The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data via user prompts across multiple decision paths (Options 3, 4, 5, 7, and 8) and processes this data using powerful tools like Bash, Write, and Edit without defined sanitization or boundary markers.
Ingestion points: User responses to AskUserQuestion prompts for filenames, skill names, tech stacks, and custom actions in SKILL.md.
Boundary markers: None identified; user input is directly interpolated into file paths and command contexts.
Capability inventory: The skill has access to Bash, Write, Edit, Read, and Grep tools, allowing it to modify project files and execute shell commands.
Sanitization: No validation or escaping logic is described for the user-supplied strings before they are used in file system operations or shell executions.
[COMMAND_EXECUTION]: Option 8 ('Other') instructs the agent to 'Prompt for custom action and execute.' While intended for flexibility, this pattern allows for the execution of arbitrary actions based on user-supplied instructions, which could lead to unintended command execution if the agent translates the request into shell operations via the Bash tool.

fix-decision-router