RepoJournal
Agent Security Agent Security
16 wires and counting

$ follow Agent Security

Keep up with Agent Security in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-10-06
stories 24

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

agent-scan upgrades mcp to 1.30.0, discovers MCP servers hidden in VS Code extensions

By RepoJournal · Filed · About Agent Security · Composed from the cited sources · methodology

Snyk's agent-scan fixes a known vulnerability in the MCP Python package and starts finding MCP servers that extensions register in code rather than in mcp.json, while Guardrails lays the groundwork for transparent provider proxying.

fix: upgrade mcp to 1.30.0 for Snyk vulnerability snyk/agent-scan

by Cursor Agent

agent-scan now pins mcp 1.30.0 to clear a flagged Snyk vulnerability in the MCP client dependency. If you installed agent-scan from source or vendor its lockfile, reinstall or bump mcp to match before your next scan.

feat: discover MCP servers in VS code extensions snyk/agent-scan

by hemang-snyk

Discovery now statically parses VS Code extension bundles for servers registered via registerMcpServerDefinitionProvider, catching setups like Pylance's HTTP provider that never appear in mcp.json. The new extension_mcp module means a --scan-all-users run reads extension files and parses JavaScript it previously ignored, which the PR notes expands attack surface; fork gating keeps false positiv...

feat(server): execute buffered guarded operations NVIDIA-NeMo/Guardrails

by Pouyanpi

Guardrails added the executor that runs request projection, input checking, dispatch, response projection, and output checking in a fixed order, and it always projects and validates guarded input even when both checker calls are disabled. This is Phase A plumbing for transparent provider proxying, not yet something you configure.

feat(server): declare private transparent proxy contracts NVIDIA-NeMo/Guardrails

by Pouyanpi

Alongside it, the private provider-neutral contracts landed: a GuardedMessage value and input/output check inputs that keep content checking separate from provider payloads and HTTP behavior. Both PRs are foundations for proxying provider traffic without exposing a provider API.

linux: a command starts once both relays listen anthropics/sandbox-runtime

by ronleizrowice-ant

The Linux sandbox previously started its command immediately after launching the two background socat relays on ports 3128 and 1080, so a command that connected right away could get its first connection refused. The start-up script now waits until the network namespace reports two TCP sockets in use before running anything.

Action items

Quick answers

What shipped in Agent Security on October 6, 2026?
Snyk's agent-scan fixes a known vulnerability in the MCP Python package and starts finding MCP servers that extensions register in code rather than in mcp.json, while Guardrails lays the groundwork for transparent provider proxying. In total, 13 commits and 11 pull requests landed.
Who contributed to Agent Security on October 6, 2026?
6 developers shipped this update, including AlgoVoi (Christopher Hopley), Pouyanpi, Hemang Sarkar, Cursor Agent, ronleizrowice-ant, and TurtIeSocks.
What were the notable Agent Security updates?
fix: upgrade mcp to 1.30.0 for Snyk vulnerability, feat: discover MCP servers in VS code extensions, and feat(server): execute buffered guarded operations.