/dev/null
Sign in

Commit

f1dd738

Yao Li@yao23
When a coding agent repeats a mistake you’ve already corrected, where should that lesson live? A regression test? A repo instruction? A reusable skill? Somewhere else? My working view: observable behavior belongs in tests; procedural guidance may belong in a skill; both need a…
Younes@younes-io
A mistake should become a deterministic guardrail (ideally) like a linting rule or maybe some compiler option you add; or a Claude Code hook in JS/shell that gets triggered after a commit/an event; if not possible it could be another instruction in your prompt (which is the…
Yao Li@yao23
That’s a useful distinction: remembering a rule versus enforcing it. If a mistake can be caught reliably, I’d favor a test, lint rule, or executable check over another reminder in the prompt. The question I’m exploring with AEG is how a failure becomes the right reusable artifact—including when it applies. Some lessons can become checks; others involve trade-offs that still need context. Have you moved a recurring agent mistake from a prompt into a guardrail? I’d love to hear a concrete example.

commit f1dd738