Package: admiralagent 0.1.0.9000

Jaime Yan

admiralagent: Spec-Driven ADaM Code Generation with LLM-Translated Layers

Translates P21-style ADaM specifications into executable 'admiral' code with validation comments. An LLM (optional, via 'ellmer') maps each spec variable's free-text derivation onto a closed vocabulary of derivation layers (structured JSON IR); a deterministic compiler renders the IR into 'admiral'/'metatools' code with 'CHECK' validation comments and writes provenance sidecars. Ships with a zero-dependency rule-based classifier as baseline and fallback; the LLM never writes R code directly and all model output is validated against the closed vocabulary before use.

Authors:Jaime Yan [aut, cre]

admiralagent_0.1.0.9000.tar.gz
admiralagent_0.1.0.9000.zip(r-4.7-any)admiralagent_0.1.0.9000.zip(r-4.6-any)admiralagent_0.1.0.9000.zip(r-4.5-any)
admiralagent_0.1.0.9000.tgz(r-4.6-any)admiralagent_0.1.0.9000.tgz(r-4.5-any)
admiralagent_0.1.0.9000.tar.gz(r-4.7-any)admiralagent_0.1.0.9000.tar.gz(r-4.6-any)
admiralagent_0.1.0.9000.tgz(r-4.6-emscripten)
manual.pdf |manual.html✨
DESCRIPTION |NEWS
card.svg |card.png
admiralagent/json (API)

# Install 'admiralagent' in R:

install.packages('admiralagent', repos = c('https://yanmingyu92.r-universe.dev', 'https://cloud.r-project.org'))

Bug tracker:https://github.com/yanmingyu92/admiralagent/issues

On CRAN:

3.48 score 2 stars 4 scripts 41 exports 3 dependencies

Last updated from:9e098d3960. Checks:9 OK. Indexed: yes.

Target	Result	Time
linux-devel	OK	202
source / vignettes	OK	219
linux-release	OK	178
macos-release	OK	165
macos-oldrel	OK	158
windows-devel	OK	181
windows-release	OK	156
windows-oldrel	OK	170
wasm-release	OK	103

Exports:aa_layers accept_model agreement_overstatement artifact_hash build_context build_system_prompt check_admiral_compat classify_variables classify_variables_dual classify_variables_llm codelist_from_data dataset_json_meta dual_disagreements evals_accuracy execute_ir ir_dependency_report layer_docs layer_names load_evals mcp_handle_request mcp_serve mcp_tool_specs mock_metacore mock_spec_adsl new_step new_variable_ir read_artifact read_dataset_json read_define read_spec read_spec_df render_program render_variable routing_plan run_agreement run_evals run_validation spec_variables validate_ir write_artifact write_program_artifact

Dependencies:digest glue jsonlite

admiralagent: Spec-to-ADaM with LLM-Translated Layers

The idea | Walkthrough on the mock ADSL spec | Accuracy story: rules vs the pilot1 oracle | MCP server: the same gates from any agent | Improvement-loop philosophy

Rendered fromadmiralagent.Rmdusingknitr::rmarkdown

Last update: 2026-09-30
Started: 2026-09-30

Help page	Topics
Layer registry	aa_layers
Model acceptance gate	accept_model
How much an args-blind label overstates agreement	agreement_overstatement
Artifact hash	artifact_hash
Build LLM context	build_context
Build system prompt	build_system_prompt
admiral compatibility check	check_admiral_compat
Classify spec variables	classify_variables
Dual-backend consensus classification	classify_variables_dual
Classify spec variables via an LLM chat backend	classify_variables_llm
Build a codelist data.frame from observed data values	codelist_from_data
Dataset-JSON header metadata	dataset_json_meta
Dual-backend disagreements	dual_disagreements
Accuracy of an eval run	evals_accuracy
Execute a Layer IR against source datasets	execute_ir
IR dependency report	ir_dependency_report
Layer documentation cards	layer_docs
Layer names	layer_names
Load the spec-to-IR classification eval corpus	load_evals
Handle one MCP / JSON-RPC 2.0 request	mcp_handle_request
Run the admiralagent MCP stdio server	mcp_serve
Tool specifications for the admiralagent MCP server	mcp_tool_specs
Build a minimal executable metacore object from a spec data frame	mock_metacore
Mock ADSL spec	mock_spec_adsl
New layer step	new_step
New variable IR	new_variable_ir
Print an agreement report	print.aa_agreement
Print a model acceptance report	print.aa_model_acceptance
Read artifact	read_artifact
Read a Dataset-JSON file	read_dataset_json
Read define.xml	read_define
Read spec workbook	read_spec
Normalize spec data frame	read_spec_df
Render program	render_program
Render a variable	render_variable
Dependency routing plan	routing_plan
Cross-tabulate classification backends at every fingerprint level	run_agreement
Run the spec-to-IR classification evals	run_evals
Run validation	run_validation
Spec rows for a dataset	spec_variables
IR validation gate	validate_ir
Write variable artifacts	write_artifact
Write program artifact	write_program_artifact

Package: admiralagent 0.1.0.9000

admiralagent: Spec-Driven ADaM Code Generation with LLM-Translated Layers

Citation

Development and contributors

Readme and manuals

Help Manual

Usage by other packages (reverse dependencies)