The Agent Skills Directory

[COMMAND_EXECUTION]: The skill allows the agent to execute arbitrary programs and shell commands within a worktree using the wt switch -x <program> syntax.- [EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading and installing external plugins (e.g., omp, claude, codex) via the wt config plugins install command.- [REMOTE_CODE_EXECUTION]: The CLI tool supports a hook system (e.g., pre-switch, post-merge) where scripts are automatically executed based on definitions in .config/wt.toml. This allows a repository to execute arbitrary code when the agent performs worktree operations.- [DATA_EXFILTRATION]: The wt list --full command is documented to perform network operations and use an external LLM to generate summaries of worktree state, which may involve sending local project metadata or code to a remote service.- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection through untrusted repository data.
Ingestion points: Repository-level configuration files (.config/wt.toml), branch names, and PR/MR metadata fetched from remote URLs.
Boundary markers: No specific delimiters or safety instructions are provided to distinguish between the skill's instructions and untrusted data in the repository.
Capability inventory: The skill can execute subprocesses via the -x flag, run automated hooks, and perform network requests for plugin management and LLM summaries.
Sanitization: No sanitization or verification of hook commands or repository configuration is mentioned before execution.

z-worktrunk