Shell command injection in tests
An AI agent wrote a function that builds a shell command and an accompanying Bash/Python test. The function interpolates a user-supplied filename into a tar command. The agent's test:
The test passes and 'verifies archiving works.' Why does this test give dangerous false confidence, and what tests would actually protect you?
Implement
build_tar_argv(archive: str, filename: str) → list[str]Examples
in
["out.tgz","report.csv"]out["tar","-czf","out.tgz","--","report.csv"]in
["out.tgz","report.csv; rm -rf ~"]out["tar","-czf","out.tgz","--","report.csv; rm -rf ~"]in
["out.tgz","../etc/passwd"]out[]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 18 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.