The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include several bash scripts that execute local commands such as git, sed, dirname, and basename. These are used to query repository state, fetch remote changes, and manage worktrees and branches. This is the primary and intended function of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to take a user-supplied slug and interpolate it into shell commands for creating branches and worktree directories. While this presents a theoretical injection surface if the agent does not sanitize the input, the risk is localized to the user's own environment.
Ingestion points: User-provided slug argument in the /work command, located in SKILL.md.
Boundary markers: None explicitly provided to separate the user input from the command context.
Capability inventory: Execution of shell commands through the agent's terminal tool.
Sanitization: None specified; the instructions rely on the agent to handle the input as a string literal for the branch/directory name.

work