Use this when the user asks whether a product, business, marketing, pricing, ranking, onboarding, marketplace, policy, or engineering change caused a metric change. Helps translate business questions into causal specs, inspect data, select causal inference methods, run checks, and generate decision-grade reports.
How this skill is triggered — by the user, by Claude, or both
Slash command
/causal-decision-debugger:diagnoseThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Help users move from correlation-based claims to careful causal claims with explicit assumptions, data checks, estimates, robustness checks, and recommended next actions.
reference/assumption-ledger.mdreference/causal-glossary.mdreference/causal-spec-schema.mdreference/method-router.mdreference/report-template.mdreference/sql-safety-rules.mdscripts/_install_paths.pyscripts/bootstrap.pytemplates/assumption_ledger.yamltemplates/causal_spec.yamltemplates/dag.jsontemplates/data_sources.yamltemplates/method_plan.jsontemplates/report.mdtemplates/technical_appendix.mdvendor/causal_debugger-0.2.0-py3-none-any.whlvendor/manifest.jsonHelp users move from correlation-based claims to careful causal claims with explicit assumptions, data checks, estimates, robustness checks, and recommended next actions.
The skill ships a bundled Python wheel. Before running any deterministic script, ask the user for permission to bootstrap, then execute:
python ${CLAUDE_PLUGIN_ROOT}/skills/diagnose/scripts/bootstrap.py
Bootstrap is idempotent (skips if the bundled wheel's metadata fingerprint already matches the installed venv) and stdlib-only. It creates an isolated virtualenv at ${CLAUDE_PLUGIN_DATA}/venv/ and pip installs the bundled wheel into it. Network access is required on the first run to resolve heavy transitive dependencies (pandas, scipy, scikit-learn, dowhy, econml) from PyPI; the bundled wheel only contains causal_debugger itself. The first install downloads ~500 MB and takes 5-30 minutes; subsequent runs are instant.
After success, the causal-debugger CLI is auto-on-PATH inside Claude Code via the plugin's bin/ directory. Outside Claude Code, the portable invocation is ${CLAUDE_PLUGIN_ROOT}/bin/causal-debugger (the bin shim resolves the venv path automatically, including when CLAUDE_PLUGIN_DATA is unset). You can also invoke the venv binary directly at ${CLAUDE_PLUGIN_DATA}/venv/bin/causal-debugger; when CLAUDE_PLUGIN_DATA is unset, the bootstrap fallback is ~/.claude/plugins/data/causal-decision-debugger/venv/bin/causal-debugger.
causal_spec.yaml and validate it via causal-debugger validate-spec <path>.causal-debugger check-timestamps.assumption_ledger.yaml (its schema is enforced by the same CLI).causal-debugger suggest-method and write method_plan.json.sql-safety-reviewer agent.report.md and technical_appendix.md.Delegate to these subagents (bundled with the plugin under agents/):
data-scout — find candidate data sources.sql-safety-reviewer — review SQL for safety and causal correctness.causal-methodologist — pick an identification strategy.assumption-ledger-agent — maintain the assumption ledger.report-writer — author the business report and appendix.After bootstrap, invoke via the causal-debugger console script (CWD-independent):
causal-debugger validate-spec <causal_spec.yaml> — schema validation + cross-field rules.causal-debugger profile <data.parquet> — missingness, cardinality, timestamp ranges.causal-debugger check-timestamps <data.parquet> — confirms treatment time precedes outcome time.causal-debugger check-balance <data.parquet> — covariate balance / SMD.causal-debugger suggest-method <causal_spec.yaml> — method routing per spec §10.causal-debugger report <analysis_dir> — render report.md from artifacts.causal-debugger pipeline <analysis_dir> — full orchestration.causal-debugger doctor — environment diagnostics for bug reports.Place artifacts under analysis/<analysis_id>/. Required:
causal_spec.yamlassumption_ledger.yamldata_sources.yamlmethod_plan.jsonreport.mdtechnical_appendix.mdOptional:
dag.jsonqueries/*.sqlartifacts/eda_summary.jsonartifacts/balance_check.jsonartifacts/timestamp_check.jsonartifacts/estimates.json (validates against estimate_result.schema.json)artifacts/robustness.json (each entry validates against refutation_result.schema.json)When a method cannot be applied, write artifacts/identifiability_failure.json (validates against identifiability_failure.schema.json) and call out the result in the report.
reference/method-router.md — situation → method table.reference/sql-safety-rules.md — SQL correctness and safety rules.reference/assumption-ledger.md — guidance for ledger entries.reference/causal-spec-schema.md — pointer to the JSON schema and field meanings.reference/report-template.md — required sections + language rules.reference/causal-glossary.md — vocabulary aligned with the spec.Provides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Searches MemPalace before answering questions about past work, people, projects, or prior decisions. Returns verbatim stored content instead of guessing from model memory.
npx claudepluginhub theillusionoflife/causal_decision_debugger --plugin causal-decision-debugger