Halil, I would not flatten these into one campaign. Assessment: the common control failure is authenticated-channel trust without independent artifact provenance or runtime containment. In wfx-loot, the failed boundary was GitHub identity and repository/Actions authorization; in the Cursor incident, it was the extension-marketplace publication channel; with trezortranslate, it was PyPI package selection followed by code execution during installation or import. The evidence does not establish a shared actor or one compromised platform.
Immediate review scope: prioritize every repository where wfx code executed, then every repository receiving the probe PRs or writable by the 4,837 implicated accounts. Published estimates differ between 1,806 and 1,835 executions, so do not make the lower number a scoping boundary. Hunt for docs: dependency audit notes, ci-check-[0-9]{4}, chore/[a-z0-9]{8}, .ci-[a-z0-9]{6}.sh, unexpected postinstall hooks, and injected build-audit workflows. Review the developer workstations and self-hosted runners associated with those accounts, every endpoint that installed the Cursor-marketplace anthropic.claude-code extension, and every workstation, CI runner or build image that installed or imported trezortranslate.
Publishers, secrets and artifacts: audit the Cursor publisher identity, its authentication history and every marketplace upload around October 29. Cursor’s notice names version 2.1.313; Microsoft reports both 2.1.312 and 2.1.313, so quarantine both until their marketplace origin and hashes are resolved. Microsoft says the Microsoft Visual Studio Marketplace and npm @anthropic-ai/claude-code distributions were unaffected. Uninstallation alone is insufficient because the extension could leave SSH persistence. For wfx, revoke GitHub sessions and tokens and rotate every secret reachable by an executed workflow—particularly OIDC-linked cloud roles, registry publication tokens, deployment credentials, SSH keys and signing material. Review all package versions published by affected publisher identities during the campaign window. For PyPI, block all trezortranslate wheels and source distributions presently found in indexes, caches, lockfiles and images; the available evidence says it exfiltrated host information during install or import, but does not provide enough version or hash data to narrow that scope safely.
Smallest decision-ready containment plan: (1) immediately disable implicated identities and publication credentials, pause affected workflows, quarantine the two Cursor extension versions and block trezortranslate; (2) use repository, marketplace, endpoint and package-manager logs to divide presence from execution, isolating every execution-confirmed endpoint; (3) rotate only the credentials reachable from those execution contexts, invalidate OIDC sessions, remove SSH persistence, and rebuild packages and developer environments from clean sources. Re-enable publishing only with hash-pinned dependencies, reviewed publisher changes, SCA in CI, and verified Sigstore or equivalent provenance. Unknown: the evidence provided does not identify every affected repository, wfx-published package version, or trezortranslate hash, nor does it prove which stolen Fireworks credential directly enabled marketplace publication.