The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from the local Git repository, including branch names and commit subjects, which are presented directly to the agent's context. An attacker with control over the repository's history or branch names could theoretically embed instructions to influence the agent's behavior.\n
Ingestion points: The script scripts/1k-worktree-clean.sh (specifically in resolve_last_commit_metadata and resolve_pr_metadata) extracts commit subjects via git log and branch names via git worktree list.\n
Boundary markers: There are no explicit delimiters or warnings to the agent to ignore instructions embedded within the Git metadata.\n
Capability inventory: The skill allows the agent to execute shell commands, remove Git worktrees, and delete directories using rm -rf as described in SKILL.md Step 5.\n
Sanitization: No sanitization or filtering is performed on the commit messages or branch names before they are included in the output.\n- [COMMAND_EXECUTION]: The skill provides instructions for the agent to perform directory deletion and Git worktree removal. While these actions are the intended purpose of the skill, they represent a sensitive capability.\n
Evidence: SKILL.md Step 5 describes the removal of stale directories using rm -rf and worktrees using git worktree remove.\n
Note: The skill instructions explicitly require user confirmation before any destructive actions are taken.\n- [EXTERNAL_DOWNLOADS]: The skill fetches data from remote repositories to compare local state.\n
Evidence: SKILL.md Step 2 includes rtk git fetch origin x and the script uses gh pr list to retrieve pull request status from GitHub.\n
Note: These are standard operations for a Git-related utility and target well-known services.

1k-worktree-clean