Prospector Studio Release for 2026.11.13 — New Plugins, StrikeKit Engagement Settings & Bug Fixes
New MCPs
- Cisco Identity Services Engine (ISE): 24 tools across endpoints, internal users, identity groups, network devices, and live session lookup by MAC address or username.
- Cisco Catalyst Center (DNA Center): 17 tools across device and client inventory, site health, assurance issues, network topology, and read-only CLI commands.
- Cisco APIC (ACI): 14 tools across tenants, application profiles, bridge domains, VRFs, EPGs, fabric nodes, faults, and events. Two general query tools reach any part of the fabric's object model.
- Cisco Secure Workload (Tetration): 13 tools across application scopes and workspaces, inventory search, flow search, sensors, and the malicious-IP threat feed.
- Cisco Meeting Server: 12 tools across system status and alarms, spaces, calls and participants, users, SIP trunks, and the dial plan.
- Cisco NSO: 8 tools across managed devices, device config, sync-from and check-sync, and alarms.
- Cisco CUBE: 7 read-only tools across voice dial-peers, interfaces, and platform state on IOS-XE voice gateways.
- Cisco Unified CM (CUCM): 4 read-only tools, including SQL queries against the configuration database and user and phone lookups.
- Tenable One (Exposure Management): 24 tools across attack path analysis, MITRE technique exports, exposure views, and asset, finding and software inventory. It is separate from the existing Tenable.io / Tenable.sc MCP.
- Elastic Security (8.11): 9 tools covering the detection rule lifecycle and alert search and status. Pinned to the 8.11 API.
- Corelight Investigator: 5 read-only tools to search alerts and detections and run log-extraction jobs. Works with all four Investigator regions.
- Splunk HEC: sends events into Splunk through the HTTP Event Collector.
All the Cisco MCPs are self-hosted, so you point them at your own appliance's address.
Updated MCPs
- Splunk: The alert tools can now fully manage native alerts. They cover scheduling, trigger conditions, suppression, retention, and paging through and deleting fired alerts. Basic authentication is supported, and searches now respect the result limit you set.
- Elasticsearch: Fixed two bugs. The MCP could only reach a fixed AWS or Elastic Cloud host; it now works with any host, including self-hosted clusters. Basic authentication credentials were never actually sent; they are now.
StrikeKit
New: a reworked Engagement Settings page. Engagement Settings now follows the same section order as the New Engagement wizard. It has editors for objectives, rules of engagement, testing windows, and scope. One floating Save / Commit bar shows what is saved and what is committed, and a review panel appears before you commit. If a section fails validation, a Fix link takes you straight to it.
Improved: a clearer plan and execution dashboard. The execution dashboard now has two panes, with a banner showing which plan version is running. Regenerating a plan asks you to confirm first. If you narrow the scope while a run is in progress, you can stop or restart the run from the same page.
Fixed: Execute Plan tells you when a run is already live. If another tab or operator had already started a run, clicking Execute Plan switched to the monitor without explanation. It could also swap the active plan version underneath the running one. The page now says a run is already in progress and leaves the plan alone.
Improved: reports you can stand behind. The technical report now builds its methodology from the engagement itself. It includes the testing window, the scope, the authorization and rules of engagement, and the techniques used. Each finding shows its evidence. Findings from scanners such as Nuclei, Nmap and OpenVAS keep the tool and target they came from. Agents can no longer mark a finding confirmed without evidence.
Fixed: fewer duplicate findings from phase agents. Phase agents could not see the findings already recorded, so they re-reported the same vulnerabilities in new words on every run. Agents now see every existing finding, grouped by target, and can search them. They re-confirm a known finding instead of creating a duplicate.
Fixed: every tool call is recorded in Activities. Activity was only recorded while someone had the Execution page open, so unattended runs recorded nothing. Two open tabs recorded every call twice. Activity is now recorded on the server for every run. The Activities page, the technical report and the compliance coverage report all read from it.
Fixed: scope exclusions can't be bypassed with other IP spellings.
An excluded IP address could be reached by writing it another way, for example as a single number or in hex. Exclusions now compare actual addresses, and ambiguous numeric hosts are blocked. Address ranges such as 10.0.0.1-9 are now enforced as ranges.
Fixed: deletes always ask for confirmation. Seven destructive controls relied on a browser prompt that could be skipped. These include deleting an engagement, target, credential, template or activity, and switching between allow-list and block-list scope. Each now opens a confirmation dialog, and the server acts only after you confirm.
Fixed: runs no longer stall on a consent prompt for built-in tools. In some runs, a built-in tool that should be auto-approved asked for consent instead. Runs with no one watching then waited forever. These tools are now recognized and auto-approved as intended.
Code Mode
New: connect to APIs from Code Mode snippets. Code Mode snippets can now call external APIs through approved egress grants. A grant is tied to the snippet it was approved for. Credentials are stored as references and added on the server side, so the snippet code never sees them, and they are redacted from output. This release also adds an Onboarding persona and alert-source tools. (Rolling out gradually, tenant by tenant.)
Fixed: more stable code execution under heavy use. A timed-out code run could hold on to capacity it never released. That leak is fixed.
Dashboards
Fixed: agent-built dashboards render at full width. Dashboards created or edited by an agent could show every panel squeezed to a sliver. Panels now render at the size the agent set.
Sign-In & Identity
Fixed: signing in from the Okta dashboard tile works end to end. Clicking the Strike48 tile in Okta could land you on the Keycloak sign-in form instead of signing you in. This happened on first login, in a private window, or after a session expired. The tile now signs you straight in. The IdP-initiated login URL shown in the identity provider panel now works and uses your tenant's own address.
Fixed: SAML sign-in with hardcoded-role mappers. On SAML identity providers, every sign-in failed with a server error if the provider had a hardcoded-role mapper. That includes any provider set up with "Apply standard mappers". Sign-in now works, and the role is assigned as configured.
ASOC
Fixed: Devo alerts forwarded through Elasticsearch reach their cases. Devo alerts read back through Elasticsearch or Quickwit could crash case correlation. The alerts were stored but never reached a case, while the run still reported success. They now correlate into cases normally.
Studio
Fixed: Studio starts when S3 TLS verification is turned off. If S3 storage was set up with TLS certificate verification turned off, Studio could fail to start and keep restarting. It now starts normally.
Fixed: the Pages editor lists your pages. In VFS mode, the Pages editor always showed "Pages 0", and opening a page's files returned a not-found error. The editor now lists every page and opens its files.
Security
- Updated the bundled SQLite library to fix RUSTSEC-2026-0459 (CVE-2026-119375).
- Fixed a StrikeKit scope bypass and unconfirmed destructive actions (see StrikeKit above).