The Agent Skills Directory

[SAFE]: The skill acts as a library of standards (SKILL.md, evidence.md, executability.md) to guide an AI agent in performing structured software testing tasks. It contains no executable entry points that trigger malicious actions. The content is purely instructional and aimed at establishing software quality engineering practices.
[SAFE]: The Python scripts scripts/scan_signals.py and scripts/validate_schema.py are designed for static analysis of test cases and target codebases. They use only standard Python libraries (except for an optional check for PyYAML) and do not perform any network operations or unauthorized file access beyond reading the files specified by the user for analysis.
[SAFE]: The 'Adjudication' (裁决) logic described in evidence.md and triage.md is a documented mechanism for resolving conflicts between different data sources (e.g., PRD vs. Code vs. User Input) by giving priority to explicit user feedback during the QA process. This is a functional feature for the specific QA framework intended to allow the user to override contradictory documentation, not a malicious prompt injection attack.
[SAFE]: The skill includes guidelines for CI/CD integration (pipeline-integration.md) which define exit codes and artifact handling for automated environments. While it mentions the --dangerously-skip-permissions flag for the Claude CLI, this is contextually provided as a technical requirement for non-interactive headless agent operation in CI pipelines, not as a bypass of security controls for malicious purposes.

core