self-improving-analytics

name

self-improving-analytics

description

Captures data quality issues, metric drift, pipeline failures, misleading visualizations, metric definition mismatches, and data freshness problems to enable continuous analytics improvement. Use when: (1) An ETL/ELT pipeline fails, (2) A metric value shows anomalous behavior, (3) Two teams define the same metric differently, (4) A dashboard shows wrong or misleading data, (5) A data freshness SLA is missed, (6) A schema change breaks downstream consumers.

Self-Improving Analytics Skill

Log analytics-specific learnings, data issues, and feature requests to markdown files for continuous improvement. Captures data quality problems, metric drift, pipeline failures, misleading visualizations, metric definition mismatches, and data freshness breaches. Important learnings get promoted to data dictionaries, metric definitions, pipeline runbooks, dashboard standards, or data quality SLAs.

First-Use Initialisation

Before logging anything, ensure the .learnings/ directory and files exist in the project or workspace root. If any are missing, create them:

mkdir -p .learnings
[ -f .learnings/LEARNINGS.md ] || printf "# Analytics Learnings\n\nData quality patterns, metric drift insights, pipeline reliability findings, visualization best practices, and governance lessons.\n\n**Categories**: data_quality | metric_drift | pipeline_failure | visualization_mislead | definition_mismatch | freshness_issue\n**Areas**: ingestion | transformation | modeling | reporting | visualization | governance | data_catalog\n\n---\n" > .learnings/LEARNINGS.md
[ -f .learnings/DATA_ISSUES.md ] || printf "# Data Issues Log\n\nPipeline failures, data quality problems, metric anomalies, visualization errors, and schema drift.\n\n---\n" > .learnings/DATA_ISSUES.md
[ -f .learnings/FEATURE_REQUESTS.md ] || printf "# Feature Requests\n\nAnalytics tools, BI capabilities, data quality automation, and governance improvements.\n\n---\n" > .learnings/FEATURE_REQUESTS.md

Never overwrite existing files. This is a no-op if .learnings/ is already initialised.

Do not log connection strings, database credentials, API keys, or PII. Prefer short summaries or redacted excerpts over raw query results or full table dumps.

If you want automatic reminders, use the opt-in hook workflow described in Hook Integration.

Quick Reference

Situation	Action
ETL/ELT pipeline fails	Log to .learnings/DATA_ISSUES.md with pipeline name and error
Metric value anomaly (spike/drop)	Log to .learnings/DATA_ISSUES.md with statistical context
Two teams define metric differently	Log to .learnings/LEARNINGS.md with category definition_mismatch
Dashboard shows wrong or misleading data	Log to .learnings/LEARNINGS.md with category visualization_mislead
Data freshness SLA missed	Log to .learnings/DATA_ISSUES.md with SLA threshold and actual delay
Schema change breaks downstream	Log to .learnings/DATA_ISSUES.md with schema diff details
NULL rate spike in key column	Log to .learnings/DATA_ISSUES.md with column and threshold
Metric silently drifts (calculation change)	Log to .learnings/LEARNINGS.md with category metric_drift
Recurring data quality pattern	Link with **See Also**, consider priority bump
Broadly applicable pattern	Promote to data dictionary, pipeline runbook, or dashboard standard
Reusable data quality check	Promote to data quality SLA or dbt test

OpenClaw Setup (Recommended)

OpenClaw is the primary platform for this skill. It uses workspace-based prompt injection with automatic skill loading.

Installation

Via ClawdHub (recommended):

clawdhub install self-improving-analytics

Manual:

git clone https://github.com/jose-compu/self-improving-analytics.git ~/.openclaw/skills/self-improving-analytics

Workspace Structure

OpenClaw injects these files into every session:

~/.openclaw/workspace/
├── AGENTS.md          # Multi-agent workflows, delegation patterns
├── SOUL.md            # Behavioral guidelines, personality, principles
├── TOOLS.md           # Tool capabilities, integration gotchas
├── MEMORY.md          # Long-term memory (main session only)
├── memory/            # Daily memory files
│   └── YYYY-MM-DD.md
└── .learnings/        # This skill's log files
    ├── LEARNINGS.md
    ├── DATA_ISSUES.md
    └── FEATURE_REQUESTS.md

Create Learning Files

mkdir -p ~/.openclaw/workspace/.learnings

Then create the log files (or copy from assets/):

• LEARNINGS.md — metric drift, definition mismatches, visualization issues, data quality patterns
• DATA_ISSUES.md — pipeline failures, freshness breaches, schema drift, metric anomalies
• FEATURE_REQUESTS.md — analytics tools, BI capabilities, automation requests

Promotion Targets

When analytics learnings prove broadly applicable, promote them:

Learning Type	Promote To	Example
Metric definitions	Data dictionary	"Active user = login within 7 days with feature interaction"
Pipeline failure patterns	Pipeline runbooks	"DST partition handling: always use UTC-based keys"
Visualization standards	Dashboard style guide	"Absolute value charts must start Y-axis at zero"
Data quality rules	Data quality SLAs	"NULL rate in PK columns must be <0.01%"
Governance patterns	AGENTS.md	"New metrics require data dictionary entry before dashboard"
Tool configuration	TOOLS.md	"dbt source freshness checks required on all external sources"

Optional: Enable Hook

For automatic reminders at session start:

cp -r hooks/openclaw ~/.openclaw/hooks/self-improving-analytics
openclaw hooks enable self-improving-analytics

See references/openclaw-integration.md for complete details.

---

Generic Setup (Other Agents)

For Claude Code, Codex, Copilot, or other agents, create .learnings/ in the project or workspace root:

mkdir -p .learnings

Create the files inline using the headers shown above.

Add reference to agent files

Add to AGENTS.md, CLAUDE.md, or .github/copilot-instructions.md:

Self-Improving Analytics Workflow

When data issues or analytics patterns are discovered:

1. Log to .learnings/DATA_ISSUES.md, LEARNINGS.md, or FEATURE_REQUESTS.md
2. Review and promote broadly applicable learnings to:
   • Data dictionaries — canonical metric definitions
   • Pipeline runbooks — failure recovery procedures
   • Dashboard standards — visualization conventions
   • Data quality SLAs — monitoring thresholds and alerts

Logging Format

Learning Entry [LRN-YYYYMMDD-XXX]

Append to .learnings/LEARNINGS.md:

## [LRN-YYYYMMDD-XXX] category

**Logged**: ISO-8601 timestamp
**Priority**: low | medium | high | critical
**Status**: pending
**Area**: ingestion | transformation | modeling | reporting | visualization | governance | data_catalog

### Summary
One-line description of the analytics insight

### Details
Full context: what data pattern was found, why it is problematic,
what the correct approach is. Include root cause analysis.

### SQL Example

**Before (problematic):**
\`\`\`sql
-- problematic query, pipeline config, or metric definition
\`\`\`

**After (correct):**
\`\`\`sql
-- corrected query, config, or definition
\`\`\`

### Suggested Action
Specific data dictionary update, pipeline fix, dashboard change, or governance rule to adopt

### Metadata
- Source: etl_failure | freshness_breach | metric_anomaly | definition_conflict | dashboard_review | reconciliation_failure | schema_drift
- Pipeline: Airflow DAG name, dbt model, Fivetran connector (if applicable)
- Warehouse: snowflake | bigquery | redshift | postgres | databricks
- Related Tables: schema.table_name
- Tags: tag1, tag2
- See Also: LRN-20250110-001 (if related to existing entry)
- Pattern-Key: metric_drift.revenue_source | data_quality.null_spike (optional)
- Recurrence-Count: 1 (optional)
- First-Seen: 2025-01-15 (optional)
- Last-Seen: 2025-01-15 (optional)

---

Categories for learnings:

Category	Use When
data_quality	NULL spikes, duplicate records, invalid values, completeness issues
metric_drift	Metric calculation silently changed due to new data source, schema change, or logic update
pipeline_failure	ETL/ELT job failure, timeout, resource exhaustion, dependency issue
visualization_mislead	Chart axis, scale, aggregation, or color choice that misrepresents data
definition_mismatch	Same metric name with different definitions across teams or dashboards
freshness_issue	Data arriving later than SLA, stale dashboards, partition delays

Data Issue Entry [DAT-YYYYMMDD-XXX]

Append to .learnings/DATA_ISSUES.md:

## [DAT-YYYYMMDD-XXX] issue_type_or_name

**Logged**: ISO-8601 timestamp
**Priority**: high
**Status**: pending
**Area**: ingestion | transformation | modeling | reporting | visualization | governance | data_catalog

### Summary
Brief description of the data issue

### Error Output
\`\`\`
Actual error message, pipeline log, query error, or anomaly description (redacted/summarized)
\`\`\`

### Root Cause
What in the pipeline, data model, or source system caused this issue.
Include the problematic query or configuration.

### Fix
\`\`\`sql
-- corrected query, pipeline config, or data quality check
\`\`\`

### Prevention
How to avoid this issue in the future (data quality test, pipeline alert, schema validation, SLA monitor)

### Context
- Trigger: etl_failure | freshness_breach | metric_anomaly | null_spike | schema_drift | rendering_error
- Pipeline: Airflow DAG name, dbt model, Fivetran connector
- Warehouse: snowflake | bigquery | redshift | postgres | databricks
- Affected Tables: schema.table_name
- Downstream Impact: dashboards, reports, or teams affected

### Metadata
- Reproducible: yes | no | unknown
- Related Tables: schema.table_name
- See Also: DAT-20250110-001 (if recurring)

---

Feature Request Entry [FEAT-YYYYMMDD-XXX]

Append to .learnings/FEATURE_REQUESTS.md:

## [FEAT-YYYYMMDD-XXX] capability_name

**Logged**: ISO-8601 timestamp
**Priority**: medium
**Status**: pending
**Area**: ingestion | transformation | modeling | reporting | visualization | governance | data_catalog

### Requested Capability
What analytics tool, automation, or capability is needed

### User Context
Why it's needed, what workflow it improves, what data problem it solves

### Complexity Estimate
simple | medium | complex

### Suggested Implementation
How this could be built: dbt macro, Airflow operator, data quality check, Looker feature, governance workflow

### Metadata
- Frequency: first_time | recurring
- Related Features: existing_tool_or_capability

---

ID Generation

Format: TYPE-YYYYMMDD-XXX

• TYPE: LRN (learning), DAT (data issue), FEAT (feature request)
• YYYYMMDD: Current date
• XXX: Sequential number or random 3 chars (e.g., 001, A7B)

Examples: LRN-20250415-001, DAT-20250415-A3F, FEAT-20250415-002

Resolving Entries

When an issue is fixed, update the entry:

1. Change **Status**: pending → **Status**: resolved
2. Add resolution block after Metadata:

### Resolution
- **Resolved**: 2025-01-16T09:00:00Z
- **Commit/PR**: abc123 or #42
- **Notes**: Added data quality test / updated pipeline runbook / fixed metric definition

Other status values:

• in_progress — Actively being investigated or fixed
• wont_fix — Decided not to address (add reason in Resolution notes)
• promoted — Elevated to data dictionary, pipeline runbook, or dashboard standard
• promoted_to_skill — Extracted as a reusable skill

Detection Triggers

Automatically log when you encounter:

ETL/ELT Pipeline Failures (→ data issue with etl_failure trigger):

• Airflow DAG task failure or retry exhaustion
• dbt model build error or test failure
• Fivetran/Stitch sync failure or schema change warning
• Spark/Databricks job timeout or OOM
• CDC replication lag exceeding threshold

Data Freshness Breaches (→ data issue with freshness_breach trigger):

• Source data >24h stale (no new rows for today's partition)
• dbt source freshness check failure (warn_after, error_after)
• Dashboard last-refresh timestamp exceeds SLA
• Downstream model depends on stale upstream

Metric Value Anomalies (→ data issue with metric_anomaly trigger):

• Metric value >3 standard deviations from 30-day rolling mean
• Day-over-day change exceeds historical maximum
• Metric drops to zero unexpectedly
• Revenue, user count, or conversion rate shows sudden discontinuity

NULL Rate Spikes (→ data issue with null_spike trigger):

• Key column NULL rate exceeds baseline by >1 percentage point
• Primary key contains NULLs (should never happen)
• Foreign key NULL rate spikes after source system change
• Required business fields (email, amount, status) contain unexpected NULLs

Schema Changes (→ data issue with schema_drift trigger):

• Column added, removed, or renamed in source system
• Data type changed (VARCHAR length, NUMERIC precision)
• Constraint added or removed (NOT NULL, UNIQUE, FK)
• Table renamed or moved to different schema

Conflicting Definitions (→ learning with definition_mismatch category):

• Same metric name in two dashboards with different values
• Team A and Team B disagree on metric logic
• Data dictionary entry contradicts actual implementation
• Looker explore definition differs from dbt model documentation

Visualization Issues (→ learning with visualization_mislead category):

• Y-axis not starting at zero for absolute value charts
• Dual Y-axis with mismatched scales
• Pie chart with too many slices (>6)
• Time series with inconsistent intervals
• Color coding that implies ranking where none exists

Priority Guidelines

Priority	When to Use	Analytics Examples
critical	Wrong data in executive dashboard or regulatory report	Revenue under-reported to board, compliance data incorrect, PII exposure in dashboard
high	Pipeline down, metric definition conflict, SLA breach	Airflow DAG failed for >4h, Marketing vs Product metric mismatch, daily report stale
medium	Data quality degradation, visualization improvement	NULL rate trending up, dashboard axis misleading, catalog entry outdated
low	Catalog update, documentation, minor improvement	Column description missing, unused dashboard cleanup, tag standardization

Area Tags

Use to filter learnings by analytics domain:

Area	Scope
ingestion	Data extraction, loading, CDC replication, API pulls, file imports
transformation	SQL transforms, dbt models, Spark jobs, data cleaning, deduplication
modeling	Dimensional modeling, entity relationships, slowly changing dimensions, grain
reporting	Scheduled reports, email digests, PDF generation, data exports
visualization	Dashboards, charts, Looker explores, Tableau workbooks, Metabase questions
governance	Metric definitions, data ownership, access control, PII classification
data_catalog	Column descriptions, table documentation, lineage, tagging, search

Promoting to Permanent Analytics Standards

When a learning is broadly applicable (not a one-off data fix), promote it to permanent standards.

When to Promote

• Same metric definition conflict surfaces across 2+ teams
• Pipeline failure pattern recurs after different source system changes
• Visualization anti-pattern found in 3+ dashboards
• Data quality rule would have prevented multiple incidents
• Freshness monitoring gap caused repeated SLA breaches

Promotion Targets

Target	What Belongs There
Data dictionary	Canonical metric definitions with owner, grain, and refresh cadence
Pipeline runbooks	Step-by-step recovery for known failure patterns
Dashboard standards	Visualization conventions (axis, colors, aggregation rules)
Data quality SLAs	Monitoring thresholds and alert configurations
CLAUDE.md	Project-specific analytics conventions for AI agents
AGENTS.md	Automated analytics workflows, data validation steps

How to Promote

1. Distill the learning into a concise definition, rule, or procedure
2. Add to appropriate target (data dictionary entry, runbook step, SLA threshold)
3. Update original entry:
   • Change **Status**: pending → **Status**: promoted
   • Add **Promoted**: data dictionary (or pipeline runbook, dashboard standard, data quality SLA)

Promotion Examples

Learning → Data dictionary entry:

Marketing "active user" = 30-day login; Product = 7-day feature interaction → 420K vs 185K discrepancy.

Promoted as: active_users_30d (Marketing, login-based) and active_users_7d (Product, interaction-based) with governance note specifying which to use for board reports.

Learning → Pipeline runbook:

Pipeline fails every DST transition — partition key uses local time, hour 2 doesn't exist.

Promoted as: "DST Partition Recovery" runbook — rerun with UTC key, verify no duplicates, migrate all partitions to UTC.

Recurring Pattern Detection

If logging something similar to an existing entry:

1. Search first: grep -r "keyword" .learnings/
2. Link entries: Add **See Also**: DAT-20250110-001 in Metadata
3. Bump priority if issue keeps recurring
4. Consider systemic fix: Recurring analytics issues often indicate:
   • Missing data quality test (→ add dbt test or Great Expectations suite)
   • Missing monitoring (→ add freshness check or anomaly detection)
   • Governance gap (→ add metric definition to data dictionary)
   • Pipeline design flaw (→ refactor ingestion or transformation logic)

Periodic Review

Review .learnings/ at natural breakpoints:

When to Review

• Before building a new dashboard or metric
• After a pipeline incident is resolved
• When the same data quality issue appears again
• Before quarterly business reviews
• During data model refactoring

Quick Status Check

# Count pending analytics issues
grep -h "Status\*\*: pending" .learnings/*.md | wc -l

# List pending high-priority data issues
grep -B5 "Priority\*\*: high" .learnings/DATA_ISSUES.md | grep "^## \["

# Find learnings for a specific area
grep -l "Area\*\*: governance" .learnings/*.md

# Find all definition mismatches
grep -B2 "definition_mismatch" .learnings/LEARNINGS.md | grep "^## \["

Review Actions

• Resolve fixed data issues
• Promote recurring patterns to data dictionaries
• Link related entries across files
• Extract reusable data quality checks as skills

Simplify & Harden Feed

Ingest recurring analytics patterns from simplify-and-harden into data quality rules or governance standards.

1. For each candidate, use pattern_key as the dedupe key.
2. Search .learnings/LEARNINGS.md for existing entry: grep -n "Pattern-Key: <key>" .learnings/LEARNINGS.md
3. If found: increment Recurrence-Count, update Last-Seen, add See Also links.
4. If not found: create new LRN-... entry with Source: simplify-and-harden.

Promotion threshold: Recurrence-Count >= 3, seen in 2+ pipelines/dashboards, within 30-day window. Targets: data dictionary entries, pipeline runbooks, dashboard standards, CLAUDE.md / AGENTS.md.

Hook Integration

Enable automatic reminders through agent hooks. This is opt-in.

Quick Setup (Claude Code / Codex)

Create .claude/settings.json in your project:

{
  "hooks": {
    "UserPromptSubmit": [{
      "matcher": "",
      "hooks": [{
        "type": "command",
        "command": "./skills/self-improving-analytics/scripts/activator.sh"
      }]
    }]
  }
}

This injects an analytics-focused learning evaluation reminder after each prompt (~50-100 tokens overhead).

Advanced Setup (With Error Detection)

{
  "hooks": {
    "UserPromptSubmit": [{
      "matcher": "",
      "hooks": [{
        "type": "command",
        "command": "./skills/self-improving-analytics/scripts/activator.sh"
      }]
    }],
    "PostToolUse": [{
      "matcher": "Bash",
      "hooks": [{
        "type": "command",
        "command": "./skills/self-improving-analytics/scripts/error-detector.sh"
      }]
    }]
  }
}

Enable PostToolUse only if you want the hook to inspect command output for pipeline errors, query failures, and data quality issues.

Available Hook Scripts

Script	Hook Type	Purpose
scripts/activator.sh	UserPromptSubmit	Reminds to evaluate analytics learnings after tasks
scripts/error-detector.sh	PostToolUse (Bash)	Triggers on pipeline errors, query failures, data quality issues

See references/hooks-setup.md for detailed configuration and troubleshooting.

Automatic Skill Extraction

When an analytics learning is valuable enough to become a reusable skill, extract it.

Skill Extraction Criteria

Criterion	Description
Recurring	Same data issue in 2+ pipelines or warehouses
Verified	Status is resolved with working fix and data quality test
Non-obvious	Required actual investigation or cross-team coordination
Broadly applicable	Not project-specific; useful across data stacks
User-flagged	User says "save this as a skill" or similar

Extraction Workflow

1. Identify candidate: Learning meets extraction criteria
2. Run helper (or create manually):

   ./skills/self-improving-analytics/scripts/extract-skill.sh skill-name --dry-run
   ./skills/self-improving-analytics/scripts/extract-skill.sh skill-name
   

3. Customize SKILL.md: Fill in template with analytics-specific content
4. Update learning: Set status to promoted_to_skill, add Skill-Path
5. Verify: Read skill in fresh session to ensure it's self-contained

Extraction Detection Triggers

In conversation: "This pipeline keeps failing the same way", "Save this data quality check as a skill", "Every warehouse has this DST issue", "This metric definition problem happens everywhere".

In entries: Multiple See Also links, high priority + resolved, definition_mismatch or pipeline_failure with broad applicability, same Pattern-Key across projects.

Multi-Agent Support

Agent	Activation	Detection
Claude Code	Hooks (UserPromptSubmit, PostToolUse)	Automatic via error-detector.sh
Codex CLI	Hooks (same pattern)	Automatic via hook scripts
GitHub Copilot	Manual (.github/copilot-instructions.md)	Manual review
OpenClaw	Workspace injection + inter-agent messaging	Via session tools

Best Practices

1. Define metrics before building — agree on definitions in the data dictionary before creating dashboards
2. Validate at ingestion — catch data quality issues as early as possible in the pipeline
3. Test transformations with known data — use static test fixtures, not just production samples
4. Document lineage — every metric should trace back to source tables through explicit joins
5. Alert on anomalies, not just failures — a pipeline that succeeds but produces wrong data is worse than one that fails
6. Use UTC everywhere — local time in partitions, timestamps, or schedules causes DST and timezone bugs
7. Version metric definitions — treat metric logic changes like code changes with review and approval
8. Separate facts from interpretation — dashboards should present data; narrative belongs in annotations
9. Review before building in same area — check .learnings/ for past issues with the same tables or metrics
10. Reconcile against source of truth — compare warehouse aggregates against GL, CRM, or billing system totals

Gitignore Options

Keep learnings local (per-analyst): add .learnings/ to .gitignore.

Track learnings in repo (team-wide): don't add to .gitignore — learnings become shared knowledge.

Hybrid (track templates, ignore entries): ignore .learnings/*.md, keep .learnings/.gitkeep.

Stackability Contract (Standalone + Multi-Skill)

This skill is standalone-compatible and stackable with other self-improving skills.

Namespaced Logging (recommended for 2+ skills)

• Namespace for this skill: .learnings/analytics/
• Keep current standalone behavior if you prefer flat files.
• Optional shared index for all skills: .learnings/INDEX.md

Required Metadata

Every new entry must include:

**Skill**: analytics

Hook Arbitration (when 2+ skills are enabled)

• Use one dispatcher hook as the single entrypoint.
• Dispatcher responsibilities: route by matcher, dedupe repeated events, and rate-limit reminders.
• Suggested defaults: dedupe key = event + matcher + file + 5m_window; max 1 reminder per skill every 5 minutes.

Narrow Matcher Scope (analytics)

Only trigger this skill automatically for analytics signals such as:

• pipeline|etl|schema drift|metric mismatch|dashboard
• lineage|warehouse|bi|attribution|anomaly
• explicit analytics intent in user prompt

Cross-Skill Precedence

When guidance conflicts, apply:

1. security
2. engineering
3. coding
4. ai
5. user-explicit domain skill
6. meta as tie-breaker

Ownership Rules

• This skill writes only to .learnings/analytics/ in stackable mode.
• It may read other skill folders for cross-linking, but should not rewrite their entries.