beaver-data

Reproducible, stage-gated data analysis for AI coding agents: requirements interrogation, Bash orchestration, user-approved statistical engines, independent cross-engine validation, and LaTeX reporting.

Part of a four-skill research suite: beaver-data (analysis), weaver-text (editing support for author-written text), honeybee-lit (general literature reviews), and wasp-lit (post-GWAS candidate gene-phenotype literature analysis).

Why a beaver?

A beaver builds dams in disciplined layers, listens for leaks, and patches weak points before they fail. beaver-data treats an analysis the same way: every stage is checked before the next is built on it.

Reproducibility and standardization

• Recorded decisions. Every request, including reanalyses and follow-ups, starts with a requirements interview ("grilling"). Estimands, exclusions, and cleaning rules are confirmed and saved in plans/ before any code runs.
• Immutable inputs. Raw data are never modified; cleaned data are regenerated by tracked scripts.
• Pinned toolchains. Each tool's version or container digest, seed, thread count, locale, and numerical options are recorded. R environments use renv; other engines use lockfiles.
• Traceable runs. Each run has a sortable YYYYMMDD_HHMMSS run ID. Git tracks scripts, and logs/analysis-index.csv links runs to outputs.
• Typed plain-text data. CSV/TSV files with explicit column types prevent NA and date auto-conversion errors.
• Independent validation. A firewalled brief lets a second engine reproduce key results without reusing the primary code, outputs, or libraries.
• One project layout and report format. Every project uses the same flat directories. Reports follow a fixed LaTeX format with machine-readable file-tracking comments.

What it does

• Stress-tests the analysis plan across decision trees before touching data.
• Runs a 30-item forensic EDA checklist and reports findings before any cleaning or modeling.
• Uses R by default, or any user-approved language, Linux tool, container, or domain pipeline.
• Tracks exploratory branches and retractions without losing provenance.
• Produces diagnostics and figures in accessible palettes, plus standalone LaTeX reports.

Package contents

beaver-data/
├── SKILL.md                              # Agent instructions
├── README.md
├── LICENSE
├── assets/beaver-data.jpg
└── references/
    ├── 01_foundations_git_and_bash.md    # Layout, Git, environments, run index, Bash
    ├── 02_grilling_and_intake.md         # Decision tree, intake records, variable dictionary
    ├── 03_forensics_and_cleaning.md      # EDA checklist, findings report, cleaning, TSV types
    ├── 04_inference_and_displays.md      # Tool selection, analysis and display plans
    └── 05_validation_and_reports.md      # Cross-engine validation, LaTeX report

Project workspace

project/
├── data/         # Immutable raw data and generated clean CSV/TSV
├── scripts/      # Tracked analysis scripts
├── plans/        # Grilling summaries, intake records, dictionaries, analysis plans
├── logs/         # Run logs, checkpoints, analysis-index.csv
├── issues/       # Edge cases and resolution records
├── figures/      # Diagnostic and report figures
├── results/      # Plotted data, model summaries, LaTeX table fragments
├── reports/      # EDA summaries and compiled reports (.tex, .pdf)
├── validation/   # Cross-engine briefs, independent implementations, reviews
└── renv/         # R environment, when R is used

Companion skills

Skill	Role
beaver-data	Approved statistical analysis, including quantitative evidence synthesis.
weaver-text	AI editing, review, and submission help for author-written manuscripts and grant proposals.
honeybee-lit	General literature reviews, from citation checks to narrative, scoping, and systematic reviews.
wasp-lit	Post-GWAS literature analysis of candidate gene-phenotype relationships.

Installation

Clone the repository, unchanged, into a skills location your agent scans:

# Pi
git clone https://github.com/chen42/beaver-data.git ~/.pi/agent/skills/beaver-data
# Claude Code (user or project scope)
git clone https://github.com/chen42/beaver-data.git ~/.claude/skills/beaver-data

In Pi, run /reload after cloning, or list the path under "skills" in ~/.pi/agent/settings.json.

Requirements: Bash, Git, LaTeX, and the analysis engines approved for each project (R by default).

License

MIT