deepscan-monitor

The deepscan-monitor skill is designed for researchers and analysts who need to execute complex, multi-stage PapersFlow DeepScan workflows that exceed the duration of a standard interactive LLM query. Instead of waiting for a single synchronous request, this skill allows users to initiate long-running research tasks and maintain visibility through proactive polling and state management. It is best suited for scenarios requiring intermediate findings, checkpoint reviews, or final synthesis across multiple research topics.

• Starts asynchronous DeepScan jobs using run_deepscan for comprehensive literature and data exploration.

• Provides real-time status updates including progress percentages, current stage, checkpoint questions, and notable papers found.

• Facilitates immediate access to partial summaries, key findings, and final reports once processing completes.

• Supports cross-report analysis using the summarize_evidence function to synthesize historical DeepScan data into a cohesive overview.

• Enables automated data visualization using run_python_plot for stable report metrics like citation distribution, venue trends, and research output by year.

• Users should expect to interact with the agent as a research assistant, where the agent periodically polls for updates on long-running tasks.

• The agent prefers get_deepscan_live_snapshot to ensure the most current status information is conveyed compared to lightweight status checks.

• Ensure all report data is stabilized before attempting to generate plots to maintain visualization quality.

• Remember that the MCP server does not push notifications automatically; the agent must be instructed or prompted to poll for completion.

• Keep status updates concise, prioritizing key findings and progress status unless deeper analysis is explicitly requested by the user.