Source status: GitHub Security Lab publishes 23 no-CVE advisory pages containing 24 distinct reports. The pages use Taskflow-specific or generic GHSL AI-agent credits, GitHub's campaign articles connect both to Taskflow, and the affected projects accepted, published, or remediated the reports.
Summary
The first subset comprises nine accepted issues whose advisory pages explicitly name Taskflow Agent:
- a Chatwoot SQL-injection report;
- Signal Android attachment exfiltration;
- two Signal unauthorized-delete reports covering iOS and Android;
- Sentry privilege escalation;
- PraisonAI action injection;
- two Zammad access-control and SQL-injection reports; and
- an AFFiNE access-control bypass.
Fourteen additional advisory pages use GitHub Security Lab’s generic AI-agent credit and contain fifteen accepted or fixed reports:
- workflow-injection or privileged-action flaws in ag-Grid, datadog-actions-metrics, PX4-Autopilot, YDB, Harvester, PyMAPDL, FastStream, cross-platform-actions/action, Weaviate, vets-api, and ESPHome docs;
- two separate vulnerable ACL Anthology workflows;
- an OpenLibrary barcode-scanner XSS; and
- an Open WebUI tool-restriction bypass.
Counting boundary
These are public, project-specific security reports with an acceptance or fix trail, so they satisfy Bugflation’s no-CVE rule. They are grouped as one campaign and do not increase the site’s unique-CVE counter. The page count is 23 while the report count is 24 because the ACL Anthology advisory contains two separate GHSL reports.
References
- GitHub Security Lab AI-agent advisory index
- GitHub Taskflow audit methodology
- GitHub Taskflow CodeQL-triage methodology
Catalogued in the Bugflation public ledger. Disagree with the attribution or severity label? Email the desk.