OpenAnolis, the community distribution project, launched SkillHub on 25 June as a place to package infrastructure know-how, including system operations, security hardening and performance tuning, as reusable skills that an AI agent can call. It ran a skill-creation campaign with AMD and now lists more than 100 skills across security, operations, AI inference and databases. The latest pick walks through pg-create-cluster, a skill that lets Claude Code build a PostgreSQL streaming replication cluster with one primary and several standbys. Rather than generating one long command, it asks one question per turn, runs host, VIP and binary-path checks to offer real machines and versions as choices, and ends by producing and executing the script. The author lists three prerequisites that break most attempts: CLup reachable locally on port 8090 or via --url, a fixed Python interpreter at /opt/csu_pyenv/bin/python carrying pycryptodome, and PostgreSQL binaries built for the host operating system, since an EL8 build on Rocky 9 fails initdb on missing libicui18n.so.60 and libssl.so.1.1. He also claims fewer than 10% of published skills get the conversational flow right.
Our takeThe useful part is not the dialogue but the constraint set wrapped around it. Ask local or remote before credentials, pin the interpreter, match binaries to the operating system: those are the same failure modes that break hand-written runbooks, now surfaced one question at a time. That also means the skill encodes one engineer's environment knowledge, and the post reports no success rate against a hand-assembled command, nor any evidence it transfers to a second cluster. More than 100 skills counts submissions. The thing to watch is whether the catalogue adds a test harness per skill; without