galaxyproject / galaxyproject/loom
galaxy_run_tool can't map a dataset collection over a single-dataset tool input (featureCounts, FastQC) -- "dataset collection supplied to single input dataset parameter"
- Dominant language
- TypeScript
- Stars
- 14
- Forks
- 12
- Avg merge
- 6d 5h
- Merged PRs (30d)
- 17
Description
When a tool's input is a single-dataset parameter (e.g. featureCounts `alignment`, FastQC `input_file`) and the agent passes a dataset collection -- the normal output shape from upstream steps like STAR or a paired trimming step -- Galaxy rejects the run:
```
400: Parameter 'alignment': dataset collection supplied to single input dataset parameter;
to run the tool over multiple inputs use map/reduce
```
This is the standard RNA-seq pattern (run featureCounts / FastQC per-sample across a collection of BAMs/FASTQs), so it comes up constantly. The agent currently has no way to express "map this tool over the collection" through `galaxy_run_tool`; it just retries the same collection-as-single-input call and re-fails.
Seen across two separate beta sessions (featureCounts with a STAR output collection; FastQC with a paired FASTQ collection), app 0.3.1, on both claude-sonnet-4-0 and deepseek-v4-flash.
Likely lives in galaxy-mcp's `run_tool` (needs Galaxy batch / map-over semantics for collection inputs), with a possible loom-side assist (guide the model to map-over, or expose batch in the input template). Related tool-input-template gaps: #198.
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.