The Agent Skills Directory

[SAFE]: The skill's primary function is prompt authoring and optimization. It provides detailed methodologies for different use cases (Modes A, B, C, D) without including malicious instructions or hidden payloads.
[EXTERNAL_DOWNLOADS]: The skill documentation includes an installation command using npx to add resources from the official soia-team repository. As this originates from the verified skill author, it is categorized as a legitimate vendor resource.
[DATA_EXPOSURE_AND_EXFILTRATION]: No evidence of hardcoded credentials, sensitive file access, or unauthorized network operations was found. The skill explicitly warns against bypassing access controls or obtaining non-public credentials in its safety reference files.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input as its primary data source. It mitigates injection risks through several layers:
Ingestion points: User-provided text or file paths (SKILL.md).
Boundary markers: Instructions to use Markdown code blocks or paired quotes to delimit data from instructions.
Capability inventory: File system and network operations are only permitted if the user explicitly triggers execution mode.
Sanitization: The 'Mode C' disambiguation process specifically identifies and clarifies the intent of sensitive terminology to prevent misinterpretation.
[COMMAND_EXECUTION]: While the skill can execute generated prompts, this behavior is 'output-only' by default. Execution requires explicit user commands like "execute after writing," and the skill is instructed to report all planned actions and risks before proceeding.

soia-dev-prompt-clarity