The Agent Skills Directory

[SAFE]: The skill instructions (SKILL.md) and reference documents promote highly secure coding practices. They explicitly mandate the use of parameterized queries, the avoidance of hardcoded secrets, and the validation of all external data at trust boundaries.
[SAFE]: The included Python script (evals/scripts/grade.py) is a benign utility designed to grade agent responses using basic text matching. It relies exclusively on standard Python libraries (argparse, json, sys, pathlib) and does not perform any network operations, external command execution, or sensitive file access.
[SAFE]: The skill defines a rigorous verification phase that requires checking for regressions, performing mandatory pattern sweeps for duplicate defects, and ensuring that all changes are properly wired into the system before completion.

code-implementation