At the time BioSpec was designed, agent skills did not exist, and MCP was relatively fragmented. We made a conscious decision to only operate within the simple tooling offered by GitHub copilot, like file search and file edit. We envisioned a few use cases that would benefit from external tooling though, and now that the space has matured more it might be worth revisting. From a maintainability perspective, providing recommended tools/skills is likely easier than re-implementing their functionality or offersing within BioSpec. The functions we noted at the time were:
- Research: Contextualising an intent/analysis/dataset bundle against prior literature research, and against biomedical databases to drive planning, hypothesis generation (and critiques), and identify research gaps. Dreaming but witihn BioSpec.
- Package discovery: Weighting up potential tool options for analysis using web search rather than an LLMs internal knowledge (a lot of biomedical tools have small userbases and low documentation)
- Integration with other data storage/backup/workbook platforms
At the time BioSpec was designed, agent skills did not exist, and MCP was relatively fragmented. We made a conscious decision to only operate within the simple tooling offered by GitHub copilot, like file search and file edit. We envisioned a few use cases that would benefit from external tooling though, and now that the space has matured more it might be worth revisting. From a maintainability perspective, providing recommended tools/skills is likely easier than re-implementing their functionality or offersing within BioSpec. The functions we noted at the time were: