agent-boundary-kit

Use before risky Claude Code edits or completion claims to catch agent boundary failures: wrong-scope edits, fallback before root cause, test-passing shortcuts, untrusted evidence leakage, oversized plans, or missing verification evidence.

Use before risky Codex edits or completion claims to catch agent boundary failures: wrong-scope edits, fallback before root cause, test-passing shortcuts, untrusted evidence leakage, oversized plans, or missing verification evidence.

Use when coding-agent work risks boundary failures: user intent or copy role confusion, fallback before root cause, named tool changes, test-passing shortcuts, scope creep, untrusted context, or completion without evidence.