | 2026-10-18 (S2): no listing or feed route found; reached by WebSearch only. |
| rapid7-research | notes | · → recipe note appended | 2026-10-18 (S3): the feed https://www.rapid7.com/rss.xml is malformed (invalid token near line 367) and the reader returns nothing; `extract https://www.rapid7.com/blog/` lists /blog/post/<slug>/ links without dates and `extract` |
| ico-uk | notes | · → recipe note appended | 2026-10-18 (S4): `ico-uk enforcement N` returns lastmod, year and month per URL; news items need the sitemap filtered on /news-and-blogs/2026/10/ via `url --direct`. |
| sec-disclosures-edgar | notes | · → recipe note appended | 2026-10-18 (S4): `sec-edgar 8k <start> <end> 1.05` is the clean sweep; item 8.01 is too noisy, and a full-text query through `url --direct` on efts.sec.gov with q, dateRange=custom, startdt and enddt is the better supplement. |
| swisscybersecurity-net | notes | · → recipe note appended | 2026-10-18 (S4): `extract` of the homepage is served through the metered reader; WebFetch lists 14 dated items and is the working recipe. |
| databreaches-net | notes | · → recipe note appended | 2026-10-18 (S4): `extract` of a post is served through the metered reader; the feed carries enough for triage. |
| fuitesinfos | notes | · → recipe note appended | 2026-10-18 (S4): WebFetch of the homepage returns only 4 entries because the list loads dynamically; the full daily list is not reachable on any transport tried. |
| theregister-security-feed | notes | · → recipe note appended | 2026-10-18 (S3): `extract` returns navigation only; WebFetch reads the article body. |
| wordfence | notes | · → recipe note appended | 2026-10-18 (S3): the feed https://www.wordfence.com/feed/ works directly and surfaced the Rank Math PSA within hours; `extract` of the blog is served through the metered reader. |
| push-security | notes | · → recipe note appended | 2026-10-18 (S3): the blog now carries buyer's guides and marketing; nothing returnable in the lookback. Review at the next audit for demotion. |
| novee-security | notes | · → recipe note appended | 2026-10-18 (S3): the blog now carries marketing; nothing returnable in the lookback. Review at the next audit. |
| splunk-advisories | notes | · → recipe note appended | 2026-10-18 (S1): SVD-2026-1006 says advisories for Splunk Enterprise, Universal Forwarder and SOAR publish on 2026-10-21; rotate this source into S1 on 2026-10-21 or 2026-10-22. |
| horizon3-ai | notes | · → recipe note appended | 2026-10-18 (S1): `extract` of the attack-blogs post URL reads the full body; the listing needs WebFetch. |
| huntress | notes | · → recipe note appended | 2026-10-18 (S3): the ConnectWise/SkyKick post is a living document with dated updates inside; re-read it for later updates. |
| zscaler-threatlabz | notes | · → recipe note appended | 2026-10-18 (S3): the feed https://www.zscaler.com/blogs/feeds/security-research surfaced the NetScaler exploitation post; `extract` reads the article. |
| huntress | last_successful_fetch | 2026-10-16 → 2026-10-18 | fetched and used |
| wordfence | last_successful_fetch | 2026-09-24 → 2026-10-18 | fetched and used |
| zscaler-threatlabz | last_successful_fetch | 2026-10-09 → 2026-10-18 | fetched and used |
| horizon3-ai | last_successful_fetch | 2026-10-08 → 2026-10-18 | fetched and used |
| anssi-fr | last_successful_fetch | 2026-10-14 → 2026-10-18 | fetched and used |
| bleepingcomputer | last_successful_fetch | 2026-10-15 → 2026-10-18 | fetched and used |
| todyl | added | · → status: candidate | S4: first-hand Microsoft 365 / Entra telemetry write-ups; independent corroborating primary for the ConnectWise Cloud Backup (SkyKick) credential abuse; no feed probed, extract reads each post |
| specterops | added | · → status: candidate | S3: Active Directory / Entra identity-tradecraft research, same-day analysis of the ConnectWise SkyKick incident; working feed https://specterops.io/feed/ |
| connectwise-trust-center | added | · → status: candidate | S3: ConnectWise Trust Center advisories, vendor PSIRT channel for MSP tooling and the primary for the Cloud Backup incident; one page stacks every advisory, no per-advisory permalink |
| znuny-advisories | notes | · → recipe note appended | main: the host answered HTTP 503 to WebFetch and reset the connection from the container while source_health.py probed it; a host outage, not a recipe fault (503 never demotes), recipe unchanged |
Coverage gaps (this run)
Sources this run's brief needed that returned no usable content via any documented recipe. Bridge-recovered or quiet-day sources do NOT appear here. (Distinct from the independent source-accessibility probe at the foot of this section, which probes all active sources regardless of what any run needed.)
No coverage gaps in this run · every source the brief needed returned usable content via its documented recipe.
Bridge invocations (this run)
6 bridge calls this run · these are successful bridge fetches (separate from "Coverage gaps" above).
- url --direct ×2
- tools/kev_window_diff.py ×1
- extract ×1
- pdf ×1
- tools/sync_products.py ×1
Verification findings · all iterations
Per-iteration finding detail. Each table is one verifier pass · what was flagged, how the main agent remediated it, and the outcome. Walking the tables top-to-bottom shows the verifier's debugging trail across iterations.
Iteration #1 NEEDS_FIXES · 17 findings (truth=8, editorial=6, advisory=3) · Claude Sonnet 5.5 · 12m 47s
| F-code | Section | Item · URL/quote | Verifier summary | Remediation · outcome |
|---|---|---|---|---|
| F4 hallucinated-fact | · | Huntress update of 10/17 5:15 PM ET: 'the access that led to the credential being added to Huntress and another app came from private Azure IPv6 space, not the IP address associated with the previous | · | |
| F3 claim-not-supported | · | ConnectWise's advisory (https://www.connectwise.com/company/trust/advisories) says only 'unauthorized access to application identities' and 'We identified the issue on October 14'; it names no Graph t | · | |
| F4 hallucinated-fact | · | (low confidence) Body and cited sources show credentials added in ONE tenant ('in one the Teams app added itself to Global Administrator and a Python script then added client secrets'; Huntress: 'In o | · | |
| F4 hallucinated-fact | · | Rank Math says only 'navigate to Plugins to check your installed Rank Math PRO version'; Wordfence says 'check when version 3.0.123 was downloaded or installed' (no Plugins screen). No cited source sa | · | |
| F4 hallucinated-fact | · | (low confidence) Rank Math: 'a limited number of users who downloaded' and 'malware that could compromise your website if the plugin was activated'; Wordfence: 'some copies' / 'a subset of version 3.0 | · | |
| F3 claim-not-supported | · | The citation terminates a clause that includes 'which CERT-FR links' (and the same clause appears in the update section: 'The alert's documentation section links an operator guide ... listed on the CE | · | |
| F3 claim-not-supported | · | CERT-FR's version table says only '16 octobre 2026 | Mise à jour des recommandations' and does not show when the Documentation link was added. The PDF is served with 'last-modified: Tue, 13 Oct 2026 0 | · | |
| F4 hallucinated-fact | · | (low confidence) CERT-FR's compromised-unit branch lists 'Rétablir l'accès uniquement via un VPN sécurisé' before 'Reconfigurer l'équipement pour sa remise en service' (the entry's own Detection line | · | |
| F8 needs-more-research | · | The body cites BleepingComputer only for 'not public'; Microsoft's confirmation is never stated. BleepingComputer (2026-10-17): 'Microsoft also observed persistence activity via a backdoor after succe | · | |
| F8 needs-more-research | · | Huntress: 'October 13 to 14: Password spraying against four law firm tenants from 144.172.64[.]128 ... curl/7.88.1 ... Two accounts were locked out by smart lockout.' The body never mentions the spray | · | |
| F9 surface-contradiction | · | That condition comes from Huntress and Todyl (Todyl, 16 Oct: the advisory 'doesn't say the credential has been rotated'). ConnectWise's latest update (18 Oct 01:00 UTC) says 'The affected shared Cloud | · | |
| F8 needs-more-research | · | The new Horizon3 section (and its PoC) shows the chain on the Linux installer ('On Linux it runs as root'; its IOC list names 'the CBS JVM (Linux)'), yet Exposure, Detection and Triage still describe | · | |
| F9 surface-contradiction | · | (low confidence) Main text (Huntress; CISA KEV: 'command injection ... via manipulation of the random parameter in ... UpdateReceivers.do', CWE-77/78) places the flaw in UpdateReceivers.do. The new se | · | |
| F10 missed-angle | · | (low confidence) The new section ties NetScaler exploitation to CVE-2026-75650 infrastructure, which the store covers in 2026-09-08/stylesmuggler-cve-2026-75650-magento-adobe-commerce-rce (registry ke | · | |
| F11 editorial-advisory | · | Unclear what boundary 'matches' what; say 'no source ties the attacks to the XWEB PRO flaws the guide cites'. The record summary says the third version 'splits the guidance by state' while the section | · | |
| F11 editorial-advisory | · | 'the entry lacked' narrates the entry's own history in a reader-facing changelog summary; state only what Horizon3 adds. | · | |
| F11 editorial-advisory | · | No earlier text defines 'second-stage checks'; name the checks (post-exploitation / persistence checks) or drop the phrase. | · |
Iteration #2 NEEDS_FIXES · 7 findings (truth=5, editorial=0, advisory=2) · Claude Sonnet 5.5 · 15m 53s
| F-code | Section | Item · URL/quote | Verifier summary | Remediation · outcome |
|---|---|---|---|---|
| F3 claim-not-supported | · | The iteration-1 re-attribution left the whole chain under 'Huntress and Todyl report'. The 162-tenant count, the ~50-tenant role self-assignments and the secrets added in one tenant are Huntress's alo | · | |
| F3 claim-not-supported | · | The sentence chains a Rank Math fact and a Wordfence fact and ends on the Wordfence link only. Wordfence never mentions the Plugins screen (it says 'Website owners should check when version 3.0.123 wa | · | |
| F4 hallucinated-fact | · | (low confidence) Huntress's 10/17 update places only 'the access that led to the credential being added to Huntress and another app' in private Azure IPv6 space; for the role self-assignments it gives | · | |
| F4 hallucinated-fact | · | Stale after the iteration-1 remediation: the Danfoss entry now dates the guide 2026-10-13 (sources[] date) and its sourcing_note says 'the date given for it is the file's last modification on the vend | · | |
| F4 hallucinated-fact | · | (low confidence; outside the entry files, but the registry summary renders on the entity page) Huntress saw credentials added in ONE tenant ('In one customer tenant ... adding new client secrets to bo | · | |
| F11 editorial-advisory | · | Truth holds (the guide ties 1.12.1 to ICSA-26-057-10, 'XWEB Pro version 1.12.1 and prior' per CISA's CSAF) but 'the only link' is ambiguous and reads as a link between the attacks and the XWEB PRO fla | · | |
| F11 editorial-advisory | · | A sourcing_note should be two sentences of provenance. Danfoss still runs to five (one more was added this run; iteration 1 flagged five) and refers to 'the update section'; ConnectWise runs to three. | · |
Iteration #3 CLEAN · 4 findings (truth=0, editorial=0, advisory=4) · Claude Sonnet 5.5 · 12m 09s
| F-code | Section | Item · URL/quote | Verifier summary | Remediation · outcome |
|---|---|---|---|---|
| F11 editorial-advisory | · | (low confidence) Huntress (https://www.huntress.com/blog/vendor-credential-enables-unauthorized-access, 10/17 update) says 'Huntress has identified almost 1,000 Entra tenants with sign-in activity fro | · | |
| F11 editorial-advisory | · | (low confidence) ConnectWise's advisory says 'Preserve evidence promptly, but do not delay urgent containment while waiting for exports.' The action puts the export 'first' on a Free tenant while the | · | |
| F11 editorial-advisory | · | (low confidence) Rank Math's delete-first Step 1 (folder, wp_options entries, host block, 'complete all the following actions before proceeding to Step 2') belongs to Option 2 (manual cleaning). Optio | · | |
| F11 editorial-advisory | · | Truth holds (CERT-FR page names no flaw or vector; the guide only cites ICSA-26-057-10 as the reason for the 1.12.1 boundary). The sentence now says the same thing twice after the iteration-2 reword; | · |
Iteration #4 CLEAN · 8 findings (truth=0, editorial=0, advisory=8) · Claude Sonnet 5.5 · 12m 26s
| F-code | Section | Item · URL/quote | Verifier summary | Remediation · outcome |
|---|---|---|---|---|
| F11 editorial-advisory | · | (low confidence) 'paid' is stated by neither cited page (Rank Math calls the file a plugin download from 'your Rank Math account'; Wordfence says only 'premium plugins' about checksums in general). No | · | |
| F11 editorial-advisory | · | 'vendor-named host' can be read as the vendor's own host; Rank Math (https://rankmath.com/kb/pro-3-0-123-security-incident/ Step 1.3) names the attacker-side domain to block. 'the malicious host Rank | · | |
| F11 editorial-advisory | · | Primary source is the stacked Trust Center advisories page (Cloud Backup update on top, older advisories such as ScreenConnect below, no per-advisory permalink; BleepingComputer and Huntress link the | · | |
| F11 editorial-advisory | · | (low confidence) 'in one' is Huntress's count and the summary attributes it to Huntress. BleepingComputer (cited) also relays Microsoft observing 'persistence activity via a backdoor ... and attempts | · | |
| F11 editorial-advisory | · | (low confidence) CISA KEV and Bitdefender label CVE-2026-105134 an OS command injection (CWE-77/78) in UpdateReceivers.do via the random parameter; Horizon3's source analysis shows the code-execution | · | |
| F11 editorial-advisory | · | Zscaler's page recommends reviewing downloads from unfamiliar IP addresses and 'unexpected PHP or CGI enablement'; the /var executable and the Perl CGI handler come from its captured payloads (telemet | · | |
| F11 editorial-advisory | · | Missing verb ('being' / 'is'): the clause is ungrammatical. The Dixell PDF says the single-source-address port forward 'doit être utilisée qu'à titre transitoire'. Trivial. | · | |
| F11 editorial-advisory | · | sourcing_note runs three sentences (and an in-parenthesis aside on the CERT-FR page); the two-sentence provenance convention would drop the store-count clause. Carried as an advisory by iterations 2 a | · |
Verification & coverage notes
The run record's narrative body, verbatim. This is where the run accounts for its own judgement calls: every borderline drop and judged-not-relevant item with its reason, dedup decisions, single-source items and their carve-outs, contradictions, and per-source coverage gaps, so nothing the run considered disappears silently.
Verification & coverage notesrun record body
2026-10-18T0256Z-intel · Sonnet 5.5 · window 26 h · 2 entries published
Verification & coverage notes
Window: gap 24.0 h from 2026-10-17T0256Z-intel, window 26 h, developing-story window 72 h; the container clock agreed with the network date (skew 0 s) and the first fetch of origin/main was current. The window is a Saturday into Sunday and thin. Two new entries and three changelog records. New: ConnectWise Cloud Backup (formerly SkyKick), high, a vendor-held Microsoft 365 application credential used against 162 or more customer tenants with directory-role self-assignments in about 50 tenants and added credentials in one (S3 and S4 surfaced it independently, merged on one incident key; Admiralty A1); Rank Math SEO PRO 3.0.123, notable, a vendor-confirmed backdoor in packages served for about eleven hours, kept short on precedent (the store carries the ShapedPlugin and Awesome Motive WordPress supply-chain entries) and single-source-victim because Wordfence only relays the vendor. Records: Danfoss and refrigeration controllers (update: CERT-FR's third version of 2026-10-16 changes the compromised-unit sequence to restart first and adds the whole-configuration check, and the Dixell France operator guide gives the XWEB PRO firmware boundary and the factory Admin account; the Detection line, summary and actions were rewritten where they stood), AhsayCBS (improvement: Horizon3's root cause and server-side hunt artifacts, first covered through the horizon3-ai lookback), Citrix NetScaler CVE-2026-88771 (improvement: Zscaler's first-party telemetry adds a Perl CGI handler and a FreeBSD binary delivery; the Magento infrastructure overlap is Zscaler's own lead, not attribution).
KEV sweep (work/2026-10-18T0256Z-intel/kev-window.txt): no CISA KEV addition inside the window (catalogue 2026.10.16 of 2026-10-16 is the newest S1 read), so no NOT COVERED, MENTION-ONLY, COVERED-STALE or RANSOMWARE row. S1 confirmed that no covered CVE moved from not-exploited to exploited and that the ENISA exploited list is fully covered.
Backlog (state/coverage_backlog.md): the one open row, Beyond Gravity, was re-gated on today's facts by S2: nothing newer than the SRF report of 2026-10-15, the vector, behaviour and data class are still not public, so an incident entry would carry no evidence-bound technique. Held unchanged on its named condition (BACS, the company or reporting names a vector, an actor or the data affected), expiry 2026-10-20; nothing changed state, so no note was appended.
borderline-drop lines:
- borderline-drop: Keycloak 26.8.1 (S1; 19 CVEs) - advisories read are Low or Moderate, EPSS 0.0044 or below, no exploitation or KEV listing, so the regular patch cycle covers it; the SAML HTTP-Redirect signature parser differential has no CVE and rests on an issue title.
- borderline-drop: Zimbra 10.1.21 advisory table (S1) - the vendor table and heise are 2026-10-01 and 2026-09-29, outside the window, with no exploitation; the password-recovery, WebDAV and OnlyOffice rows carry no CVE or severity, and the covered Zimbra entry already directs readers to 10.1.21. Reopen on Rapid7's next part or a CVE with severity.
- borderline-drop: October Windows updates default the NTLMv1-derived single sign-on block to Enforce (S3; Microsoft KB5066470) - a vendor hardening default with an outage consequence, no attacker behaviour to map, outside the window; an operations notice, not a hunt, patch or block decision.
- borderline-drop: IDC Frontier sixth report of 2026-10-18 (S4) - service-status change only on a
routine, out-of-region entry; no new cause, route or actor. - borderline-drop: FBI PSA I-101326-PSA on North Korean IT-worker facilitators (S2) - US-written, no Swiss nexus, adds little to the three covered DPRK IT-worker entries.
- borderline-drop: FBI and CISA PSA on election and campaign officials, heise's CDU Heilbronn mailbox hijack, Europol's digital-forensics report, Truesec commentary, Arkadia RAIDEN (S2, S3) - US-midterm or German party focus, no vector or actor, or commentary without telemetry; incident floor.
- borderline-drop: German universities (Halle-Wittenberg, HU Berlin, LMU), the ChimeraZ burst of French forum claims, the Atexo/Docaposte supplier breach, Eurofiber France, Inter-Etab, ABC Bank, Eichstätt, Brenner Base Tunnel (Everest claim), the French hacktivist DDoS wave, ETS California, Cyber Daily exclusives, BAE Systems (S4) - claims without victim statements (PD-6), no vector or actor (incident floor), or out of nexus.
- borderline-drop: Elastic Apex Accelerator, Group-IB KGB RAT, Report URI skimmer write-ups, Noma SneakPeek, Any.Run lure statistics, Unit 42's monthly ranking, Volexity part 3, the MALFEX npm package, iVerify recap, The Register on building automation, CSA Lab notes (S3) - no constituency nexus, no new behaviour, or vendor statistics.
- borderline-drop: Fortinet FG-IR-26-176 to -185, the October Microsoft, Adobe and Palo Alto remainders, Juniper JSA112186, Chrome 155, phpMyFAQ 4.1.9, HPE ClearPass, Splunk SVD-2026-1006 (S1) - none exploited and none KEV-listed; unchanged calls on the facts earlier fires used. Splunk says its advisories publish 2026-10-21, so splunk-advisories should be rotated into S1 on 2026-10-21 or 2026-10-22.
Single-source and carve-outs: Rank Math is single-source-victim (the vendor discloses its own distribution incident and Wordfence relays it; Admiralty A2); the Horizon3 and Zscaler additions are first-party analyses of their own reverse engineering and telemetry and carry no verification change on their entries. Reduced-confidence inclusions: none. Identifier omission: Wordfence's page lists CVE-2026-111862 for the Rank Math incident, but the NVD API returned no record and the CVE Services API a 404 on 2026-10-18T03:27Z, so no cves[] record was written for it.
Contradictions: Huntress's count of 162 confirmed tenants (as of 2026-10-16 19:00 UTC) is the number cited; The Hacker News reports 173 for Huntress and the entry does not repeat it. Todyl's post is dated 2026-10-15 on its page yet references ConnectWise's advisory of 2026-10-16, so it was updated after first publication; the entry cites the page date and says Huntress and Todyl run living posts.
Source-recipe notes (for the next audit): ConnectWise's Trust Center page and Wordfence's blog were read through the metered reader (extract fell to jina); S2 spent about five reader fallbacks, four of them on guessed feed paths, which the source records now describe. The Dixell France guide is undated; the 2026-10-13 date given is the file's last modification on the vendor's server (HEAD request), disclosed in the Danfoss entry's sourcing note.
Coverage gaps: prodaft (no dated listing on any transport), apa-ots (no listing or feed route), hpe-networking-psirt (single advisory readable, no listing), hu-berlin.de notice (Anubis challenge, read only as quoted by borncity).
Essential-coverage: all 21 essential records were attempted and returned content; ncsc-ch-security-hub was an edit scan with no post edited since 2026-10-16T06:56Z.
Excluded as recent attempts: 90 standard or candidate records were held out of the rotation slices because the previous two fires attempted them (belt and braces over the rotation cursor); S4's slice took the twelve oldest-cursor breach and ransomware sources regardless, because that pool is small and entirely attempted by the previous two fires.
Self-identification: main agent Claude Sonnet 5.5 (claude-sonnet-5-5) from the harness-injected model line; all four research sub-agents reported Sonnet 5.5 from their own prompt line.
Verification: four iterations. Iteration 1 (NEEDS_FIXES, 8 truth and 6 editorial findings, 17 with the advisories) and iteration 2 (NEEDS_FIXES, 5 truth) were fixed in place; iterations 3 and 4 are two consecutive CLEAN verdicts from independent cold passes, the second a full-ledger confirmation (141 of 141 claims). After iteration 4 the main agent applied three trivial wording fixes from its advisories (dropped 'paid' from the Rank Math summary, 'the malicious host Rank Math names', a missing verb in the Danfoss update section), none changing a sourced claim, and declined the other five advisories with the reasons recorded on the iteration. The Trust Center page for ConnectWise returned 403 to every live transport in the later passes, so those claims were checked against the body the gate saved at 2026-10-18T03:36Z.
Source health: tools/source_health.py probed 249 records, 247 relevant, one stale (threatpost, a demoted record) and one unreadable; the one UNSOLVED row, znuny-advisories (a candidate that has never been fetched), is a host outage: the site answers HTTP 503 to WebFetch and resets the connection from the container, so the recipe is unchanged and a note says to re-probe next fire.
← Operations dashboard · run-record contract: docs/pipeline.md