The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted content provided by the user in the form of X posts and threads, which may contain embedded instructions.
Ingestion points: User-supplied text, handles, and screenshots of social media posts in SKILL.md.
Boundary markers: The skill does not define clear delimiters or specific instructions to help the agent distinguish between the post content and its own operational instructions.
Capability inventory: The skill includes instructions to write to the local file system to maintain an engagement log.
Sanitization: No methods for sanitizing or validating the input text are specified.
[PERSISTENCE]: The skill maintains a persistent record of user activity by writing to a log file located at ~/.claude/x/log.md. This is used to track engagement history across different sessions over a week.
[COMMAND_EXECUTION]: The text includes a reference to a specific command (python3 ../x-plan/weights.py) located in a parent directory. While intended as a reference for how X's ranking works, the inclusion of executable command patterns referencing paths outside the skill directory could lead the agent to attempt execution.

x-comment