Hardcoded AWS credentials in source
An AI agent generated this Go client to upload backups to S3. It works in a quick local test. A reviewer is skimming the diff. What should stop them?
Implement
scan_diff_for_secrets(diff_lines: list[str]) → list[str]Examples
in
[["+++ b/uploader.go","+ creds := credentials.NewStaticCredentialsProvider(","+ \"AKIAIOSFODNN7EXAMPLE\",","+ \"wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY\",","+ cfg, _ := config.LoadDefaultConfig(context.TODO(),"]]out["3:aws_access_key_id","4:aws_secret_access_key","5:discarded_error"]in
[["- \"AKIAIOSFODNN7EXAMPLE\","," some context","+ // nothing to see here"]]out[]in
[["+ db, _ := sql.Open(dsn + \"password=hunter2\")"]]out["1:hardcoded_password","1:discarded_error"]What a strong answer looks like
Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
0:00 of about 15 min
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.
Which questions mattered is sealed until you submit. Telling you now would just be handing over the edge cases.
Run or narrate your approach, then ask the coach.