You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Risk dimensions: Exposure amplification: Medium (sandbox is the security boundary for agent execution) · Patchability: Medium (base-image bumps, cadence-dependent) · Detectability: High (weekly scans exist) · Operational fragility: Medium (accumulating unaddressed base-image CVEs across multiple open per-version issues) · Ownership confidence: Medium (burn-down issue [container-image-scan] Container CVE burn-down #52657 exists but individual per-version issues keep proliferating)
Finding
The last 7 days of container-image-scan automation opened/kept-open numerous per-version findings for gh-aw-firewall components (agent, squid, cli-proxy, api-proxy) plus related images (mcpg, github-mcp-server, serena-mcp-server, node:lts-alpine): #53075, #53073, #53072, #53071, #52858, #52652, #52456, #52455, #51710, #51707, #51329, #51328, #51022, #51021, #51020. An aggregate burn-down tracker already exists (#52657), but individual per-scan issues continue to open without visible linkage/closure against it, risking triage drift and aging CVE backlog in the sandbox's own network-egress-control images.
Confirm the gh-aw-firewall base-image bump cadence keeps pace with weekly scan frequency so CVEs don't accumulate across concurrently-open image versions.
Assign an explicit owner/rotation for triaging new container-image-scan issues as they land, rather than leaving them unassigned.
SLA urgency
High — this is the sandbox that isolates agentic workflow execution; unaddressed base-image CVE backlog reduces the security margin of the primary containment boundary.
Discussion report
See the full UK AI Open Code Risk & Resilience Governance discussion report for this run (created alongside this issue) for the complete asset graph, control verification, and full risk-scoring tables.
Tier and risk-scoring
Finding
The last 7 days of container-image-scan automation opened/kept-open numerous per-version findings for
gh-aw-firewallcomponents (agent, squid, cli-proxy, api-proxy) plus related images (mcpg, github-mcp-server, serena-mcp-server, node:lts-alpine): #53075, #53073, #53072, #53071, #52858, #52652, #52456, #52455, #51710, #51707, #51329, #51328, #51022, #51021, #51020. An aggregate burn-down tracker already exists (#52657), but individual per-scan issues continue to open without visible linkage/closure against it, risking triage drift and aging CVE backlog in the sandbox's own network-egress-control images.Remediation action
SLA urgency
High — this is the sandbox that isolates agentic workflow execution; unaddressed base-image CVE backlog reduces the security margin of the primary containment boundary.
Discussion report
See the full UK AI Open Code Risk & Resilience Governance discussion report for this run (created alongside this issue) for the complete asset graph, control verification, and full risk-scoring tables.