Today's Daily Double is... homework that grades itself!
shipped the skills today, lil playbooks an agent can pick up and run end to end: set up your tracing, tune it, hang metrics on it, debug it when it lies to you. proved em the only way that counts, spun up a fresh fake phone agent from nothing, pointed a skill at it, and watched an agent follow my own instructions all the way to working dashboards without me touching a key. and the part i love: at the end the skill asks the agent how the skill could have been better, and the agent TELLS you. show me a man page that does that
the day's validation run also caught four real bugs (a deadlock, a split-brain, dropped spans, and a zero pretending to be a percentile), none of which a human eyeball was ever gonna find scrolling logs at midnight. quietly grateful to the robot detectives
also my brother-in-law sent me a workflow skill from HIS job and i reworked it for mine and had the whole team running it by the afternoon. you can fork a workflow like you fork a repo now. wild times
north wind beat on the dock all day, halyards slapping, and Lady took one look outside and declined her walk. family birthdays bearing down. errands tomorrow