The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the semantius command-line interface to interact with the platform's CRUD API (e.g., read_module, read_entity, read_field). These operations are scoped to retrieving schema metadata for the purpose of documentation and are consistent with the skill's stated purpose.
[DATA_EXFILTRATION]: While the skill reads sensitive schema information, including RBAC configurations (permissions and hierarchies), this data is written to a local markdown file within the workspace for documentation purposes. There are no network operations targeting external or non-whitelisted domains that would suggest exfiltration.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute external scripts. It relies on internal skill inter-dependencies (e.g., use-semantius, semantic-model-analyst) and standard CLI tools provided within the environment.
[PROMPT_INJECTION]: No attempts to override system prompts or bypass safety filters were found. The instructional language is focused on maintaining consistency in documentation (e.g., US English spellings, specific singular-subject grammar rules).
[INDIRECT_PROMPT_INJECTION]: The skill processes data from a live Semantius instance, which constitutes an external ingestion point. The instructions emphasize rendering this data "verbatim" or "byte-for-byte" (particularly for JSON logic and validation rules). While this represents a surface for indirect prompt injection if the source data were maliciously crafted, the skill functions as a faithful conduit for documentation, and the subsequent audit phase (Stage 5) includes checks for malformed logic.

semantic-model-optimizer