The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local maintenance and setup scripts (e.g., done-cleanup.sh and start-worktree.sh) via the bash shell. It also leverages osascript to automate the macOS Terminal, executing commands like cd and claude in new sessions to set up the workspace.
[PROMPT_INJECTION]: The skill is susceptible to command injection because raw user input from the $ARGUMENTS variable is interpolated into a single-quoted bash command string. Malicious input containing single quotes could break out of the intended string context to execute unauthorized shell commands on the host system.
[PROMPT_INJECTION]: The skill exhibits an indirect injection surface with the following mandatory evidence:
Ingestion points: User-provided task descriptions via the $ARGUMENTS variable and issue metadata (including branch names) retrieved from Linear MCP tools.
Boundary markers: While data is wrapped in single quotes within shell scripts, there is no escaping of internal quotes or use of specific delimiters to prevent data from being interpreted as instructions.
Capability inventory: The skill can execute bash scripts, manage git worktrees, modify files in the repository's .claude directory, and automate macOS applications via AppleScript.
Sanitization: No explicit sanitization, validation, or escaping logic is performed on the task description or tool-provided branch names before they are used in high-privilege shell operations.

start