run-reproducible-r-analysis

Files (2)

• SKILL.md
• shennong.skill.yaml

SKILL.md (first lines)

# Run a reproducible R analysis

The unified Runtime image preinstalls commit-pinned `Shennong` and
`ShennongData` packages. Optional analytical backends are not implicit global
state and must remain in the Project environment lock.

1. Require an active Project before executing code. If the conversation is personal, explain that Runtime work needs a Project and help the user choose one.
2. Call `plan.propose` with the smallest useful analysis, artifact, and verification steps. Mark one step `in_progress`, then pass its exact `plan_id` and `plan_step_id` to `runtime.submit_job`.
3. For an R built-in dataset, do not claim that an upload is required. Use the dataset by name from the reviewed R script.
4. Write a reviewable R script under `project://current/scripts/` and write outputs under `project://current/results/`.
5. Submit a bounded `cpu-small` Runtime job with `Rscript --vanilla`, explicit argv, inputs, outputs, timeout, seed when relevant, and the R environment requirements.
6. Read the terminal job status and logs. Do not claim success until expected output artifacts exist and are non-empty.
7. Register durable plots, tables, and scripts with their producing job, environment, command, and content digests.
8. Run `analysis.validate` when the result contains biomedical or inferential claims. For purely demonstrative built-in data, clearly label the output as a software demonstration.

Never execute host paths or arbitrary mounts, and never bypass Runtime approval.

From zerostwo/shennong-os, Apache-2.0.

More from zerostwo/shennong-os