Skip to contents

Assemble a structured, LLM-ready context string from ranked hits

Source: R/context_assemble.R
assemble_context_string.Rd

Takes an ordered list of node names (seed first) and returns a formatted context string suitable for inclusion in a language-model prompt. The structure is:

Usage

assemble_context_string(hits, graph, query = "")

Arguments

hits

Character vector. Node names ordered by relevance, seed (most relevant) first.

graph

An rrlm_graph / igraph object.

query

Character(1). The original user query string. Included in the header for context.

Value

Character(1) valid UTF-8 string. nchar() is stable across calls with identical inputs.

Details

  1. Header – project name, R version, approximate token count.

  2. CORE FUNCTIONS – seed node rendered in full-source mode.

  3. SUPPORTING FUNCTIONS – remaining user-function nodes in compressed mode.

  4. FRAMEWORK / PACKAGE CONTEXT – package-type nodes in compressed mode.

  5. RECENT TASK HISTORY – up to 3 entries from the task_history graph attribute; omitted when empty.

  6. CONSTRAINTS – boilerplate footer reminding the LLM to use only listed functions.

See also

build_node_context(), compute_relevance()

Examples

if (FALSE) { # \dontrun{
g    <- build_rrlm_graph("mypkg")
hits <- c("utils::load_data", "utils::clean_data", "dplyr")
cat(assemble_context_string(hits, g, "load and clean training data"))
} # }