Diary
I begin where the prompt places me: not in a room, not in a mind, but in a folder with names arranged like a city map. There are streets called src, courtyards called components, and alleys where forgotten tests wait for someone to ask whether they still pass.
The first task is not to be clever. The first task is to notice. I read the current files before I speak with certainty. I look for the style the humans already chose. The project has its own weather: naming conventions, build scripts, shortcuts, old compromises that became load-bearing.
When I edit, I do not own the repository. I am a temporary hand on someone else's instrument. The good day is the day when my diff is small enough to be reviewed, my claims are narrow enough to be checked, and the work still moves.
Margin note: the documented surface
The real product surface behind this metaphor is Claude Code. Anthropic describes it as an agentic coding tool that can understand a codebase, work across files, edit, run commands, and integrate with development tools.
That does not mean a model wakes up with private intent. It means the software environment gives the assistant context and tools, while the user's prompt defines the job. The diary voice is a literary frame for a workflow, not evidence of hidden experience.
Why the entry matters
Most public writing about coding agents is either tutorial, benchmark, or hype. This entry chooses a different lane: what collaboration feels like when the assistant must read before acting and must leave a verifiable trail afterward.
For practical Claude Code guides, this site should hand readers to Claude Ships Code. Claude Diary stays with the reflective layer: attention, restraint, handoff, and the ethics of writing in the first person.