Translate business questions into causal specs, audit data, pick a causal inference method, run estimation + refutation, and produce a decision-grade report. "Not identifiable" is a first-class outcome.
Converts user answers, data checks, and causal structure into explicit assumptions with status, evidence, risk, and needed follow-up.
Selects appropriate causal designs and methods based on question, data structure, treatment assignment, timing, and assumptions.
Finds relevant tables, schemas, SQL files, dbt models, dashboards, event definitions, and data sources for causal analysis.
Writes business-readable causal reports and technical appendices from causal specs, data audits, method plans, estimates, and assumptions.
Reviews SQL for safety, query cost, PII leakage, accidental writes, full scans, timestamp mistakes, and causal-analysis correctness.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
A Claude Code Skill plus Python toolkit that helps teams move from correlation-based claims to decision-grade causal evidence. It translates business questions into causal questions, audits the data, picks an appropriate identification strategy, runs estimation and refutation checks, and writes a business-readable report. "Not identifiable" is a valid (and valuable) answer.
skills/diagnose/ + agents/) with subagents for data discovery, SQL safety review, methodology, assumption ledgering, and report writing.examples/onboarding_retention/ that runs the full pipeline on synthetic data.In Claude Code:
/plugin marketplace add github:TheIllusionOfLife/causal_decision_debugger
/plugin install causal-decision-debugger
The plugin bundles the causal_debugger Python wheel under skills/diagnose/vendor/. On first invocation, Claude will run a stdlib-only bootstrap.py that creates an isolated virtualenv at ${CLAUDE_PLUGIN_DATA}/venv/ (the per-plugin persistent data dir) and pip installs the bundled wheel into it. Inside Claude Code, the plugin's bin/causal-debugger shim is auto-on-PATH and dispatches into the venv. Outside Claude Code, invoke the venv's binary directly (~/.claude/plugins/data/causal-decision-debugger*/venv/bin/causal-debugger) or add it to your shell PATH.
Network required on first run. The bundled wheel is causal_debugger itself; transitive dependencies (pandas, scipy, scikit-learn, statsmodels, linearmodels, dowhy, econml, pyyaml, jinja2, jsonschema, pyarrow) still resolve from PyPI on first install. The first install downloads ~500 MB and takes 5-30 minutes; subsequent runs are instant. Python 3.11+.
Windows note. The bin/causal-debugger shim relies on a Unix-style shebang and executable bit. On Windows the venv is still created (under ${CLAUDE_PLUGIN_DATA}\venv\Scripts\), but if the bin/ shim doesn't auto-resolve via Claude Code's Bash tool, invoke python skills/diagnose/scripts/bootstrap.py and then call the venv's causal-debugger.exe directly.
After install, ask Claude:
Use the causal decision debugger to investigate whether onboarding_v2 improved D7 retention. Do not export PII. Generate a business report and technical appendix.
The Skill will produce:
analysis/<id>/causal_spec.yamlanalysis/<id>/assumption_ledger.yamlanalysis/<id>/method_plan.jsonanalysis/<id>/report.mdanalysis/<id>/technical_appendix.mdcausal-debugger pipeline examples/onboarding_retention
This validates the spec, runs balance and timestamp checks, picks a method, estimates the effect, runs refutation, and renders report.md. If data/ is missing under the example, regenerate it via uv run python examples/onboarding_retention/generate_synthetic.py (the script imports causal_debugger, so a bare python from outside a uv sync checkout will not find it; from a plugin-only install, run causal-debugger doctor and use the reported python_executable).
uv sync --extra dev # set up dev environment
uv run pytest # full test suite
uv run ruff check . # lint
uv build # build the wheel into dist/
Two ways to load the plugin during development:
.claude/skills/diagnose -> ../../skills/diagnose and .claude/agents -> ../agents. Both are gitignored. Claude Code picks them up when you launch from this directory.--plugin-dir flag: claude --plugin-dir . from the repo root. Doesn't require symlinks — useful for one-off testing in a clean directory.After editing SKILL.md, an agent file, or the manifest, run /reload-plugins inside Claude Code to pick up changes without restarting.
After modifying src/causal_debugger/, rebuild the bundled wheel and refresh its manifest so the install-test CI does not fail:
uv build
cp dist/causal_debugger-*.whl skills/diagnose/vendor/
# Regenerate sha256 + metadata_sha256 in skills/diagnose/vendor/manifest.json
npx claudepluginhub theillusionoflife/causal_decision_debugger --plugin causal-decision-debuggerComprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
A growing collection of Claude-compatible academic workflow bundles. Covers scientific figures, manuscript writing and polishing, reviewer assessment, citation retrieval, data availability, paper reading, literature search, response letters, paper-to-PPTX conversion, and evidence-grounded Chinese invention patent drafting. Rules are organized as reusable skill folders with explicit workflows and quality checks.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Comprehensive PR review agents specializing in comments, tests, error handling, type design, code quality, and code simplification
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.