The Agent Skills Directory

[SAFE]: The skill provides instructions for the agent to perform deterministic extraction of legal citations using regular expressions and rule-based logic rather than LLM inference. This approach reduces the risk of hallucinations or manipulation by the input text.
[DATA_EXPOSURE_AND_EXFILTRATION]: No exfiltration vectors were identified. The skill's configuration limits tools to Read and specifies local data residency with no PII egress. There are no hardcoded credentials or sensitive file paths accessed.
[INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted user documents, the risk is minimal due to the following:
Ingestion points: Text from legal documents processed via the Read tool.
Boundary markers: None explicitly defined in the instructions, but the output format is strictly structured.
Capability inventory: Limited to reading and text processing; no write or execution capabilities.
Sanitization: The use of regex-based pattern matching acts as a filter, only extracting content that matches specific legal citation formats.
[EXTERNAL_DOWNLOADS]: The skill references the eyecite project by Free Law Project on GitHub for attribution and architectural inspiration. This is a neutral reference to a well-known open-source repository and does not involve runtime execution of remote code.

ekstraktor-cytatow-pl