The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from user briefs and existing codebases. It lacks explicit input sanitization or boundary markers for this data. However, the skill specifically restricts its own capabilities to read-only operations using tools like rg (ripgrep) and explicitly forbids modifying the filesystem, running builds, or installing packages. This significantly mitigates the risk of an indirect injection leading to unauthorized system changes during the planning phase.
[COMMAND_EXECUTION]: The skill uses rg and shell_command to explore the codebase. These are standard read-only operations for a coding assistant and are used here within a narrow, non-destructive scope as defined by the skill's instructions.

planning-work