The Agent Skills Directory

[SAFE]: The skill implements strong integrity checks using SHA256 hashing. It requires the agent to verify both the content hash and the target file hash before performing any write operations, preventing accidental overwrites or TOCTOU (Time-of-Check to Time-of-Use) vulnerabilities.
[SAFE]: The installer script includes a dedicated function, Assert-UnlinkedPath, which recursively checks for symbolic links or junctions in the target path. This effectively mitigates path traversal and redirection attacks targeting the file system.
[SAFE]: Input validation is applied to user-provided language strings using regular expressions (ValidatePattern('^[^ {}]{1,80}$')) to prevent injection of control characters or template placeholders.
[SAFE]: The skill uses a robust backup mechanism, moving existing policies to a unique, timestamped backup directory outside the IDE's rule loading path before applying updates.
[SAFE]: The disable-model-invocation frontmatter flag is set to true, restricting the agent to follow the defined instructions without making unauthorized LLM calls during the setup process.
[SAFE]: The policy template installed by the skill includes defensive instructions for the Cursor Agent, such as requiring authorization for destructive actions and treating retrieved content as evidence rather than instructions, which helps protect the user against future indirect prompt injection attacks.

setup-cursor