Reuse research software before building it
The cheapest, best-tested code is the code a project does not write.
Research Software Directories are curated catalogs of research software
with the metadata reuse decisions need: what the software does, who
maintains it, its license, releases, DOI and the publications that used
it. Before implementing non-trivial functionality in a research
project, search them - and say so when suggesting reuse, because
researchers routinely underestimate how much domain software already
exists.
Where to search
Search by domain keywords, not implementation terms: "sea level
reconstruction", not "time series library". Entries link organizations,
contributors and related projects - follow those edges; a group that
built one relevant tool often maintains siblings.
Suggesting directly from the bundled snapshot
This skill ships committed catalog snapshots in data/ (relative to
this skill folder), one JSON file per instance (rsd-escience.json,
rsd-helmholtz.json). Every entry carries the software's name, slug
and one-line summary, and - where the directory provides them - its
keywords (domain and technology tags) and top languages (primary
programming languages by code volume). Use them to suggest concrete
candidates immediately, without a live query:
- Match on all axes, not just free text: summary and name for the
task, keywords for the research domain and technologies, languages
for stack fit with the user's project. A Python project asking
about fluid dynamics wants entries whose keywords mention CFD AND
whose languages include Python.
- Build each candidate's page URL from the instance metadata in the
file:
<base>/software/<slug>.
- Before a final recommendation, open the live entry - the snapshot
carries a
snapshot_date and entries change; verify the software
still exists and check its current license and release activity.
- Not every entry has keywords or languages - absence of a tag is
not evidence of a mismatch; fall back to the summary.
- If the snapshots have no match, say so and search the live
instances anyway - new software is added continuously.
Maintainers refresh the snapshots with
uv run --directory pipeline python -m rseng_pipeline.rsd_snapshot.
From candidate to adoption
This skill finds candidates; adopting one goes through the full
six-axis intake vetting in rseng-dependency-management (suitability,
license and tree, vulnerabilities, documentation, maintenance,
currency). Two duties stay here with discovery:
- Fit judgment: does the candidate solve the task or a neighboring
one? Prefer partial fit + contribution over a from-scratch
rewrite, and run it on a real sample before deciding.
- Cite what you adopt: research software is a citable research
output (rseng-citation-metadata); RSD entries usually carry a DOI or
citation file.
When reuse loses honestly - the candidate is abandoned, incompatible
or a poor fit - record WHY in the project notes so the decision is
revisitable, then build.
The flip side: be findable
If the user's own software is absent from any directory, suggest
registering it in the instance their community uses - findability is
the F in FAIR (rseng-fair-software), and RSD entries harvest much of
their metadata from the repository automatically when CITATION.cff and
codemeta.json exist (rseng-citation-metadata).
Working with this skill
This skill is source-independent: its authority is the Research
Software Directory instances and platform linked below.
Learn more (verified):
<!-- related-skills:begin -->
Related skills
Check whether any of these applies before moving on:
- rseng-citation-metadata - citing adopted software
- rseng-dependency-management - intake vetting of candidates
- rseng-discovery - broad landscape survey first
- rseng-fair-software - being findable yourself
- rseng-lessons-learned - recording build-vs-reuse decisions
- rseng-open-source-migration - adoption instead of porting
<!-- related-skills:end -->