Speaking of personal projects, I wandered off on a sidequest to create my own CLI tool for copyediting my text. The idea is to have something that can take text in, prompt an LLM to proofread, and ship the result back. Following UNIX norms, the baseline is to read and write to standard input and standard output. After that, it’s bells and whistles.
This one took a slightly odd path but I’ve got something out the door: copyedit_ai. Okay, nothing fancy and probably only of use to me. Here’s how I did it.
- Started off with Claude just to generate a Bash script with this prompt
Use the Python llm cli package from Simon Willison to create a bash script that reads from standard input or a file and then prompts a model to copyedit the text from stdin. The prompt should be something like …
- Gave the script a test drive
- Requested a few features, like passing extra arguments
- Felt decent about the script
- Created a new GitHub repo from a cookiecutter that aligns with a lot of my personal preferences
- Stuffed the script into the repository
- Prompted Claude Code to take this plan and generate an implementation plan
- Pushed the code up to GitHub
- Pointed Claude Code Web at the repo and told it “Make it so!”, more or less
- Iterated with Claude Code Web to reach the tool’s current state
I think this will turn out quite useful. Even if it doesn’t pan out, going through the process struck me as just one of many reasonable ways to approach agentic coding. I sort of like having a chatbot generate bash as a first draft. Plop it into a container of guidelines and guardrails. Then let an agent cook.
Speaking of personal utilities, I’m fiending for a simple way to use the 1Password CLI or SDK to grab api key secrets and install them in the current shell (ai_key_op?). In the age of LLMs, api key fatigue is real. Maybe a little TUI action as a bonus. Feels like yet another job for agentic coding.