Stats
Actions
Tags
From denario
Run the Denario research pipeline end-to-end over MCP — lay out a project from a data description, then idea → methods → results → paper, configured via params.yaml. Use when the user wants a full research run (not just one stage).
How this skill is triggered — by the user, by Claude, or both
Slash command
/denario:run-pipeline <project_dir> [data description or path] [hints: --steps=4 --vlm --no-citations …]<project_dir> [data description or path] [hints: --steps=4 --vlm --no-citations …]This skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Drive the full Denario pipeline for the project at `$0` through the **denario MCP
Drive the full Denario pipeline for the project at $0 through the denario MCP
tools. Read the explain skill first if you need the tool
details. Confirm any underspecified choice with the user before launching a long
run — don't guess the science.
Prepare the project (avoid denario_setup unless the user wants a GitHub repo):
mkdir -p $0/Iteration0/input_files$0/Iteration0/input_files/data_description.md (from $ARGUMENTS, a referenced file, or by asking).params.yaml at $0/params.yaml. Start from Denario/tests/params_multiprovider.yaml (or denario-scientists/data/params.yaml) and set, in the Analysis module:
max_n_steps: 3 or 4 (small unless the user wants more),gemini-3.5-flash, gpt-5.4, claude-sonnet-4-6) — and ensure each is in denario/llm.py's registry,enable_vlm_review: true (+ vlm_model: gemini-3.1-flash-lite) if --vlm.GOOGLE_API_KEY, and OPENAI_API_KEY/ANTHROPIC_API_KEY for those providers).Idea → denario_idea(project_dir=$0). Then read input_files/idea.md and show a 1-line summary.
Methods → denario_methods(project_dir=$0). Summarize methods.md.
Results → denario_results(project_dir=$0). This is long (minutes) and runs cmbagent_lg.
/tmp/denario-mcp.log for the [cmbagent_lg] banners and deep_research COMPLETE (n/n steps).denario_results(restart_at_step=N) resumes from the checkpoint (N = last fulfilled step + 1, read from experiment_output/logs/deep_research_run.json).Paper → denario_paper(project_dir=$0, add_citations=False) (pass add_citations=True only if --citations). Confirm paper.pdf exists.
Report back: the idea (1 line), n results steps + whether all fulfilled, number of plots propagated, and the paper.pdf path + size. Note any stage that halted and why (from the log / denario_status).
denario_eda and the cmbagent-keyword paper path need the legacy cmbagent package (not in the cmbagent-lg venv) — skip them.max_n_steps small for iteration; raise it once the science is dialed in.npx claudepluginhub borisbolliet/denario-claude-plugin --plugin denarioGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.