The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around the use of the wt (or git-wt) command-line interface to manage repository state. It also facilitates the execution of other binaries or agents (e.g., claude, opencode) using the wt switch --execute flag, which allows passing arbitrary instructions to those processes.
[COMMAND_EXECUTION]: The skill utilizes lifecycle hooks (e.g., pre-start, post-start, pre-commit) defined in .config/wt.toml. These hooks are designed to execute arbitrary shell commands such as environment setup (cp), dependency management, and running development servers. The skill explicitly instructs the agent to validate and run these hooks.
[EXTERNAL_DOWNLOADS]: The configuration examples suggest the use of package managers (e.g., pnpm install) to fetch external dependencies during the post-start lifecycle phase.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and acting upon instructions defined in external configuration files (.config/wt.toml).
Ingestion points: The skill reads project-specific configuration from .config/wt.toml and user-level configuration from ~/.config/worktrunk/config.toml to determine behavior and hooks (SKILL.md).
Boundary markers: The instructions recommend using wt hook show --expanded and validating commands against actual scripts before execution, though it lacks strict enforcement of boundaries for untrusted repo content.
Capability inventory: The skill allows for shell command execution via hooks, launching external agents via the --execute flag, and performing file system operations like copying files.
Sanitization: The skill relies on manual validation steps, such as checking wt config show and wt hook show, to ensure configuration integrity.

worktrunk-workflows