메뉴
Data cleaning, profiling, transformation, and quality gates -- prepares raw data for visualization and analysis
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/fabioc-aloha/Alex_ACT_Visual_Storytelling --skill data-preparation명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Clear documentation through visual excellence
Ingestion patterns for CSV, JSON, REST API, SQL, Excel, and Parquet -- guides an LLM through loading data from any common source
Render data dashboards as pure ASCII art in monospace text -- the cheapest, most portable delivery method. No rendering engine, no SVG, no browser. LLM-native output with predictable character geometry.
SKILL.md 표시 중
| type | skill |
| lifecycle | stable |
| inheritance | inheritable |
| name | data-preparation |
| description | Data cleaning, profiling, transformation, and quality gates -- prepares raw data for visualization and analysis |
| tier | standard |
| applyTo | **/*data*,**/*clean*,**/*transform*,**/*prep*,**/*pivot*,**/*aggregate* |
| currency | 2026-08-15T00:00:00.000Z |
| lastReviewed | 2026-08-15T00:00:00.000Z |
Prepare raw data for visualization. This module fires after ingestion and before chart selection. It does not analyze or visualize; it makes data ready for those steps.
State why you are skipping. Silent skips hide data problems.
Before cleaning, understand what you have. Run these checks on every column:
| Check | What to look for |
|---|---|
| Row count | Is it what you expected? Zero rows = ingestion failed |
| Column types | Dates stored as strings? Numbers as text? |
| Null count | Which columns have nulls? What percentage? |
| Unique count | Cardinality: is this a dimension (low) or measure (high)? |
| Min / Max / Mean | For numeric columns: are ranges plausible? |
| Sample values | Eyeball 5 rows. Do they look right? |
Report the profile as a table before proceeding. The user (or orchestrator) needs to see it to decide what cleaning is needed.
Apply fixes in this order. Each fix is a decision; state what you chose and why.
| Signal | Action |
|---|---|
| Exact duplicate rows | Remove. Log count removed. |
| Near-duplicates (same key, different values) | Flag for user decision. Do not silently drop. |
| Strategy | When to use |
|---|---|
| Drop row | Null is in a must-have column and row is < 5% of data |
| Fill with default |
| Column has an obvious default (e.g., 0 for optional count) |
| Fill with median/mode | Numeric column, nulls are < 10%, distribution is stable |
| Leave as-is | The visual can handle nulls (e.g., gap in a line chart) |
Never fill nulls silently. State the strategy and count affected.
| Problem | Fix |
|---|---|
| Date as string | Parse to date. State the format detected. |
| Number as string | Convert. Check for non-numeric values first. |
| Boolean as string | Map "yes/no", "true/false", "1/0" to boolean. |
| Mixed types in one column | Flag for user. This usually means the schema is wrong. |
| Context | Action |
|---|---|
| Outlier is a data error (negative revenue, future date) | Remove or fix. Log. |
| Outlier is real but extreme | Keep. Note it for the visual (it may need axis adjustment). |
| Unsure | Flag for user. Do not remove real data silently. |
Reshape the data to match what the target visual needs.
Aggregate when the raw data is too granular for the visual.
| Visual need | Aggregation |
|---|---|
| Monthly trend | Group by month, sum or average the measure |
| Category comparison | Group by category, sum the measure |
| Proportion | Group by category, compute share of total |
| Distribution | Bin the values, count per bin |
Always state the grain change: "Aggregated from daily to monthly by summing revenue."
| Need | Direction |
|---|---|
| One column per category (wide format) | Pivot |
| One row per observation (long format) | Unpivot |
Most visualizations prefer long format. Pivot only when the delivery module explicitly needs wide format (e.g., side-by-side bar chart in ASCII).
Add columns only when the brief requires a metric that is not in the raw data:
Name computed columns clearly. Do not overwrite source columns.
Every metric passed downstream must carry enough lineage to reproduce it:
| Formula | Source fields | Grain | Units | Rounding | Baseline |
|---|---|---|---|---|---|
| (current - previous) / previous | revenue, date | month | percent | 1 decimal | previous month |
Apply these rules:
Before passing data downstream, assert:
| Gate | Assertion | Fail action |
|---|---|---|
| No nulls in key columns | Dimensions and must-have measures are null-free | Push back: data is not ready |
| Row count is plausible | Within expected range (not 0, not 10x expected) | Push back: ingestion may have failed |
| Types are correct | Every column has the expected type after coercion | Push back: type fix needed |
| Grain is correct | One row per expected unit (e.g., one per month per region) | Push back: dedup or aggregate needed |
If any gate fails, push back to the orchestrator with a specific message: "Quality gate failed: 12 nulls in revenue column. Recommend: drop rows (< 1% of data)."
Pass to the next module:
| Mall plugin | This module's role |
|---|---|
data-analysis | Runs after this module; assumes clean data |
data-quality-monitoring | Monitors ongoing quality; this module does initial prep |
data-visualization | Designs the visual; needs correctly shaped data from this module |
| Do not | Do instead |
|---|---|
| Clean silently | Log every change with count and reason |
| Remove outliers without context | Keep real data; flag for user if unsure |
| Aggregate without stating the grain change | "Aggregated from X to Y by Z" |
| Add computed columns the brief did not ask for | Only add what the questions require |
| Skip profiling because "the data looks fine" | Profile always; it costs one table and catches surprises |
