The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts input from the user via the $ARGUMENTS placeholder. This provides a vector for indirect prompt injection where a user might attempt to supply instructions instead of the expected data to override agent behavior. The skill attempts to mitigate this by using <arguments> tags and a directive to treat the content as data rather than instructions.
Ingestion points: The $ARGUMENTS placeholder in SKILL.md captures all text following the command.
Boundary markers: The skill uses <arguments> XML-style tags and explicit text instructions to separate user data from the system prompt.
Capability inventory: The skill is granted access to the Bash tool, allowing it to execute system commands like git revert.
Sanitization: There is no explicit sanitization of the input before it is used, relying instead on the language model's ability to follow the provided boundary instructions.
[COMMAND_EXECUTION]: The skill is configured to use the Bash tool to perform its primary function of reverting git commits. This is a powerful capability that is intended for the skill's operation but requires the agent to correctly interpret arguments to avoid unintended command execution.

gsd-undo