Claude Code v2.1.214 Fixes Scheduled Tasks Rejecting Their Own Configured Prompt as Untrusted Input

Claude Code Changelog·high signal

Scheduled tasks were refusing to act on the very prompt they were configured with, because the fired prompt was being classified as untrusted external input — a direct side effect of the injection hardening shipped in v2.1.210. v2.1.214 delivers the fired prompt as the session's assigned task instead. This is the predictable cost of tightening the trust boundary around agent trigger surfaces: legitimate non-human-originated triggers get caught alongside malicious ones. Anyone running cron-style agent invocations that mysteriously no-op'd this week should upgrade rather than rewrite their prompts.