Contract
- Input: ML use cases, feature candidates, latency requirements, SLA.
- Output: feature store architecture + feature definitions + monitoring plan.
- Scope: designs feature store; execution requires approval.
- Rule: designs feature store; execution requires approval.
- Rule: documented standards override defaults; explicit project rules take precedence.
- Rule: if blocked by missing context or dependencies, surface the blocker before proceeding.
Provenance
| Question |
Answer |
| What is the source of truth? |
The user's request, originating spec/issue, and the skill's declared outputs |
| What is in scope? |
Work covered by the skill's acceptance criteria and completion rules |
| What is explicitly out of scope? |
Files, behaviors, and decisions outside the skill's declared boundary |
| Who or what consumes this artifact afterward? |
The next skill in the workflow or the user |
| What evidence proves it is done? |
Completion criteria met, artifact saved, validation passed |
| What risk remains? |
Subjective judgment calls, missing context, or external dependency failures |
Artifact
Emit FeatureStoreArtifact as both:
- JSON:
.agents/skills/platform/artifacts/feature-store/{identifier}.json
- Markdown view: same filename with
.md extension
Feature Store
Design a feature store — online/offline, feature engineering, serving, monitoring, and drift detection.
Process
1. Use case and requirements
- Inference latency: real-time (ms), batch (minutes/hours).
- Feature types: categorical, numerical, embedding, time-series.
- Freshness: real-time, near-real-time, daily.
- Team: data scientists, ML engineers, platform team.
Completion criterion: use cases and requirements saved.
2. Architecture design
- Online store: Redis, DynamoDB, Cassandra, Feast.
- Offline store: S3, GCS, data warehouse, Delta Lake.
- Serving layer: REST / gRPC / feature server.
- Transformation: batch (Spark, dbt) vs streaming (Flink, Kafka Streams).
Completion criterion: architecture diagram saved.
3. Feature definitions and schemas
- Feature: name, description, type, entity, value type, TTL.
- Entities: user, item, session, device.
- Features: user features (history, demographics), item features (attributes, popularity).
- Point-in-time correctness: avoid leakage with event-time joins.
Completion criterion: feature definitions and schemas saved.
4. Feature engineering
- Reusable transformations: standardise, normalise, encode, aggregate.
- Feature pipelines: batch backfill vs streaming updates.
- Testing: unit tests for transformations, data quality checks.
Completion criterion: transformation logic documented.
5. Serving and latency
- Online serving: low-latency lookups, caching, TTL eviction.
- Consistency: online vs offline alignment, point-in-time joins.
- Scalability: sharding, replication, load balancing.
Completion criterion: serving design saved.
6. Monitoring and drift detection
- Feature quality: null rate, type violations, out-of-range.
- Feature drift: distribution shift (KS test, PSI, Wasserstein).
- Alerting: thresholds, notifications, on-call runbook.
- Model impact: link feature drift to model performance degradation.
Completion criterion: monitoring and drift plan saved.
Completion
- the skill's completion criteria are explicitly checked
- the artifact is saved and validated
- any blockers or skipped validations are documented
- the next consumer is identified or the work is handed off
@include .agents/skills/platform/contract-base.xml
1---2name: "feature-store"3category: "data"4maturity: "stable"5version: "1"6description: "Feature store (online/offline, feature engineering, serving, monitoring, drift)."7license: "MIT"8capabilities:9 - apply feature store methods within explicit data boundaries10 - analyze feature store (online/offline, feature engineering, serving, monitoring, drift) with assumptions, constraints, and failure modes11 - produce and validate a data artifact with evidence, decisions, and next steps12outputs: "feature store architecture + feature definitions + monitoring plan."13sideEffects: []14dependencies: []15stopCondition: "Architecture diagram saved; feature definitions documented; monitoring plan present."16risk: "medium"17trustTier: "3"18maxIterations: "6"19promptVersion: "2.0"20artifactType: "data"21modelTier: "reasoning"22evaluators: ["behavioral", "regression"]23fixturesPath: .agents/skills/platform/fixtures/behavioral/feature-store.json24diataxis: "how-to"25tags: ["data"]26compatibility: []27approvalRequired: "true"28approvalFor: []29metadata:30 source: https://skillmd.com/skills/quantumquirkxyz/feature-store31---3233## Contract3435- Input: ML use cases, feature candidates, latency requirements, SLA.36- Output: feature store architecture + feature definitions + monitoring plan.37- Scope: designs feature store; execution requires approval.38- Rule: designs feature store; execution requires approval.39- Rule: documented standards override defaults; explicit project rules take precedence.40- Rule: if blocked by missing context or dependencies, surface the blocker before proceeding.4142## Provenance4344| Question | Answer |45|---|---|46| What is the source of truth? | The user's request, originating spec/issue, and the skill's declared outputs |47| What is in scope? | Work covered by the skill's acceptance criteria and completion rules |48| What is explicitly out of scope? | Files, behaviors, and decisions outside the skill's declared boundary |49| Who or what consumes this artifact afterward? | The next skill in the workflow or the user |50| What evidence proves it is done? | Completion criteria met, artifact saved, validation passed |51| What risk remains? | Subjective judgment calls, missing context, or external dependency failures525354## Artifact5556Emit `FeatureStoreArtifact` as both:57- JSON: `.agents/skills/platform/artifacts/feature-store/{identifier}.json`58- Markdown view: same filename with `.md` extension596061# 62# 63# Feature Store6465Design a **feature store** — online/offline, feature engineering, serving, monitoring, and drift detection.6667## Process6869### 1. Use case and requirements70- Inference latency: real-time (ms), batch (minutes/hours).71- Feature types: categorical, numerical, embedding, time-series.72- Freshness: real-time, near-real-time, daily.73- Team: data scientists, ML engineers, platform team.7475**Completion criterion:** use cases and requirements saved.7677### 2. Architecture design78- Online store: Redis, DynamoDB, Cassandra, Feast.79- Offline store: S3, GCS, data warehouse, Delta Lake.80- Serving layer: REST / gRPC / feature server.81- Transformation: batch (Spark, dbt) vs streaming (Flink, Kafka Streams).8283**Completion criterion:** architecture diagram saved.8485### 3. Feature definitions and schemas86- Feature: name, description, type, entity, value type, TTL.87- Entities: user, item, session, device.88- Features: user features (history, demographics), item features (attributes, popularity).89- Point-in-time correctness: avoid leakage with event-time joins.9091**Completion criterion:** feature definitions and schemas saved.9293### 4. Feature engineering94- Reusable transformations: standardise, normalise, encode, aggregate.95- Feature pipelines: batch backfill vs streaming updates.96- Testing: unit tests for transformations, data quality checks.9798**Completion criterion:** transformation logic documented.99100### 5. Serving and latency101- Online serving: low-latency lookups, caching, TTL eviction.102- Consistency: online vs offline alignment, point-in-time joins.103- Scalability: sharding, replication, load balancing.104105**Completion criterion:** serving design saved.106107### 6. Monitoring and drift detection108- Feature quality: null rate, type violations, out-of-range.109- Feature drift: distribution shift (KS test, PSI, Wasserstein).110- Alerting: thresholds, notifications, on-call runbook.111- Model impact: link feature drift to model performance degradation.112113**Completion criterion:** monitoring and drift plan saved.114115## Completion116117- the skill's completion criteria are explicitly checked118- the artifact is saved and validated119- any blockers or skipped validations are documented120- the next consumer is identified or the work is handed off121---122@include .agents/skills/platform/contract-base.xml