The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external Git repositories, such as branch names, commit messages, and file paths, which could be crafted to influence the agent's behavior during cleanup operations.
Ingestion points: The skill reads branch metadata via git branch -vv (SKILL.md line 38), commit history via git log (SKILL.md line 54), and repository structures via find ~/Projects (SKILL.md line 88).
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore embedded commands within the repository data.
Capability inventory: The agent can execute branch deletions via git branch -d (SKILL.md line 44) and maintenance tasks like git gc (SKILL.md line 102).
Sanitization: No specific sanitization or escaping mechanisms for branch names or commit messages are described.
Mitigation: The skill enforces strong safety rules (SKILL.md lines 105-111), requiring explicit user confirmation and dry-run previews for all actions, which significantly reduces the risk of automated exploitation.
[COMMAND_EXECUTION]: The skill utilizes several powerful shell commands to perform its tasks.
It executes a file system scan using find ~/Projects -maxdepth 3 -name .git -type d to identify repositories.
It uses a complex pipe involving git rev-list, git cat-file, awk, sort, and head (SKILL.md line 73) to analyze repository history for large files.

git-cleanup