The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/init_workspace.py uses subprocess.run to execute git commands (git config, git rev-parse) to automatically detect repository and branch names. These operations are limited to metadata retrieval and use static argument lists, which is a common and safe pattern for development-oriented agent skills.
[DATA_EXPOSURE]: The skill manages all intermediate task data, including literature review findings and agent review drafts, within a hidden local directory (.bensz-api). It incorporates a dedicated validation script (scripts/validate_report.py) that checks the final output report for any accidental leakage of these internal file paths or workspace locations before delivery to the user.
[REMOTE_CODE_EXECUTION]: While the skill interacts with external resources such as URLs and research papers to generate ideas, it does not download or execute arbitrary code from these sources. Execution is confined to internal dependency skills and local Python scripts provided within the skill package.
[PROMPT_INJECTION]: The skill inherently processes untrusted user-provided materials (documents, URLs, data). To mitigate indirect prompt injection, it uses structured prompts for secondary review agents (documented in references/agent-review-prompt.md) and a multi-round iterative review process to ensure the final research hypothesis remains focused on the scientific objective.

research-idea