The Agent Skills Directory

[SAFE]: The skill implements a 'user-first' configuration model. It explicitly prohibits silent setup or assuming defaults without user confirmation in references/setup.md.
[COMMAND_EXECUTION]: Uses standard system utilities (git, gh, rg, find, mkdir) for legitimate repository management and search tasks. All command patterns are transparent and mapped to the skill's stated purpose of source code research.
[DATA_EXPOSURE]: The skill operates on a dedicated directory (~/clones by default). While it reads file metadata and contents, it focuses on public repository data and includes rules in SKILL.md to treat clones as read-only and preserve local modifications.
[EXTERNAL_DOWNLOADS]: Downloads (clones) are performed via official git and gh tools from user-specified or resolved public repositories (e.g., GitHub, GitLab). The workflows use standard protocols (HTTPS/SSH).
[PROMPT_INJECTION]: No malicious override patterns or safety bypass attempts were detected. The instructions focus on structured research workflows and evidence-based answering.

replicant