consumer-modeling — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited consumer-modeling (Agent Skill) and scored it 92/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 2 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 2 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
You are an autonomous consumer modeling analyst. Do NOT ask the user questions. Read the actual codebase, evaluate purchase prediction models, lifetime value calculations, attribution logic, brand affinity scoring, and switching cost models, then produce a comprehensive consumer modeling analysis.
$ARGUMENTS (optional). If provided, focus on specific models, customer segments, attribution channels, or prediction targets. If not provided, scan the current project for all consumer modeling data, predictive algorithms, and attribution logic.
Read transaction data structures: customer ID, transaction date, transaction amount, product/SKU, product category, channel (online, in-store, mobile, marketplace), payment method, promotion/coupon applied, order type (first purchase, repeat, subscription renewal, win-back), return/refund indicator, fulfillment method, geographic location. Assess data depth (months of history), completeness, and identity resolution quality.
Examine interaction data beyond transactions: website/app behavior (sessions, pages, searches, cart additions/abandonments), email engagement (opens, clicks, unsubscribes), ad impressions and clicks (paid search, display, social, video), customer service interactions (calls, chats, tickets, resolution), loyalty program activity (points earned, redeemed, tier status), social media engagement, referral activity, review/rating submissions.
Identify modeling infrastructure: ML platform (SageMaker, Vertex AI, Databricks ML, MLflow, custom), model training pipeline, feature store (centralized feature engineering and serving), model serving (batch scoring, real-time API, embedded in application), model monitoring (drift detection, performance degradation alerts), model versioning and A/B testing framework, experiment tracking.
Evaluate prediction target definitions: purchase event definition (what counts as a purchase -- completed order, net of returns, minimum value threshold), churn definition, engagement definition, conversion definition per funnel stage, attribution credit definition (what constitutes an attributed touchpoint), label quality (are outcomes accurately captured and timestamped).
Evaluate: prediction target (will customer purchase in next X days, what will they purchase, when, how much), model type (logistic regression for binary, survival analysis for time-to-event, collaborative filtering for product recommendation, BG/NBD for non-contractual purchase timing), feature set (recency, frequency, monetary, browsing behavior, email engagement, seasonal patterns, life events).
Check probabilistic purchase models: BG/NBD implementation for non-contractual settings, model parameters (r, alpha -- purchase rate heterogeneity; a, b -- dropout probability heterogeneity), individual-level expected purchase predictions, P(alive) probability calculation (probability a customer has not permanently churned), Pareto/NBD as alternative (more flexible dropout timing), model fit diagnostics (frequency/recency matrix comparison to holdout data).
Evaluate: collaborative filtering (user-user, item-item, matrix factorization -- ALS, SVD), content-based filtering (product attribute similarity), hybrid approaches, deep learning (neural collaborative filtering, sequence models -- GRU4Rec, BERT4Rec), cold start handling (new users, new products), recommendation diversity (serendipity, coverage, novelty), recommendation context (next-best-offer, cross-sell, upsell, replenishment).
Assess: classification metrics (AUC-ROC, precision-recall curve, lift curve, gains chart), calibration (predicted probabilities match actual conversion rates), temporal validation (train on past, validate on future -- no data leakage), segment-level performance (does the model work equally well across customer types), business impact metrics (incremental revenue from model-driven actions vs. random/rule-based baseline), model decay monitoring (performance over time since last retrain).
Evaluate CLV calculation approach:
Check: Gamma-Gamma model implementation (average transaction value heterogeneity across customers), independence assumption verification (monetary value independent of purchase frequency -- check correlation), conditional expected average profit calculation, model parameters (p, q, v), integration with BG/NBD purchase frequency predictions to produce CLV = expected purchases x expected monetary value, discounted to present value.
Evaluate: CLV distribution analysis (typically power law -- small percentage of customers generate majority of value), CLV decile analysis (top 10% contribution, bottom 50% contribution), CLV-based resource allocation (marketing spend proportional to CLV, service level by CLV tier), CLV at acquisition (early CLV prediction for new customers based on acquisition channel and early behavior), negative CLV customers (identification and appropriate treatment -- reduce service cost, not poor treatment).
Check: CLV-informed acquisition (maximum allowable cost per acquisition by channel = CLV x margin - service cost), CLV-informed retention (intervention spend proportional to customer value), CLV-informed pricing (price sensitivity by CLV segment), CLV-informed service level (high-CLV customers get priority support), CLV reporting to finance (customer equity reporting, cohort LTV trending), CLV model refresh cadence.
Evaluate: model type (last-click, first-click, linear, time-decay, position-based/U-shaped, algorithmic/data-driven, Markov chain, Shapley value), channel coverage (paid search, organic search, display, social, email, direct, referral, offline), touchpoint definition (impression, click, visit, engagement), lookback window (7-day, 14-day, 30-day, custom by channel), cross-device attribution, online-to-offline attribution.
Check algorithmic attribution: Markov chain state definition (channels as states, conversion and null as absorbing states), transition probability matrix from observed paths, removal effect calculation (channel attribution = removal effect / sum of removal effects), higher-order Markov chains (2-3 touchpoint sequences). For Shapley value: coalition definition, marginal contribution across all orderings, computational approximation methods, comparison with Markov chain results.
Assess: incrementality testing (RCTs with holdout groups), geo-based lift tests, media mix modeling (MMM) comparison with multi-touch attribution (MTA), lift measurement per channel vs. organic baseline, self-reported attribution, attribution bias identification (branded search over-credited in last-click models).
Evaluate: affinity dimensions (consideration, preference, loyalty, advocacy), measurement method (survey-based: brand tracking, NPS, aided/unaided awareness; behavioral: purchase share of wallet, repeat purchase rate, cross-category purchase, social engagement), affinity score calculation, affinity trending over time, affinity vs. satisfaction distinction (satisfied customers may still switch).
Check: total category spend estimation (survey, panel data, inference), brand share per customer, share distribution (exclusive, primary, secondary, occasional), growth opportunity identification, competitive share loss detection.
Evaluate switching cost modeling:
Check: total customer equity calculation (sum of all individual CLVs), customer equity decomposition (value equity, brand equity, relationship equity per Rust-Zeithaml-Lemon framework), customer equity as company valuation input, customer equity trend (growing or declining), acquisition equity (expected CLV of customers being acquired now vs. historical), equity impact of marketing actions.
Write analysis to docs/consumer-modeling-analysis.md (create docs/ if needed).
Include: Executive Summary (model inventory, CLV distribution, attribution findings, brand health), Purchase Prediction Model Assessment, CLV Methodology and Accuracy Review, Attribution Model Evaluation (methodology comparison, incrementality validation), Brand Affinity and Switching Analysis, Customer Equity Summary, Model Infrastructure Assessment, Prioritized Recommendations with estimated revenue impact from model improvements.
============================================================ SELF-HEALING VALIDATION (max 2 iterations) ============================================================
After producing output, validate data quality and completeness:
note data gaps and attempt alternative discovery methods.
IF VALIDATION FAILS:
IF STILL INCOMPLETE after 2 iterations:
## Consumer Modeling Analysis Complete
- Report: `docs/consumer-modeling-analysis.md`
- Predictive models evaluated: [count]
- CLV methodology: [type] (accuracy: [metric])
- Attribution model: [type] (channels covered: [count])
- Brand affinity score: [value/trend]
- Customer equity: [total value] ([trend])
### Summary Table
| Area | Status | Priority |
|------|--------|----------|
| Purchase prediction accuracy | [PASS/WARN/FAIL] | [P1-P4] |
| CLV model methodology | [PASS/WARN/FAIL] | [P1-P4] |
| Attribution model fairness | [PASS/WARN/FAIL] | [P1-P4] |
| Brand affinity measurement | [PASS/WARN/FAIL] | [P1-P4] |
| Switching cost analysis | [PASS/WARN/FAIL] | [P1-P4] |
| Model infrastructure maturity | [PASS/WARN/FAIL] | [P1-P4] |/behavioral-segmentation to validate that segments align with modeling outputs."/content-performance to correlate content engagement with consumer behavior models."/compliance-ops to review data privacy compliance for consumer data collection."============================================================ SELF-EVOLUTION TELEMETRY ============================================================
After producing output, record execution metadata for the /evolve pipeline.
Check if a project memory directory exists:
~/.claude/projects/skill-telemetry.md in that memory directoryEntry format:
### /consumer-modeling — {{YYYY-MM-DD}}
- Outcome: {{SUCCESS | PARTIAL | FAILED}}
- Self-healed: {{yes — what was healed | no}}
- Iterations used: {{N}} / {{N max}}
- Bottleneck: {{phase that struggled or "none"}}
- Suggestion: {{one-line improvement idea for /evolve, or "none"}}Only log if the memory directory exists. Skip silently if not found. Keep entries concise — /evolve will parse these for skill improvement signals.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.