The Agent Skills Directory

[SAFE]: No security issues were detected. The skill consists entirely of instructional text and metadata configuration for an AI agent to perform content auditing within a local environment.
[INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted user input (prompts being audited), the risk is minimal due to the skill's design.
Ingestion points: The skill ingests user-provided prompts via standard interaction for auditing purposes (SKILL.md).
Boundary markers: The skill uses structured output templates and markdown headers to separate the analysis from the potentially malicious input content.
Capability inventory: The skill only utilizes the Read tool. It lacks network access, file writing, or subprocess execution capabilities.
Sanitization: The skill enforces a strict human-in-the-loop requirement ('Bramka człowieka'), requiring users to manually review and approve all outputs line-by-line before use.
[DATA_EXFILTRATION]: The skill is explicitly designed to prevent data leakage. It contains no network tools, hardcoded credentials, or external destinations. The pii-egress: none and data-residency: local settings provide further security constraints.
[COMMAND_EXECUTION]: No shell commands, script generation, or dynamic code execution patterns were identified.

tajemnica-preflight-pl