Repository navigation
Releases
v2.0.2 — CatBoost selection guidance
Documentation release.
Changed
- README: added CatBoost selection guidance to the agent-ensemble section — tool-scout's practical recommendation from the Telco Churn validation run. Prefer CatBoost on categorical-heavy data (many high-cardinality categorical columns, few numeric features): native categorical handling and ordered target encoding that guards against leakage, often edging out LightGBM at the cost of slower training. Keep LightGBM for large mostly-numeric data or when training speed matters; dispatch tool-scout to verify the best fit for your specific data.
Full changelog: see CHANGELOG.md
v2.0.1 — Reliable leakage signals in model-diagnostician
Patch release refining the model-diagnostician agent after an end-to-end validation run on the Telco Customer Churn dataset.
Changed
model-diagnosticianleakage-recheck signals reordered by reliability:- Anomalous score jump (AUC/R² spikes when a feature is added) — primary suspicion.
- Feature-vs-label consistency check (>95% ⇒ the feature is a relabeling/downstream of the target).
- Importance via gain/SHAP, not split-count.
- Documented that a binary leakage feature can be buried under continuous features in default split-count importance, making split-count unreliable for leakage detection — the gain/SHAP view (plus the score-jump and consistency signals) is what reliably exposes it.
- Updated both the portable role-prompt (
references/agents/model-diagnostician.md) and the plugin wrapper (agents/model-diagnostician.md).
Full changelog: see CHANGELOG.md
v2.0.0 — Agent ensemble & risk-proportional leakage detection
Major release introducing a 4-agent ensemble and reframing data-leakage detection to be risk-proportional.
Added
- 4-agent ensemble for more comprehensive, rigor-checked analysis (portable role-prompts in
references/agents/as the single source of truth, plus Claude plugin wrappers inagents/):methodology-architect— pre-modeling analysis blueprint; DL-aware (DL is a must-be-justified candidate on tabular data, GBDT usually wins); emits a leakage-risk rating + list.model-diagnostician— adversarial post-evaluation failure hunting; explicit overfitting (train-vs-test) check distinguished from leakage; model-family-agnostic.tool-scout— web-search-backed library/model recommendation (LightGBM is the default, not automatically best); clearly-flagged offline fallback.professor-reviewer— professor-perspective final review with grade + remediation checklist; does not auto-rerun.
.claude-plugin/plugin.jsonso the repository can be used as a Claude Code plugin (agents available natively).- Agent-ensemble sections in
SKILL.md,METHODOLOGY.md, andREADME.md.
Changed
- Reframed data-leakage detection from an absolute, always-mandatory step to risk-proportional screening: the screening question is always asked (near-zero cost), but investigation depth scales with data provenance. Aligned with the existing "match method to constraint" principle.
Full changelog: see CHANGELOG.md
v1.1.0 - Rename to Predictive Analytics
Highlights
This release rebrands the skill to better reflect its core — predictive analytics and machine learning on structured (tabular) data — with naming that reads clearly for non-technical audiences such as MBA and finance users.
Changed
- Renamed the skill from
structured-data-analysistopredictive-analytics. The skillname, document titles, repository URLs, and the project directory were all updated to match.
Added
- README now includes the repository link, clone instructions, and project-status badges.
Full details in CHANGELOG.md.
v1.0.0 — Initial release
First public release of the Structured Data Analysis Skill — a complete, reusable, multi-model and language-adaptive methodology for end-to-end machine learning analysis of structured (tabular) data.
Highlights
- Dual-format architecture —
SKILL.md(Claude entry point with YAML header) andMETHODOLOGY.md(plain-prompt entry point for ChatGPT, Gemini, DeepSeek, Qwen, Kimi, or any LLM via API), sharing one platform-neutralreferences/core. - Language-adaptive output — the analysis is produced in whatever language the user asks in, chart labels included.
- Seven deep-dive guides covering the full workflow:
- Data import & cleaning (with data-leakage detection)
- Exploratory data analysis (with chart-font setup)
- Feature engineering
- Modeling (classification & regression)
- Evaluation, validation & diagnosis (cross-validation, grid search, residual diagnosis, data-driven thresholds)
- Clustering (with PCA-based high-dimensional visualization)
- Interpretability, math expression & output
- Embedded "analysis spirit" — leakage-first, honest reporting, baseline comparison, method-matches-constraint, data-driven thresholds, cross-validate and interpret.
- Domain-agnostic — works for customer churn, demand forecasting, risk prediction, site selection, sensor data, and more. A city-noise dataset is the running example.
Getting started
- Claude: install
SKILL.mdas a Skill, or paste its contents into the chat. - Other LLMs: paste the contents of
METHODOLOGY.mdas a system prompt.
See the README for full usage.