LSM-Copilot: A Skill-Flow Agent for Fluorescence Microscopy Analysis

Ruofan Liu, Pengcheng Chen, Eric J Seibel

Keywords: bioimage analysis, microscopy, agent skills, scientific agents

Abstract:

Fluorescence and confocal laser scanning microscopy produce large, heterogeneous image collections whose analysis often requires manual coordination across vendor readers, segmentation models, statistical scripts, and reporting tools. We present LSM-Copilot, an
agent suite that turns raw microscopy files and a natural-language analysis goal into auditable masks, tables, figures, and reports for two evaluated microscopy workflows: spot localization and 3D LSM object quantification. The system is organized as a Skill-Flow: a thin host agent routes work through three self-contained skills. Search, Process, and Interpret, while the skills package prompts, tools, domain knowledge, and verification logic. On public fluorescence spot-detection benchmarks, the same LSM-Copilot skills run in Claude Code, Codex, Cursor, and OpenClaw with Claude 4.7 Opus, GPT-5.5 xhigh, Composer 2, and DeepSeek V4 Pro. All four hosts independently recover evidence-backed method plans and obtain matching final F1 scores on Spotiflow, deepBlink, and 3D simulated spot-localization tasks. The prompt gives no benchmark source. hints, and the skills contain no hard-coded method names. On private 3D raw LSM crystallization data, the same skills work without web search: each host writes and runs a local 3D analysis pipeline and agrees with Imaris reference measurements at about the five-percent level. These results support a narrow claim: skill-level packaging can make both search-guided method discovery and fully local microscopy analysis portable across host agents on the evaluated tasks.

Presentation Mode: Yes, at least one author will attend and present in person.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 2