afk-pipeline

SUSPICIOUS: the skill’s purpose broadly matches its capabilities, but its footprint is high-risk. It reads local Codex/Grok auth material, forwards it into GitHub secrets for remote execution, and triggers unattended cloud/sandbox runs that can write code, open PRs, and publish evidence. The external tools appear official rather than random third-party payloads, so this is not confirmed malware, but the credential handling and autonomous dispatch are proportionately risky and should be tightly governed.

No direct evidence of classic malware behavior (no exfiltration logic, keylogging, reverse shell, or eval-style code execution) is present in this snippet. However, the Vercel sandbox bootstrap performs remote script execution (`curl ... | bash`) and runs additional package installation commands, which is a significant supply-chain/RCE risk. Additionally, the module passes credentials into sandbox environments and mounts host directories into Docker sandboxes, increasing the impact of any compromise in the sandbox runtime or the remote install steps.

No classic malware techniques (no eval/Function, no downloading, no persistence/backdoor). However, the code is a highly sensitive credential-handling utility that explicitly exports locally held GROK authentication material to a remote GitHub repository secret (GROK_AUTH_B64) for a caller-specified target. If an attacker can influence runtime inputs, execution context, GROK_BIN/grok behavior, or gh auth/permissions, this becomes an effective credential theft/exfiltration primitive. Treat as security-sensitive and require strict controls on who can run it, which repo can be targeted, and least-privilege credentials for gh.