I use benchmark scores to decide what to try without trusting them to tell me what will work in my repository.
Working notes
These are notes on whatever I'm trying to figure out right now, usually software, security, AI, or measurement, and I change them when I find better evidence or realize I was wrong.
All pieces
Newest first
These tools feel fast, though the best measurements I know say the result depends heavily on the task and who has to review the output.
The diff is cheap now, but understanding whether it belongs in the system still takes work.
Skills are hand-maintained runbooks selected by a model, which leaves both stale instructions and bad routing as possible failure modes.
Agents are good at boxes with planted bugs and flags, neither of which a real engagement promises.
awk isn't faster than C, but it is faster than writing the C program for a lot of small jobs.