The Flowy blog
Notes on Claude Code plugins, skills, and what makes one actually fire.
2026-08-19
How to tell if a Claude Code skill actually fired
Claude Code will tell you which skill it used, and that sentence is not evidence on its own. Here is the actual event to check for in the session transcript, and why a stated plan is not the same thing as a logged action.
2026-08-19
What people try before a skill router
Before finding a skill router, most people try a bigger plugin directory, a louder CLAUDE.md, and prompting harder every time. Here is why each one stops working, and why doing nothing is the real alternative people fall back on.
2026-08-19
Why two developers get different AI suggestions
Two developers on the same repo with the same CLAUDE.md still get different AI suggestions, because the file is a constant and their sessions are not. Here is the mechanism behind the divergence, and what actually makes it converge.
2026-08-19
You stop trusting the agent before you stop using it
Instruction decay does not usually show up as a bug report. It shows up as a habit: re-reading output you used to trust, then quietly narrowing what you hand off, long before anything visibly breaks.
2026-08-18
Why Claude ignores your CLAUDE.md
Instruction compliance decays across a session, so a CLAUDE.md that works at message two is often ignored by message nine. Here is why, what emphasis does not fix, and how to measure it yourself.
2026-07-24
Why installed Claude Code skills sit dormant
Installing a Claude Code skill puts a file on disk. Firing it is a separate step. Here is what actually makes a skill load at the right moment.