The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface (Category 8).
Ingestion points: The skill ingests user-supplied citation lists and retrieves data from external sources via WebSearch and Consensus MCP, specifically querying api.crossref.org, doi.org, and academic search engines.
Boundary markers: While the skill follows a strict internal protocol (Steps A-E), there are no explicit delimiters or instructions provided to the agent to disregard potential instructions embedded within the retrieved academic metadata or search snippets.
Capability inventory: The skill uses Bash, Read, Write, and WebSearch tools. A malicious search result could potentially influence these tools if the model treats the external data as instructions.
Sanitization: There is no evidence of sanitization or filtering for the external content before it is processed by the agent to generate reports.
[SAFE]: Data Transmission and Search Operations.
The skill communicates with well-known academic services (doi.org, api.crossref.org). These operations are consistent with the skill's stated purpose of verifying academic citations and do not represent unauthorized data exfiltration.
[SAFE]: Scripting and Utilities.
The Python utility scripts/tw_edu_doc_utils.py is used for Word document styling and does not perform any network operations, subprocess executions, or sensitive file system access.

tw-edu-citation-checker