Composes a short, citable account of a run: a design header (kind, run id,
the participating agents, and a compact workflow summary), the model and
token paragraph drafted by LLMR::report() over the run's
call-level rows, and a one-line limits note stating that the results are
model-conditioned and are not estimates of a human population.
Arguments
- x
An
agent_runor an Agent. Convert other results withas_agent_run()first.- ...
May include
task, a one-clause description of what the model was asked to do (spliced into the methods paragraph).
Value
An object of class agent_report: a character vector with a print
method that cat()s the lines.
Details
When some calls did not report their sent, received, or total token counts, the report says so beside the token sums. When the call records cannot be read, or the methods paragraph or the claim-language check cannot be produced, the report warns and says in its text what is missing.
Examples
a <- agent("Aria", LLMR::llm_config("groq", "openai/gpt-oss-20b"), quiet = TRUE)
report(a, task = "to answer a factual question")
if (FALSE) { # \dontrun{
a$chat("Hello")
report(a, task = "to answer a factual question")
} # }