The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs network requests to established and reputable scholarly metadata services, including Crossref (api.crossref.org), DataCite (api.datacite.org), PubMed (eutils.ncbi.nlm.nih.gov), Semantic Scholar (api.semanticscholar.org), OpenAlex (api.openalex.org), and OpenCitations (api.opencitations.net). These connections are used exclusively for verifying publication details and are considered safe under the well-known services recognition rule.
[DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration was detected. The skill only transmits scholarly identifiers (e.g., DOIs) to authoritative databases for verification purposes. The code includes specific precautions to prevent API keys or contact information from being written to logs or output artifacts.
[COMMAND_EXECUTION]: There are no instances of arbitrary or unsafe command execution. The shell commands mentioned in the documentation are provided as examples for the user to execute the provided Python scripts in a development or audit environment.
[PROMPT_INJECTION]: The instruction set does not contain any patterns intended to bypass agent safety filters or manipulate core AI behavior. The guidelines provided in the SKILL.md file are functional and integrity-focused rules for academic auditing.
[CREDENTIALS_UNSAFE]: No hardcoded credentials or API keys were found. The scripts are designed to fetch necessary credentials (such as mailto addresses or OpenAlex API keys) from environment variables, following standard security best practices for secret management.
[INDIRECT_PROMPT_INJECTION]: The skill processes user manuscripts and external API responses, which constitutes an attack surface. However, the skill implements defensive mechanisms, such as strict regex-based identifier normalization and path traversal checks in locators, to mitigate potential data poisoning risks. Furthermore, it explicitly labels semantic support verdicts as requiring human review.

light-citation