The Agent Skills Directory

[SAFE]: The skill's operations are limited to reading and editing local files within a project-specific directory (thoughts/shared/questions/). No sensitive system paths are accessed.
[SAFE]: No network operations (e.g., curl, wget, fetch) are present in the instructions. All data processing and file updates are performed locally.
[SAFE]: There are no patterns of remote code execution, package installation, or dynamic script execution. The skill relies on standard text editing tools.
[SAFE]: The instructions include safety guidelines for the agent to actively reframe biased or implementation-heavy questions, which helps maintain the integrity of the research process and prevents indirect prompt injection through leading user feedback.
[SAFE]: The frontmatter configuration uses disable-model-invocation: true and a specific model, which limits the operational scope and reduces the risk of autonomous model-to-model exploitation.

iterate-research-questions