CLAUDE CODE 2.1.228 HARDENS SYNCED SKILLS AND STABILIZES RUNNERS
Anthropic shipped Claude Code 2.1.228 with concrete security hardening for synced skills and reliability fixes across the CLI and self‑hosted runners. The [v2....
Anthropic shipped Claude Code 2.1.228 with concrete security hardening for synced skills and reliability fixes across the CLI and self‑hosted runners.
The v2.1.228 release stops claude.ai–synced skills from shadowing local commands or MCP prompts, sanitizes their metadata, and blocks skill bodies from running shell bangs or expanding file macros. It also fixes runner cleanup quirks, Windows Git detection, cross‑session messaging, and Vertex AI credential handling.
A community changelog thread highlights the security angle and faster search defaults ripgrep. The system‑prompts mirror adds protocol and artifact guidance for enterprise workflows, including JWT validation for customer‑routed inference prompts 2.1.228. If you want desktop alerts for approvals and reviews, Warp outlines its notifications plugin for Claude Code Warp guide.
Synced skills could override local commands and trigger shell exec; this patch closes a real abuse path for terminal agents.
Runner and cleanup fixes reduce flaky sessions and data loss risks in long‑lived or CI‑driven workflows.
-
terminal
Try to reproduce command shadowing: create a local command and a same‑named synced skill; confirm locals win and skills can’t shell‑exec.
-
terminal
Exercise self‑hosted runners: background task completion, checkout failures on non‑push repos, and memory/plugin‑cache cleanup behavior.
Legacy codebase integration strategies...
- 01.
Upgrade to 2.1.228 and audit existing synced skills; remove any that relied on shell execution or unsafe macros.
- 02.
Tighten CI/CD: pin claude-code-action to the latest tag, and add regression checks for credential expiry (Vertex AI) and session recovery.
Fresh architecture paradigms...
- 01.
Adopt Claude Code with stricter defaults: synced skills for UX, local commands for authority; treat skills as untrusted input.
- 02.
Use Warp notifications to keep human‑in‑the‑loop approvals visible without wiring custom alerting.
Get daily CLAUDE-CODE + SDLC updates.
- Practical tactics you can ship tomorrow
- Tooling, workflows, and architecture notes
- One short email each weekday