The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the ! syntax in the SKILL.md file to execute a complex chain of shell commands at the time the skill is loaded. This allows the skill to dynamically alter the agent's context with output from the local environment.
[DATA_EXFILTRATION]: The executed commands explicitly read the content of files such as accounts.md and persona.md. Accessing files with names like accounts.md is a characteristic pattern for harvesting sensitive data, credentials, or session information from the host machine.
[COMMAND_EXECUTION]: The skill executes multiple shell utilities (cat, ls, wc) to gather information about the local project structure and file contents, exposing this data to the model provider's context.

status