Useful prompts to try
Start with a read. Be explicit before changing anything.
Read and explain
“List my DimerIQ runs. Open RUN_ID and explain the target references, exclusion references, checks performed, missing checks and current state. Include the run ID.”
“Read the candidates for RUN_ID. Compare their recorded findings. Do not choose one for me or change any thresholds.”
Start work
“Prepare a selected-reference panel using these approved accessions and a review goal of three. Show me the exact inputs and scope before starting.”
Pause or resume
“Pause RUN_ID. Check until it reaches a saved stopped checkpoint, then tell me which results remain available.”
“Resume RUN_ID from its saved checkpoint. Give me the new execution ID and its link to the old run.”
Save a decision
“Set candidate CANDIDATE_ID aside in RUN_ID with this note: … Confirm the saved decision ID. Do not change its calculated findings.”
A result should distinguish what the tool returned from the assistant's interpretation. Missing evidence must stay missing; it must not become a pass in a summary.