AI Services Market › Use cases › Software & DevOps
Software & DevOpsDone with Claude Code Oct 20, 2026
Use Claude Code as co-developer with AGENTS.md and logging
The author uses Claude Code for planning, coding, testing, and debugging, with AGENTS.md and numerical logging to improve accuracy and context handling.
- What They Did
- The author spent about seven months building ThinkTerm, a Rust terminal multiplexer on top of WezTerm, with Claude Code acting as a co-developer. Claude Code handled planning, code writing, tests, and benchmarks, while the author reviewed the output; hard rules lived in AGENTS.md or Claude.md, and for UI work without a framework Claude saved frames as PNGs because it could not see the window.
- How It Went
- The result was an open-source app for macOS, Linux, and Windows that the author says lets them manage agent sessions across machines. The main shortcoming they call out is debugging accuracy: Claude guessed wrong repeatedly on one bug until they made it log actual numeric sizes.
- Worth Knowing
- The most concrete catch is to force Claude to log numbers and real measurements instead of inferring them; the author says that fixed a bug after five wrong guesses.
Try it yourself with Claude Code
Create an AGENTS.md file in [repo path] that describes the project's structure, how to run and test it, coding conventions, and rules for you to follow. Then help me with [feature or bug] by planning first, then coding, testing, and debugging, adding clear numbered log lines so you can verify behavior from the output. Stop when the tests pass and show me a summary of the changes.
Do this with a skill.
These AI Services Market playbooks do this exact job.