The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to execute internal Python scripts (citation_tracker.py, framework_recommender.py, cross_search_aggregator.py) and a system command for ZIP validation of the output document. These executions are scoped to the skill's local directory and the user's home directory for state persistence.
[EXTERNAL_DOWNLOADS]: The skill requires the 'docx' Node.js package and an active Consensus MCP connection. The network operations are limited to academic paper searches via the Consensus service, which is the primary data source for the skill.
[DATA_EXFILTRATION]: While the skill communicates with an external API (Consensus), the communication is limited to user-initiated search queries. The session data stored locally in the user's home directory contains only research metadata (queries, citation counts, paper titles) and does not involve sensitive user credentials or system data.
[PROMPT_INJECTION]: The skill contains behavioral constraints and 'grill-me' checkpoints to ensure the agent follows the research framework and budget. These instructions are functional and do not attempt to bypass safety filters or override agent constraints.
[CREDENTIALS_UNSAFE]: No hardcoded secrets or credentials are present. Authentication for the external research service is expected to be handled through the environment's MCP configuration.

litreview