The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to resolve and execute shell commands from worktrees/worktree-modes.yaml. These commands are used to stop services and clean up resources during a worktree teardown.
[REMOTE_CODE_EXECUTION]: By treating a file within the repository as the 'source of truth' for executable commands, the skill creates a vector for arbitrary code execution if the repository content is untrusted, such as in the context of analyzing external pull requests.
[PROMPT_INJECTION]: The skill processes untrusted data from repository files to determine its execution logic, creating a surface for indirect prompt injection.
Ingestion points: worktrees/worktree-modes.yaml and AGENTS.md within the repository workspace.
Boundary markers: The instructions lack requirements for the agent to validate or sandbox the commands retrieved from the configuration file.
Capability inventory: The agent can execute shell commands and perform filesystem operations like removing directories and worktrees.
Sanitization: No sanitization or validation of the commands resolved from the configuration file is mentioned in the skill instructions.

teardown-worktree