장익수
Agents
Most broken skills are not broken. They load, they validate, and they are never invoked - because the description does not match how anyone actually asks. Covers the one-character YAML mistake that makes frontmatter load empty with no error, how to write a description as a match target rather than a summary, and how to prove invocation in two steps instead of hoping.
Keep a scheduled job alive across restarts, prove it actually ran, and get told when it didn't. The assistant's built-in scheduler is session-only and expires in seven days, so anything you set to run daily dies quietly. This covers where the schedule really belongs, the environment differences that make a scheduled command run and do nothing, how to make a job prove it is still alive, and the guard a screen-touching job needs.
Make a rule the assistant actually follows by moving it out of the prompt and into a hook that runs every time. When you have said the same thing three times and the behaviour still drifts, this is the fix. Covers which hook events carry the value, the exit-code rule that wedges sessions, how to test a hook with temporary settings so nothing is left behind, a skeleton that cannot crash your turn, and the escape hatch for when one does.
Hand work to subagents and get back results you can actually act on. A subagent that fails will often produce a plausible answer anyway - URLs that 404, numbers nobody checked, all reading as confidently as a correct report. This covers what delegates well and what quietly does not, the five things a brief needs before the output is usable, why asking for the dead ends tells you how carefully the work was done, and what to re-verify by hand before acting.
Stop the assistant from refusing work it could actually have done. "I can't" is an expensive sentence - if it is wrong, people stop asking, including for the things you could have done. This is the four-step check that catches a wrong no before it reaches the person waiting: has this already been solved here, is the wall real or just the first path, is this genuinely theirs to do, and have you actually asked the tool instead of guessing at its limits.
Stop guessing whether Claude Code can do something. This skill lists the features that look like they work but quietly do not - session-only cron jobs, connectors that show 'connected' with zero usable tools, skills that load empty because of one quote mark in the YAML. Every claim was verified by running it, and each section ends with the exact command that settles the question on your own machine.
만든 영상을 올리기 전에 냉정하게 혹평해 드립니다. 영상을 실제로 보고, 몇 초 지점에서 시청자가 빠져나가는지, 첫 삼 초가 왜 약한지, 화면과 자막 중 무엇이 촌스러운지를 구간별로 짚어 드립니다. 좋다는 말 대신 고칠 것만 말합니다. 릴스, 쇼츠, 광고 영상 모두 됩니다.