Build an LLM prompt from structured Shennong evidence

Usage

sn_build_prompt(
  evidence,
  task = c("annotation", "de", "enrichment", "results", "figure_legend",
    "presentation_summary"),
  style = NULL,
  audience = c("scientist", "clinician", "general"),
  language = c("en", "zh"),
  background = NULL,
  output_format = c("llm", "human"),
  include_json_schema = FALSE
)

Arguments

evidence: A structured evidence list created by sn_prepare_*_evidence().
task: Interpretation task type.
style: Optional style instruction, for example "manuscript".
audience: Intended audience such as "scientist".
language: Output language.
background: Optional user-supplied study background or biological context to inject into the prompt.
output_format: One of "llm" for a model-ready prompt bundle or "human" for a human-readable markdown brief.
include_json_schema: Whether to request structured JSON output.

Value

A prompt bundle with system, user, and messages.

Examples

evidence <- list(
  task = "annotation",
  cluster_summary = data.frame(cluster = "0", top_markers = "CD3D, TRAC")
)
prompt <- sn_build_prompt(evidence = evidence, task = "annotation")
names(prompt)
#> [1] "output_format" "task"          "system"        "user"         
#> [5] "messages"      "evidence"