Agent instructions (Cursor, Copilot, and similar)

April 18, 2026 · View on GitHub

Use this file together with llms.txt and CONTRIBUTING.md.

Primary toolkit (agentic SDRF)

Use sdrf-skills as the default setup for agent-assisted annotation and validation (skills + assistant rules + workflows). Follow its setup docs, then apply changes in this repository under datasets/ and confirm with parse_sdrf validate-sdrf as below.

Scope

  • In scope: datasets/**/*.sdrf.tsv, sandbox/**/*.sdrf.tsv (WIP only), and doc/CI updates only when asked.
  • Canonical vs sandbox: validated accessions belong in datasets/. If parse_sdrf validate-sdrf fails or the file is incomplete, use sandbox/ (see sandbox/README.md) until it passes locally.
  • Out of scope: Changing the SDRF specification or templates here; that belongs in bigbio/proteomics-sample-metadata.

Required checks before suggesting a PR

  1. Layout: datasets/{ACCESSION}/{ACCESSION}.sdrf.tsv (or same folder, extra .sdrf.tsv only when justified by the project).

  2. Run validation:

    pip install --upgrade "git+https://github.com/bigbio/sdrf-pipelines.git@main"
parse_sdrf validate-sdrf --sdrf_file "datasets/<ACCESSION>/<file>.sdrf.tsv" --use_ols_cache_only

  3. PR description must include public evidence (PX / PMID / repository URL) for sample and file mappings.

Agentic annotation — do / do not

  • Do keep edits minimal, accession-scoped, and aligned with archive metadata.
  • Do prefer ontology terms and template columns consistent with the chosen SDRF template (ms-proteomics unless the dataset clearly needs another layer).
  • Do not fabricate raw file names, replicate IDs, or sample relationships.
  • Do not bulk-reformat unrelated accessions.

If uncertain, leave a comment in the PR for human reviewers instead of guessing.