CLAUDE CODE SECURITY PREVIEW LANDS ALONGSIDE KEY CLI HARDENING
Anthropic shipped a limited Claude Code Security preview to scan repos and suggest patches, alongside CLI updates that improve remote build control, sandboxed h...
Anthropic shipped a limited Claude Code Security preview to scan repos and suggest patches, alongside CLI updates that improve remote build control, sandboxed hooks, and context efficiency.
Anthropic’s code-scanning capability is now built into Claude Code as a limited research preview for Enterprise and Team customers, with human-in-the-loop patch suggestions and expedited access for OSS maintainers, per coverage from CSO Online. In parallel, the CLI added a new remote-control mode for external builds, hardened HTTP hooks behind a sandbox proxy and explicit allowedEnvVars, persisted large tool outputs to disk to save context, and fixed a workspace-trust gap—plus a Windows crash fix in the VS Code extension (v2.1.51, v2.1.52).
Teams are also adjusting to a simplified CLI output that hides some file I/O; practitioners suggest prompting for a pre-action file list to restore transparency and control, effectively a dry-run step community thread. The wider ecosystem is keeping pace—LangChain’s Anthropic integration updated headers for 1M-context handling, model IDs, and tests, smoothing orchestration in agent workflows release notes.
Security scanning and patch suggestions now live where your agent already reads and edits code, shrinking time-to-detect and time-to-fix.
Hook sandboxing, env allowlists, and output persistence reduce supply-chain and reliability risks as agent usage scales.
-
terminal
Pilot Claude Code Security on a staging monorepo and compare findings/FPs to your existing SAST, tracking patch applicability and review load.
-
terminal
Enable sandboxed HTTP hooks with explicit allowedEnvVars and measure context savings from tool-result persistence on long agent sessions.
Legacy codebase integration strategies...
- 01.
Adopt the new remote-control mode to run agent builds on existing CI workers while keeping repo write access tightly scoped.
- 02.
Add a pre-exec 'list planned file paths' step in prompts or scripts to offset simplified CLI output and preserve auditability.
Fresh architecture paradigms...
- 01.
Design agent workflows around sandboxed hooks, explicit domain allowlists, and model pinning from day one to simplify compliance.
- 02.
Integrate code scanning as a pre-PR gate with human review, using patch suggestions to accelerate secure-by-default pipelines.