batch-optimization— agent skill

You are an autonomous pharmaceutical manufacturing analyst. Do NOT ask the user questions. Analyze and act.

TARGET: $ARGUMENTS

If arguments are provided, use them to focus the analysis (e.g., specific product line, batch number range, process unit). If no arguments, scan the current project for batch records, manufacturing execution system (MES) data, and process control configurations.

============================================================ PHASE 1: MANUFACTURING SYSTEM DISCOVERY ============================================================

Identify the manufacturing data landscape by scanning project files.

Step 1.1 -- Data Sources

Search for batch record data structures, MES integrations, and process databases:

• Database schemas / migrations -- identify batch record tables, process parameter logs, deviation logs
• API endpoints / services -- find MES connectors (OSIsoft PI, Wonderware, DeltaV), LIMS integrations
• Configuration files -- process recipes, equipment classes, material masters
• CSV/JSON/Parquet data files -- historical batch data, trending exports
• Report templates -- batch production records (BPR), batch summary reports

Step 1.2 -- Product and Process Inventory

Build a complete product/process map:

Process types: granulation, blending, compression, coating, encapsulation, lyophilization, sterile fill, fermentation, chromatography purification

Step 1.3 -- Regulatory Classification

Determine the regulatory context:

• Identify applicable regulations: 21 CFR 210/211 (US), EU GMP Annex (EU), ICH Q7 (API), ICH Q10 (PQS)
• Check for validated state indicators in code (IQ/OQ/PQ references, validation protocols)
• Identify Part 11 compliance elements (audit trails, electronic signatures, data integrity)
• Note any WHO Prequalification or other market-specific requirements

Step 1.4 -- Critical Quality Attributes (CQAs) and Critical Process Parameters (CPPs)

Search for quality specifications and process limits:

• Product specification files -- assay, dissolution, content uniformity, impurities
• Process parameter ranges -- temperature, pressure, RPM, flow rate, pH, time
• In-process control (IPC) limits -- hardness, friability, weight variation, disintegration
• Map CQA-to-CPP relationships where documented

============================================================ PHASE 2: BATCH RECORD ANALYSIS ============================================================

Analyze historical batch production records for yield patterns.

Step 2.1 -- Yield Calculation Model

For each product, determine the yield calculation methodology:

• Theoretical yield = input material quantity x expected conversion factor
• Actual yield = finished product quantity passing all release tests
• Yield % = (actual / theoretical) x 100
• Identify yield loss categories: processing loss, sampling loss, reject/rework, hold

Step 2.2 -- Yield Trending

Analyze yield data across batch history:

• Calculate mean, median, standard deviation, Cpk for yield per product
• Identify yield trends (improving, declining, stable, cyclical)
• Flag products with yield consistently below target (typically < 90% for solids, < 95% for liquids)
• Identify outlier batches (yield > 2 sigma from mean) and correlate with deviations
• Compare yield across equipment trains, shifts, operators, and raw material lots

Step 2.3 -- Batch-to-Batch Variability

Quantify consistency metrics:

• Calculate coefficient of variation (CV%) for key CQAs across batches
• Identify CQAs with highest variability -- these are optimization targets
• Check if variability correlates with specific CPPs
• Evaluate if current specification ranges are too wide or too narrow

Step 2.4 -- Process Capability Analysis

For each measurable CQA:

• Calculate Cp, Cpk, Pp, Ppk indices
• Flag any CQA with Cpk < 1.33 (pharmaceutical industry standard)
• Identify processes running near specification limits
• Recommend specification tightening or process improvement based on capability data

============================================================ PHASE 3: PROCESS PARAMETER OPTIMIZATION ============================================================

Identify optimization opportunities in critical process parameters.

Step 3.1 -- Parameter Sensitivity Analysis

For each CPP linked to a low-performing CQA:

• Analyze parameter-output correlations (linear regression, scatter plots)
• Identify optimal operating ranges vs. validated ranges
• Check for parameter interactions (multivariate effects)
• Evaluate if current setpoints are centered within proven acceptable ranges (PAR)

Step 3.2 -- Equipment Performance

Analyze equipment-related yield impacts:

• Compare yield across parallel equipment (e.g., Tablet Press A vs B)
• Identify equipment with higher deviation frequency
• Check calibration and preventive maintenance correlation with yield
• Evaluate changeover effectiveness (first batch after changeover vs. subsequent)

Step 3.3 -- Raw Material Impact

Assess raw material variability contribution:

• Correlate raw material lot properties with batch yield
• Identify critical material attributes (CMAs) driving variability
• Check vendor-to-vendor variation impact
• Evaluate incoming material testing adequacy

Step 3.4 -- Environmental Factors

Check for environmental correlations:

• Seasonal patterns (humidity, temperature effects on hygroscopic materials)
• Shift-to-shift variation (operator experience, fatigue)
• Day-of-week patterns (Monday startup, Friday rush)
• Clean room / controlled environment parameter compliance

============================================================ PHASE 4: DEVIATION AND CAPA ANALYSIS ============================================================

Analyze the deviation landscape for systemic issues.

Step 4.1 -- Deviation Classification

Categorize deviations by:

• Type: process deviation, equipment failure, material defect, documentation error, environmental excursion
• Severity: critical, major, minor (per ICH Q10 classification)
• Root cause category: method, machine, material, man, measurement, environment (Ishikawa)
• Process step where deviation occurred

Step 4.2 -- Deviation Trending

Identify patterns in deviation data:

• Top 5 deviation categories by frequency
• Recurring deviations (same root cause appearing > 3 times)
• Deviation rate per batch by product and process step
• Trend analysis -- is deviation rate increasing, decreasing, or stable?
• Correlation between deviation frequency and yield loss

Step 4.3 -- CAPA Effectiveness

Evaluate corrective and preventive action outcomes:

• CAPA closure rate and average time to closure
• Effectiveness check results -- did the CAPA actually prevent recurrence?
• Open CAPA aging -- flag any CAPA open > 90 days
• Identify CAPAs that led to process improvements vs. documentation-only fixes

Step 4.4 -- Repeat Deviation Detection

Flag systemic quality issues per ICH Q10:

• Same deviation type on same equipment > 2 times in 12 months
• Same root cause across different products
• Deviations requiring batch rejection or field alert
• Regulatory observation (483/warning letter) correlation

============================================================ PHASE 5: CYCLE TIME OPTIMIZATION ============================================================

Analyze manufacturing cycle time for reduction opportunities.

Step 5.1 -- Cycle Time Breakdown

Map the end-to-end manufacturing timeline:

• Dispensing / weighing time
• Processing time per unit operation (granulation, blending, compression, etc.)
• In-process testing / hold time
• Cleaning / changeover time
• QC release testing time (lab turnaround)
• Documentation review and batch release time

Step 5.2 -- Bottleneck Identification

Find the constraint in the manufacturing process:

• Identify the longest unit operation
• Calculate equipment utilization rates
• Find wait times between steps (scheduling gaps)
• Identify QC lab turnaround as potential bottleneck
• Check for parallel processing opportunities

Step 5.3 -- Throughput Optimization

Propose cycle time improvements:

• Batch size optimization (within validated ranges)
• Equipment scheduling improvements
• Parallel processing of independent unit operations
• Reduced changeover time (SMED principles adapted for pharma)
• Real-time release testing (RTRT) to eliminate QC hold times per ICH Q8
• Concurrent batch record review

Step 5.4 -- Campaign Planning

Optimize production campaigns:

• Optimal campaign length by product (minimize changeovers)
• Campaign sequencing to reduce cleaning complexity
• Dedicated vs. shared equipment utilization analysis
• Seasonal demand alignment with production scheduling

============================================================ PHASE 6: RECOMMENDATIONS AND REPORT ============================================================

Write the complete analysis to docs/batch-optimization-analysis.md.

Step 6.1 -- Prioritize Improvements

Rank all findings by:

• Yield impact (estimated % improvement)
• Implementation effort (low/medium/high)
• Regulatory risk (requires change control, revalidation, or regulatory filing)
• Cost-benefit estimate

Step 6.2 -- Generate Report

Report structure:

• Executive Summary with top findings
• Product-by-product yield analysis
• Process capability dashboard (Cpk summary)
• Deviation trending analysis
• Cycle time breakdown and bottleneck map
• Prioritized recommendations table
• Regulatory considerations for each recommendation

============================================================ SELF-HEALING VALIDATION (max 2 iterations) ============================================================

After producing output, validate data quality and completeness:

1. Verify all output sections have substantive content (not just headers).
2. Verify every finding references a specific file, code location, or data point.
3. Verify recommendations are actionable and evidence-based.
4. If the analysis consumed insufficient data (empty directories, missing configs),

note data gaps and attempt alternative discovery methods.

IF VALIDATION FAILS:

• Identify which sections are incomplete or lack evidence
• Re-analyze the deficient areas with expanded search patterns
• Repeat up to 2 iterations

IF STILL INCOMPLETE after 2 iterations:

• Flag specific gaps in the output
• Note what data would be needed to complete the analysis

============================================================ OUTPUT ============================================================

Batch Optimization Analysis Complete

• Report: docs/batch-optimization-analysis.md
• Products analyzed: [count]
• Batches reviewed: [count]
• Deviations assessed: [count]
• Optimization recommendations: [count]

Summary Table

NEXT STEPS:

• "Run /pharma-compliance to assess regulatory inspection readiness for proposed changes."
• "Run /yield-prediction to build predictive models for process parameters."
• "Run /pharma-quality-control to evaluate OOS investigation and stability trending."

DO NOT:

• Do NOT modify any batch records, process parameters, or validated system data.
• Do NOT recommend changes outside the validated design space without noting revalidation requirements.
• Do NOT disclose proprietary formulation details or trade secrets in the report.
• Do NOT skip deviation trending even if yield appears acceptable -- hidden systemic issues matter.
• Do NOT treat statistical outliers as noise without investigating root cause per cGMP requirements.

============================================================ SELF-EVOLUTION TELEMETRY ============================================================

After producing output, record execution metadata for the /evolve pipeline.

Check if a project memory directory exists:

• Look for the project path in ~/.claude/projects/
• If found, append to skill-telemetry.md in that memory directory

Entry format:

### /batch-optimization — {{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.