Securing CI/CD in an agentic world: Claude Code Github action case

Microsoft Threat Intelligence discovered a prompt‑injection pathway in the Claude Code GitHub Action that, under specific conditions, could be abused to exfiltrate GitHub Actions workflow secrets. The blog details the attack chain, responsible disclosure to Anthropic, the mitigations applied, and prescriptive guidance to secure AI‑powered CI/CD (least privilege tokens, pin action versions, sanitize inputs, isolate agentic runs, use OIDC and short‑lived credentials). This is a supply‑chain/agentic‑AI risk that can lead to credential and secrets exposure if not properly mitigated.