Sparkey

Provides time-limited, self-revoking SSH access for AI agents using certificate TTL, user expiry, forced command restrictions, and scheduled automated cleanup.

This skill appears to do what it says: it creates short‑lived SSH sessions by generating keys/certificates, provisioning local helper accounts and cleanup jobs, then revoking them. However, it must be run as root and it creates a persistent CA private key on the operator host — if that key is compromised, attackers can mint certificates for any server trusting that CA. Before installing/using: (1) review the scripts line-by-line and test in a disposable VM; (2) run setup-ca.sh on a dedicated, hardened operator host (or HSM) and restrict access to /etc/ssh/agent_ca; (3) prefer dry-run mode and audit logs; (4) never run on a production host without approval — the scripts modify users, systemd/at jobs, and /usr/local; (5) require explicit human confirmation before any agent-initiated grant operation. If you are uncomfortable running root-level scripts supplied by an unknown source, do not install or run them.

Type: OpenClaw Skill Name: sparkey Version: 1.2.1 The 'sparkey' skill bundle provides a robust framework for granting temporary, restricted SSH access to AI agents using a defense-in-depth approach (certificate TTL, OS account expiry, and a restricted dispatch shell). Analysis of scripts like grant-access.sh and the restricted shell logic shows intentional security controls, such as input sanitization against shell metacharacters, path traversal checks using readlink -f, and automated cleanup via at or systemd-run. The code is well-documented, lacks any evidence of data exfiltration or unauthorized persistence, and aligns perfectly with its stated purpose of providing auditable, time-boxed access.