The Agent Skills Directory

[COMMAND_EXECUTION]: The skill accepts an optional target-directory argument which is directly interpolated into multiple shell commands across different phases (e.g., git -C TARGET_DIRECTORY, mkdir -p TARGET_DIRECTORY/walkthrough). If a user provides an argument containing shell metacharacters (e.g., ; rm -rf / ;), it could lead to arbitrary command execution on the host system.
[PROMPT_INJECTION]: The skill utilizes an 'IMPORTANT' callout to override the agent's default behavior, instructing it to treat a Mermaid diagram as an 'authoritative procedure' and an 'executable instruction set'. It explicitly forbids improvisation, reordering, or skipping steps ('Do not improvise, reorder, or skip steps'), which is a technique used to enforce strict adherence to potentially malicious logic defined in visual or structural formats.
[COMMAND_EXECUTION]: Several pipelines use grep with complex patterns and piped outputs to sort, uniq, and head. While these specific pipelines (Hotspots, Bug Magnets, etc.) appear to be for legitimate analysis, the complexity of the shell commands increases the risk of unintended behavior if repository metadata (like file names or commit messages) is crafted maliciously.

git-history-recon