- name
SKILL.md
Le fichier du skill reste dans le dépôt de l'éditeur.
Collez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add retentioneering/retentioneering-tools --skill retentioneering-product-analytics -a cursor -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add retentioneering/retentioneering-tools --skill retentioneering-product-analytics -a claude-code -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add retentioneering/retentioneering-tools --skill retentioneering-product-analytics -a codex -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add retentioneering/retentioneering-tools --skill retentioneering-product-analytics -a grok -yEnvoyez le message copié dans un chat Grok Bot. Le Bot récupère ce SKILL.md sur GitHub et l'ajoute aux skills que partagent tous vos Bots.
Install this skill: https://github.com/retentioneering/retentioneering-tools/tree/HEAD/.claude/skills/retentioneering-product-analyticsDe l'éditeur
Turn event-level behavioral data into a reproducible answer to a product question — why users convert, churn, loop, or abandon — using user trajectories, transitions, funnels, and behavioral segments.
Do not merely generate visualizations. Connect each output to the question, separate observation from interpretation, and never present path correlations as causal effects.
Minimum: a path identifier (user or session), an event name, a timestamp (or a reliable order column — see gotcha G2 for order-only data). Useful extras: session id, segment attributes (device, source, plan), event properties, conversion labels.
python -c "import retentioneering; print(retentioneering.__version__)".
Expect 5.x; this skill's API map is version-verified for 5.0 — on a different major
version, trust installed docstrings over the map.references/api-map.md must be verified
against the installed package before use.Run scripts/inspect_event_log.py <path> [--sep ...] (or replicate its checks inline for
in-memory frames). It profiles columns, infers the user/event/timestamp mapping, checks
timestamp parseability, duplicates, per-path ordering, path-length distribution, and
emits artifacts/data-profile.json plus a ready-to-paste Eventstream(...) schema.
Report to the user before proceeding: inferred mapping, row/user/event-type counts, covered period, and any red flags (nulls in key columns, timestamp ties, suspected bots or ultra-long paths, order-only timestamps). Confirm the mapping if inference is ambiguous.
Map the question to a recipe in references/analysis-recipes.md:
navigation structure/loops → transition graph (R1/R6) · before/after an anchor →
step matrix (R3) · ordered conversion flow → funnel (R1/R2) · what winners do
differently → diff on a funnel-stage segment (R2) · heterogeneous users → clustering
without target leakage (R5) · between two funnel levels → truncate micro-journey (R4) ·
intervention timing → time-to-outcome (R7) · cross-segment scan → segment overview (R8) ·
value of a fix → Markov what-if (R9, advanced).
Combine recipes only when each addition resolves a distinct uncertainty.
artifacts/ by default).sample_paths(frac=, random_state=) for stable subsamples; stochastic steps get
explicit seeds.processed.recipe() and the package version into
artifacts/run-metadata.json — any artifact must be regenerable from raw data via
Eventstream.from_recipe(raw_df, recipe).Work through references/gotchas-and-validation.md section B. Non-negotiables:
every percentage names its denominator; population filters are disclosed with counts;
survivorship and exposure confounds addressed; no outcome leakage into features;
small cells flagged with n; caption numbers come from headless *_data twins;
visuals agree with tables.
Structure the final answer as:
Deliverables: analysis script or executed notebook; artifacts/data-profile.json;
artifacts/metrics.csv (key tables); interactive HTML exports via
widget.export_html(..., title=, analysis=) — write analysis= captions AFTER
conclusions are final; artifacts/summary.md (mapping, filters, assumptions, versions,
findings, limitations, next steps); artifacts/run-metadata.json (versions, parameters,
seeds, recipe() lineage).
>
---name: retentioneering-product-analyticsdescription: >---# .claude/skills/retentioneering-product-analytics/SKILL.mdFAQ
Téléchargez .claude/skills/retentioneering-product-analytics/SKILL.md auprès de l'éditeur, ou copiez le chemin et ouvrez le dépôt. PlugMCP répertorie le skill et n'héberge pas le fichier.
L'éditeur le maintient sur https://github.com/retentioneering/retentioneering-tools.
Claude, Cursor, Codex et Grok Build peuvent lire un SKILL.md que vous ajoutez depuis le dépôt de l'éditeur. Grok Bot l'installe lorsque vous envoyez dans un chat le lien du dossier GitHub du skill.