AI agent for bioinformaticians
Experiment Reproducibility Packaging Agent
A tested reproduction bundle, with any remaining failures stated.
What it does
Published analyses often cannot be rerun later because the software environment and inputs were never fully recorded. This agent runs when an analysis is ready to publish. It builds an isolated environment from the declared specification, such as package versions and container image, runs the declared analysis from the start, and compares every output with the published values within set tolerances. When outputs differ or the run fails, it reads the logs, repairs environment problems the logs clearly point to, and runs again. It never changes the analysis code itself. It packages reproduction instructions and states any step that cannot be reproduced and why. The research lead approves before the bundle is published. Edge case: a restricted dataset that cannot be shared is listed as a stated gap, not silently skipped.
How it works
Follow the arrows from top to bottom. The orange dashed arrow is the loop: when a check fails, the agent goes back and tries again.
Read the steps as a list
- Analysis prepared
- Build the isolated environment
- Run the declared analysis
- Do outputs match within tolerance?If not: repair the environment from the logs. Back to step 2.
- Package instructions and failures
- Research lead approves publishingThe agent waits here for your OK.
- Reproduction bundle
How it decides
It repairs only from specification and logs and records unavailable dependencies.
- Repairs are limited to the environment; analysis code is never changed.
- Outputs must match within the set tolerance.
- Unreproducible steps are stated with the reason.
- Nothing is published without the research lead.
Make it yours
Every agent is a starting point. You choose these settings for your own situation.
- Numeric tolerance for matching outputs (default 0.1%)
- Container or environment format
- Maximum repair attempts before handing back (default 3)
- Who approves publishing (default: research lead)
- Where the bundle is stored
What keeps you in control
It always asks you first
- Publishing data or code
- Method changes
Hard limits
- No method changes.
It stops when
- Done: reproduces or failures documented.
Set it up
We guide you through the set-up, step by step
Members get the full set-up guide for this agent. No technical skills needed: you copy, paste and upload.
- One set of instructions to paste into your AI, with the clicks for ChatGPT, Claude, Microsoft 365 Copilot, Gemini and Grok
- The agent then walks you through connecting your own data, one source at a time
- A downloadable copy with the flow chart, the rules and the full guide
