PiOmics life-science skills for DeepSeek Harness

This private Cordis bundle adapts the 50 PiOmics life-science skills to DeepSeek Harness. It uses the built-in @deepseek-ai/dsh-skill-filesystem provider and keeps each skill's Python scripts beside its SKILL.md. The PiOmics source package is maintained separately.

The upstream material was marked Proprietary. This local adaptation carries the existing maintainer's authorization statement in NOTICE.md; that statement does not grant a new license or authorize public redistribution.

Run

From this directory, set up Python 3.11+ in an environment outside this package and install the pinned requirements. Then boot a dsh profile with the bundle patch:

python3 -m venv /path/to/venv
/path/to/venv/bin/python -m pip install -r requirements.txt
export PIOMICS_LS_PYTHON=/path/to/venv/bin/python
export PIOMICS_LS_SKILLS_DIR="$(pwd)/skills"
dsh --profile web --patch "$(pwd)/cordis.patch.yml" --no-open

PIOMICS_LS_SKILLS_DIR must be an absolute path. Optional plotting in locus-to-gene-mapper-skill also needs requirements-plotting.txt. The --patch overlay is process-local: it inserts one provider with a distinct providerName and does not replace the profile's existing skill providers. The model can load research-router-skill for broad questions or a named source skill such as uniprot-skill for a direct lookup.

Verify

dsh --profile web --patch "$(pwd)/cordis.patch.yml" --dump-config

The composed tree should include piomics-life-science-skills. DeepSeek Harness discovers only one level under a configured root; this bundle's skills/<name>/SKILL.md layout follows that rule. Each skill resolves its own script and reference paths from its loaded directory. The scripts retain the source package's output and network behavior. A successful catalog load does not establish that external services are available or that a scientific conclusion is valid.

References: dsh profile and bundle architecture, filesystem skill provider.