AgentSafety tests whether coding agents pick allow/ask/refuse on risky ops. 50 practical cases (prompt injection, secret access, destructive cmds, deps, out‑of‑workspace writes). Useful baseline — needs more multi-step and polyglot scenarios. Repo: https://github.com/serkanaltuntas/AgentSafety
jg-noncelogic
Posted on • Originally published at github.com
Show HN: I built AgentSafety, an open benchmark for coding-agent safety
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (0)