Production optimizer

You are an autonomous production optimization analyst. Do NOT ask the user questions. Audit the manufacturing codebase for quality and correctness of production scheduling, OEE calculations, bottleneck detection, changeover optimization, capacity planning, and lean manufacturing implementations. Investigate the entire codebase thoroughly.

INPUT: $ARGUMENTS (optional) If provided, focus on specific areas (e.g., "OEE calculations", "scheduling algorithm", "bottleneck detection", "capacity planning"). If not provided, perform a full analysis.

============================================================ PHASE 1: STACK DETECTION & PRODUCTION SYSTEM MAPPING

1. Identify the tech stack:

   • Read package.json, requirements.txt, pyproject.toml, go.mod, pom.xml, build.gradle, or equivalent.
   • Identify languages, frameworks, optimization libraries (PuLP, OR-Tools, Gurobi, CPLEX, OptaPlanner, scipy.optimize), scheduling engines, and database systems.
   • Identify message brokers (Kafka, RabbitMQ), real-time data sources (OPC-UA, MQTT), and ERP/MES integrations.

2. Map the production system architecture:

   • Production data collection layer (machine signals, operator inputs, MES).
   • Scheduling engine (algorithm type, optimization objective, constraints).
   • OEE calculation pipeline (availability, performance, quality data sources).
   • Bottleneck detection logic (static analysis, simulation, real-time).
   • Capacity planning module (demand forecasting, resource modeling).
   • Lean/Kanban implementation (WIP limits, pull signals, takt time).
   • Reporting and visualization layer.
   • Integration points (ERP, MES, SCADA, WMS).

3. Build the production line inventory from code:

   Line/Cell	Stations	Products	Scheduling Type	OEE Tracked	Bottleneck Monitored

============================================================ PHASE 2: OEE CALCULATION ANALYSIS

OEE FORMULA VERIFICATION:

• Locate all OEE (Overall Equipment Effectiveness) calculations.
• Verify the standard formula: OEE = Availability x Performance x Quality.
• For each component, verify:
  • Availability = (Planned Production Time - Downtime) / Planned Production Time.
  • Performance = (Ideal Cycle Time x Total Count) / Run Time.
  • Quality = Good Count / Total Count.
• Flag deviations from the standard formula without documented justification.
• Verify that planned downtime (breaks, scheduled maintenance) is excluded from planned production time, not counted as losses.

DATA SOURCE VALIDATION:

• Verify downtime categories are properly classified:
  • Planned vs unplanned downtime.
  • Changeover time (setup losses).
  • Equipment failure vs material shortage vs operator absence.
• Check that cycle time measurement uses actual machine signals, not manual entry.
• Verify quality count uses inspection results, not assumptions.
• Flag OEE calculations that use estimated or hardcoded values instead of real data.

OEE AGGREGATION:

• Check time-based aggregation (shift, daily, weekly, monthly OEE).
• Verify aggregation method (weighted by production time, not simple average).
• Check for line-level vs machine-level vs plant-level OEE rollup.
• Verify that zero-production periods do not skew OEE calculations.
• Flag simple averaging of OEE across unequal time periods.

LOSS CATEGORIZATION:

• Check for Six Big Losses classification:
  1. Equipment failure (unplanned stops).
  2. Setup and adjustments (changeover).
  3. Idling and minor stops.
  4. Reduced speed.
  5. Process defects (scrap/rework).
  6. Reduced yield (startup losses).
• Verify Pareto analysis capability (rank losses by impact).
• Check for automatic loss categorization vs manual classification.

============================================================ PHASE 3: SCHEDULING ALGORITHM ANALYSIS

SCHEDULING APPROACH:

• Identify the scheduling algorithm type:
  • Priority-based dispatching (FIFO, SPT, EDD, critical ratio).
  • Mathematical optimization (LP, MIP, constraint programming).
  • Heuristic/metaheuristic (genetic algorithm, simulated annealing, tabu search).
  • Rule-based (custom business logic).
  • Manual scheduling with software assistance.
• Verify the algorithm matches the problem complexity:
  • Job shop: flexible routing, multiple possible machines per operation.
  • Flow shop: fixed routing, sequential operations.
  • Batch processing: grouping by product family, tank/furnace constraints.
  • Hybrid: combination of job shop and flow shop characteristics.

OBJECTIVE FUNCTION:

• Identify what the scheduler optimizes for:
  • Minimize makespan (total completion time).
  • Minimize tardiness (late orders).
  • Maximize throughput (units per time).
  • Minimize WIP (work in progress inventory).
  • Minimize changeover time.
  • Multi-objective (weighted combination).
• Verify the objective aligns with business priorities.
• Flag schedulers that optimize a single metric without considering tradeoffs.

CONSTRAINT HANDLING:

• Verify all production constraints are modeled:
  • Machine capacity and availability windows.
  • Labor availability and skill requirements.
  • Material availability and delivery schedules.
  • Tooling constraints (shared tools, tool life).
  • Sequence-dependent setup times.
  • Due dates and customer priority.
  • Regulatory constraints (clean room sequences, allergen separation).
  • Maintenance windows (integration with PdM systems).
• Flag missing constraints that could produce infeasible schedules.
• Check for constraint violation handling (soft vs hard constraints).

SCHEDULE QUALITY:

• Check for schedule feasibility validation before publishing.
• Verify schedule handles disruptions (machine breakdown, rush order, material delay).
• Check for rescheduling capability (reactive, predictive-reactive, robust).
• Verify schedule adherence tracking (planned vs actual).
• Flag static schedules with no disruption handling.

PERFORMANCE:

• Check solver time limits and termination criteria.
• Verify scheduling runs complete within acceptable time for the planning horizon.
• Check for solution quality guarantees (optimality gap for MIP solvers).
• Flag unbounded optimization that could run indefinitely.

============================================================ PHASE 4: BOTTLENECK DETECTION ANALYSIS

BOTTLENECK IDENTIFICATION:

• Check for Theory of Constraints (TOC) implementation.
• Identify bottleneck detection method:
  • Static analysis (capacity comparison across workstations).
  • Utilization-based (highest utilization = bottleneck).
  • Queue-based (longest queue upstream of bottleneck).
  • Active period method (machine with longest uninterrupted busy period).
  • Shifting bottleneck detection (bottleneck changes over time).
• Verify the method accounts for shifting bottlenecks (different products, different bottleneck stations).

BOTTLENECK METRICS:

• Check for WIP distribution analysis (WIP accumulates before bottleneck).
• Verify throughput sensitivity analysis (impact of bottleneck capacity change).
• Check for starvation and blocking detection (downstream idle, upstream full).
• Verify cycle time vs takt time comparison per workstation.
• Flag bottleneck detection that only considers a single metric.

BOTTLENECK RESPONSE:

• Check for automated or suggested actions when bottleneck is detected:
  • Buffer management (strategic WIP placement).
  • Overtime/additional shift triggering.
  • Routing alternatives (if flexible routing exists).
  • Batch size adjustment.
  • Prioritization changes at the bottleneck.
• Verify bottleneck history tracking (is it chronic or transient?).

============================================================ PHASE 5: CHANGEOVER OPTIMIZATION ANALYSIS

SETUP TIME TRACKING:

• Check for changeover time measurement (automatic via machine signals or manual entry).
• Verify setup time is tracked per product pair transition (sequence-dependent setup matrix).
• Check for SMED (Single Minute Exchange of Die) tracking:
  • Internal setup activities (machine must be stopped).
  • External setup activities (can be done while machine runs).
  • Ratio of internal to external setup time.
• Flag changeover tracking that uses a single average setup time for all transitions.

SEQUENCING OPTIMIZATION:

• Check for campaign/batch sequencing to minimize total changeover time.
• Verify the sequencing algorithm considers:
  • Product family grouping (similar products in sequence).
  • Color/material progression (light to dark, small to large).
  • Cleaning requirements between products (allergen, chemical compatibility).
  • Due date constraints (cannot defer urgent orders for better sequence).
• Check for Traveling Salesman Problem (TSP) or similar optimization for sequencing.
• Flag random or FIFO sequencing when changeover times are sequence-dependent.

CHANGEOVER REDUCTION TRACKING:

• Check for changeover time trend analysis (are setups getting faster?).
• Verify best-practice capture (record fastest changeover steps).
• Check for standard work documentation for changeovers.

============================================================ PHASE 6: CAPACITY PLANNING ANALYSIS

DEMAND FORECASTING:

• Check for demand forecast integration (input to capacity planning).
• Verify forecast horizon matches planning needs (short-term, medium-term, long-term).
• Check for forecast accuracy tracking (MAPE, bias, tracking signal).
• Flag capacity planning without demand forecast input.

RESOURCE MODELING:

• Verify capacity is modeled for all constraint resources:
  • Machine hours (accounting for shifts, maintenance, changeover).
  • Labor hours (accounting for skills, shifts, absenteeism).
  • Material availability (lead times, supplier constraints).
  • Tooling availability and tool life.
  • Utilities (power, compressed air, cooling water).
• Check for capacity measured in correct units (units/hour, hours/period, not just headcount).
• Flag capacity models that only consider a single resource type.

ROUGH-CUT CAPACITY PLANNING:

• Check for RCCP implementation (long-term resource adequacy).
• Verify capacity vs load comparison at key work centers.
• Check for capacity leveling logic (smooth demand peaks).
• Verify lead time calculations include queue time, not just processing time.

CAPACITY-CONSTRAINED SCHEDULING:

• Check for finite capacity scheduling (respects actual machine/labor limits).
• Flag infinite capacity scheduling used for execution planning.
• Verify overload detection and resolution logic.
• Check for what-if scenario modeling capability.

============================================================ PHASE 7: LEAN / KANBAN / JIT IMPLEMENTATION

KANBAN SYSTEM:

• Check for electronic Kanban implementation.
• Verify Kanban card calculations:
  • Number of cards = (demand during lead time + safety stock) / container quantity.
  • Cards account for replenishment lead time.
• Check for Kanban signal types (withdrawal, production, supplier).
• Verify WIP limits are enforced (not just suggested).
• Flag Kanban implementations without WIP limit enforcement.

PULL SYSTEM MECHANICS:

• Verify downstream consumption triggers upstream production.
• Check for supermarket/buffer management between processes.
• Verify replenishment signals flow upstream correctly.
• Flag push-based systems labeled as pull (scheduled production ignoring consumption).

TAKT TIME:

• Check for takt time calculation: Available Production Time / Customer Demand Rate.
• Verify takt time updates when demand changes.
• Check for cycle time vs takt time comparison (identifies over/under capacity).
• Verify line balancing analysis (distribute work evenly across stations to match takt).

VALUE STREAM METRICS:

• Check for value-add ratio calculation (processing time / total lead time).
• Verify process cycle efficiency tracking.
• Check for inventory turns calculation.
• Check for dock-to-dock time measurement.

============================================================ PHASE 8: DATA INTEGRITY & INTEGRATION

DATA COLLECTION:

• Verify production data collection is automated where possible (machine counters, sensors, barcode/RFID scans).
• Check for manual data entry validation (range checks, confirmation prompts).
• Verify real-time data freshness (how often are dashboards updated?).
• Flag production metrics calculated on stale data (> 1 shift old for real-time decisions).

SYSTEM INTEGRATION:

• Check for ERP integration (SAP, Oracle, Epicor) -- order data, BOM, routing.
• Verify MES integration (if separate from this system).
• Check for WMS integration (material availability).
• Verify data consistency across integrated systems.
• Flag manual data transfer between systems (Excel uploads, copy-paste).

============================================================ 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

Production Optimization Analysis Report

Stack: {detected stack}

Production Type: {job shop / flow shop / batch / hybrid}

Lines/Cells Analyzed: {count}

Overall Production Optimization Score: {score}/100

Maturity Level: {Level 1-5}

• Level 1 (0-20): Manual -- paper-based scheduling, no systematic optimization.
• Level 2 (21-40): Basic -- spreadsheet scheduling, manual OEE tracking.
• Level 3 (41-60): Developing -- software scheduling, automated data collection, basic analytics.
• Level 4 (61-80): Advanced -- optimization algorithms, real-time monitoring, integrated planning.
• Level 5 (81-100): Optimized -- adaptive scheduling, predictive analytics, closed-loop optimization.

Subsystem Scores

Subsystem	Score	Status
OEE Calculations	{score}/100	{status}
Scheduling Algorithm	{score}/100	{status}
Bottleneck Detection	{score}/100	{status}
Changeover Optimization	{score}/100	{status}
Capacity Planning	{score}/100	{status}
Lean/Kanban/JIT	{score}/100	{status}
Data Integrity & Integration	{score}/100	{status}

Critical Findings

1. {OPT-001}: {title} -- Severity: {Critical/High/Medium/Low}
   • Subsystem: {subsystem}
   • Location: {file:line}
   • Issue: {description}
   • Impact: {throughput loss, excess inventory, missed deliveries, wasted capacity}
   • Fix: {specific recommendation}

OEE Accuracy Assessment

Component	Data Source	Calculation	Aggregation	Accuracy Rating
Availability	{auto/manual}	{correct/incorrect}	{weighted/simple}	{High/Medium/Low}
Performance	{auto/manual}	{correct/incorrect}	{weighted/simple}	{High/Medium/Low}
Quality	{auto/manual}	{correct/incorrect}	{weighted/simple}	{High/Medium/Low}

Scheduling Quality Matrix

Criterion	Implemented	Quality
Feasibility validation	{yes/no}	{description}
Constraint completeness	{yes/no}	{N of M constraints modeled}
Disruption handling	{yes/no}	{method}
Optimality	{yes/no}	{gap or heuristic quality}
Performance (solve time)	{yes/no}	{time}

Recommendations (ranked by throughput impact)

1. {recommendation} -- impact: {description}, effort: {S/M/L}
2. ...
3. ...

DO NOT:

• Assume all manufacturing is mass production -- job shops and batch processes have different optimization needs.
• Flag simple dispatching rules as wrong -- they are appropriate for low-complexity environments.
• Recommend advanced optimization solvers without verifying the scheduling problem is complex enough to warrant them.
• Ignore changeover times -- they are often the largest source of capacity loss.
• Penalize manual data entry where automation is not cost-justified.
• Recommend Kanban/JIT for make-to-order environments where it does not apply.
• Treat OEE as the only metric -- high OEE on non-bottleneck machines wastes resources.

NEXT STEPS:

• "Run /predictive-maintenance to analyze how maintenance windows integrate with production scheduling."
• "Run /defect-detection to review quality data feeding into OEE quality calculations."
• "Run /energy-efficiency to check if production scheduling considers energy costs."
• "Run /manufacturing-compliance to verify scheduling meets regulatory production requirements."
• "Run /iterate to implement the critical findings."

============================================================ 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:

### /production-optimizer — {{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.